**4.3 TLC/HPTLC methods**

In general practice, thin layer chromatography (TLC) is not preferred for the accurate determination of very low residual analyte level. However, this technique is still used for the determination of related substances in the pharmacopoeial monographs for amiodarone, bromazepam, carmustine, ifosamide, indoramin, and tolnaftate (Elder *et al.*, 2008).

Nevertheless, there are several examples of its use in association with determining levels of the epoxyl alkaloid, including scopolamine in extracts of *Datura stramonium*. Sass and Stutz (1981) used TLC to determine residual sulfur and nitrogen mustards (beta haloethyl compounds) in a variety of substrates in which the sensitivities in the microgram range were typically achievable. High performance thin layer chromatography (HPTLC) was used for monitoring the degradation products of rifampicin, including the hydrazones (25-desacetyl rifampicin (DAR)) and rifampicin quinone (RQU). Finally, it was concluded that the method is suitable for routine quality control and stability analyses, especially in the developing world (Jindal *et al.* 1994).

#### **4.4 Capillary electrophoresis methods**

Jouyban and Kenndler (2008) reviewed the applicability of capillary electrophoresis (CE) methods for the analysis of pharmaceutical impurities. In addition, they discussed the applications of these methods in various groups of compounds such as chemotherapeutic agents, central nervous system (CNS) drugs, histamine receptor and cardiovascular drugs.

The main advantage of CE techniques is their selectivity; thus, they are suitable for the analysis of complex herbal products. Bempong *et al.* (1993) reported the separation of 13-cis and all-trans retinoic acid and their photo-degradation products (including all-trans-5, 6 epoxy retinoic acid, 13-cis-5, 6-epoxy retinoic acid) using both capillary zone electrophoresis (CZE) and micellar electrokinetic chromatography (MEKC) methods. A Chinese research group reported the development of CE methods for the simultaneous determination of some hydrazine related impurities (Liu *et al.*, 1996).

Hansen and Sheribah (2005) evaluated a series of electrically driven separation techniques: CZE, MEKC, and microemulsion electrokinetic chromatography (MEEKC) for the determination of residual alkylating impurities in bromazepam API. However, the poor sensitivity of the techniques posed a problem even when specialized detection cells (e.g. bubble or Z-cells) were used. Mahuzier *et al.* (2001) demonstrated the poor sensitivity of CE based methods, in comparison to other separation methods. The problem of limited sensitivity of CE methods can be solved either by the use of detection methods with sensitivity higher than UV absorption or by pre-concentration of the analytes (Jouyban and Kenndler, 2008).

#### **4.5 Enhancing methods**

404 Toxicity and Drug Testing

procedure involving the formation of a benzalazine derivative was developed for monitoring the residual levels of hydrazine in hydralazine and isoniazid APIs, tablets, combined tablets, syrups, and injectable products in which nitrogen selective detection was

In addition, Carlin *et al.* (1998) adapted a previously published method for monitoring a benzalazine derivative using GC with electron capture (EC) detection. The LOQ was 10 ppm and the method was linear over the range of 10-100 ppm. The inter-day residual standard deviation (RSD) based on six measurements at analyte levels of 10 ppm was 15%; however, this improved slightly at increased analyte concentrations of 25 and 100 ppm, to 9.5% and

Nevertheless, non-volatile API does not partition into the headspace and therefore does not enter the GC system; as a result, headspace injection becomes the preferred choice whenever

In general practice, thin layer chromatography (TLC) is not preferred for the accurate determination of very low residual analyte level. However, this technique is still used for the determination of related substances in the pharmacopoeial monographs for amiodarone,

Nevertheless, there are several examples of its use in association with determining levels of the epoxyl alkaloid, including scopolamine in extracts of *Datura stramonium*. Sass and Stutz (1981) used TLC to determine residual sulfur and nitrogen mustards (beta haloethyl compounds) in a variety of substrates in which the sensitivities in the microgram range were typically achievable. High performance thin layer chromatography (HPTLC) was used for monitoring the degradation products of rifampicin, including the hydrazones (25-desacetyl rifampicin (DAR)) and rifampicin quinone (RQU). Finally, it was concluded that the method is suitable for routine quality control and stability analyses, especially in the developing

Jouyban and Kenndler (2008) reviewed the applicability of capillary electrophoresis (CE) methods for the analysis of pharmaceutical impurities. In addition, they discussed the applications of these methods in various groups of compounds such as chemotherapeutic agents, central nervous system (CNS) drugs, histamine receptor and cardiovascular drugs. The main advantage of CE techniques is their selectivity; thus, they are suitable for the analysis of complex herbal products. Bempong *et al.* (1993) reported the separation of 13-cis and all-trans retinoic acid and their photo-degradation products (including all-trans-5, 6 epoxy retinoic acid, 13-cis-5, 6-epoxy retinoic acid) using both capillary zone electrophoresis (CZE) and micellar electrokinetic chromatography (MEKC) methods. A Chinese research group reported the development of CE methods for the simultaneous determination of some

Hansen and Sheribah (2005) evaluated a series of electrically driven separation techniques: CZE, MEKC, and microemulsion electrokinetic chromatography (MEEKC) for the determination of residual alkylating impurities in bromazepam API. However, the poor sensitivity of the techniques posed a problem even when specialized detection cells (e.g. bubble or Z-cells) were used. Mahuzier *et al.* (2001) demonstrated the poor sensitivity of CE

bromazepam, carmustine, ifosamide, indoramin, and tolnaftate (Elder *et al.*, 2008).

used (Matsui *et al.*, 1983).

11.3%, respectively.

possible (Liu *et al.*, 2010).

**4.3 TLC/HPTLC methods** 

world (Jindal *et al.* 1994).

**4.4 Capillary electrophoresis methods** 

hydrazine related impurities (Liu *et al.*, 1996).

Alternatively, the structure of the molecule as well as its properties can be altered to enhance detectability which in turn will help to achieve the desired sensitivity. This is especially true for GIs that lack structural features for sensitive detection (Bai *et al.*, 2010; Liu *et al.*, 2010). A number of general approaches could be considered, some of which are explained below:

#### **4.5.1 Chemical derivatization**

This method is generally used for stabilizing reactive GIs and for introducing a detection specific moiety for enhanced detection, i.e. chromophore for UV. Also, this method sometimes produces a single compound for several GIs; thus, it becomes non-specific which can be considered as an advantage in determining a group of structurally related compounds (Liu *et al.*, 2010). Bai *et al.* (2010) introduced a chemical derivatization method for analyzing two alkyl halides and one epoxide. The objective of the three derivatization reactions is to generate a strong basic center by introducing an amine functional group. All three derivatization products are good candidates for electrospray ionization (ESI)-MS owing to the high proton affinity or the permanent charge.

#### **4.5.2 Coordination ion spray-MS**

Owing to their structural features, several analytes are not amenable to atmospheric pressure ionization methods, such as the ESI method. Alkali metal ions such as Li+, Na+, and K+ can form complexes with some organic molecules in the gas phase; this fact could be used as a solution for the analytes subjected previously (Liu *et al.*, 2010).

#### **4.5.3 Matrix deactivation**

The matrix deactivation approach is a chemical approach to stabilize unstable/reactive analytes. It is based upon the hypothesis that the instability of certain GIs at trace level is caused by the reaction between the analytes and reactive species in the sample matrix. Thus, controlling the reactivity of the reactive species in the sample matrix would stabilize the unstable/reactive GI analytes (Liu *et al.*, 2010).

As an example the alkylators are reactive unknown impurities which possess mainly nucleophilic characteristics. Their reactivity can be attenuated by either protonation or scavenging approaches. Sun *et al.* (2010) reported a matrix deactivation methodology for improving the stability of unstable and reactive GIs for their trace analysis. This approach appears to be commonly applicable to techniques like direct GC–MS and LC–MS analyses, or coupled with chemical derivatization as well.

#### **5. Genotoxicity prediction**

The concept of using structural alerts to predict potential genotoxic activity for identied impurities is now well established; however, the concordance between such alerts and biologically relevant genotoxic potential (in the context of genotoxic impurities) could be

Genotoxic Impurities in Pharmaceuticals 407

(molar refractivity), and ELUM0 (energy of the lowest unoccupied molecular orbital, indicating electrophilicity) were the most significant factors to be considered for

Benigni *et al.* (2005) showed that the QSAR models could correctly predict–– based only on the knowledge of the chemical structure––the genotoxicity of simple and unsaturated aldehydes. The active and inactive compounds were separated based on the hydrophobicity

Bercu *et al.* (2010) used *in silico* tools to predict the cancer potency (TD50) of a compound based on its structure. SAR models (classification/regression) were developed from the carcinogenicity potency database using MULTICASE and VISDOM (a Lilly Inc. in-house

It is commonly accepted that the carcinogenicity of chemicals is owing to their genotoxicity and, in fact, the mutation and carcinogenesis data are practically coincident. Thus, the two endpoints were collapsed into one ''genotoxicity'' classification, in which QSAR analysis was applied. Now the question remains as to how to predict non-genotoxic carcinogenicity. In fact, it cannot be well approached until some mechanistic understanding of nongenotoxic carcinogenesis is achieved. At this time, this approach is unable to grasp the structural features of non-genotoxic carcinogens (Ashby, 1990; Cunningham *et al.*, 1998;

The other limitation to currently available QSARs is the lack of models for organometallics, complex mixtures (e.g. herbal extracts), and high molecular weight compounds such as polymers (Valerio, 2009). However, the QSAR predictive software offers a rapid, reliable, and cost effective method of identifying the potential risk of chemicals that are well represented in QSAR training data sets, even when experimental data are limited or lacking (Kruhlak *et al.*, 2007). These models should be further developed/validated by employing

Since 2007, following the EMEA suspension of the marketing authorization of viracept (nelfinavir mesylate), genotoxic impurities have become a common issue for health concerns. Thus, regulatory agencies have made several attempts to construct a systematic method for controlling and analyzing GIs. However, several points must be considered for

One of the main problems is the very conservative limit regulated by agencies (1.5 µg/day). Bercu *et al.* (2009) calculated the permissible daily exposure (PDE) for EMS, which was the first GI of concern in 2007, as 0.104 mg/day. This value was found to be about 70-fold higher than the TTC level of 1.5 µg/day currently applied to EMS based on the generic linear back extrapolation model for genotoxins acting via non-threshold mechanisms. Other literatures highlighted this conservative limit as well (Gocke *et al.*, 2009b; Elder *et al.*, 2010a; Snodin, 2010). In addition, Gocke *et al.* (2009b) reported that the accidental exposure of viracept patients did not result in an increased likelihood for adverse genotoxic, teratogenic or

In addition to the challenge of setting a more pragmatic limit for GIs, the development of extremely sensitive and robust analytical methods that can adequately monitor GIs at very low levels is very difficult. Also, the pharmaceutical industry has no long-term experience in the use of these methodologies within the factory setting. Thus, analysts make attempts to

new mechanistic findings and using newly reported experimental data.

achieving a general view on the regulation of GIs.

discriminating between genotoxins and nongenotoxins.

(log P) and bulkiness (MR) properties.

software).

Benigni *et al.*, 2005).

**6. Conclusion** 

cancerogenic effects.

highly imperfect. Structural alerts are defined as molecular functionalities (structural features) that are known to cause toxicity, and their presence in a molecular structure alerts the investigator to the potential toxicities of the test chemical. Nevertheless, the assumption that any impurity with a structural alert is potentially DNA-reactive and thus subject to the default TTC limit may often lead to unnecessary restrictive limits. From a resource and time table viewpoint of a new drug production, the experimental determination of genotoxicity is not feasible for millions of drug candidates in the pharmaceutical industry. Thus, compounds identified as potential hazards by *in silico* methods would be high priority candidates for confirmatory laboratory testing (Kruhlak *et al.*, 2007; Snodin, 2010).

*In silico* toxicology is the application of computer technologies to analyze existing data, model, and predict the toxicological activity of a substance. In sequence, toxicologically based QSARs are mathematical equations used as a predictive technique to estimate the toxicity of new chemicals based upon a model of a training set of chemicals with known activity and a defined chemical space (Valerio, 2009).

Ashby and Tennant (1991) reported some correlations of electrophilicity with DNA reactivity (assessed by Ames-testing data) for about 300 chemicals and elucidated the concept of structural alerts for genotoxic activity in the 1980s/1990s. Using a database of >4000 compounds, Sawatari *et al.* (2001) determined correlations between 44 substructures and bacterial mutagenicity data. A high proportion of genotoxic compounds were found for electrophilic reagents such as epoxides (63 %), aromatic nitro compounds (49 %), and primary alkyl monohalides (46 %). In a retrospective analysis of starting materials and intermediates involved in API syntheses, the most common structurally alerting groups were found to be aromatic amines, aromatic nitros, alkylating agents and Michael acceptors (Snodin, 2010).

One of the strengths of QSAR models is that they contribute to a mechanistic understanding of the activity, and, at the same time, they constitute practical tools to predict the activity of further, untested chemicals solely based on chemical structure (Benigni *et al.*, 2005). Another strength of QSAR models is that they are strictly data-driven, and are not based on a prior hypotheses. On the other hand, high-quality experimental data must be used to build the training data set. As error (e.g. incorrect molecular structure or erroneous data from toxicology studies of a chemical) is introduced into the model, amplification of that error is generated and represented in the prediction (Benigni *et al.*, 2005; Valerio, 2009).

Cunningham *et al.* (1998) investigated a SAR analysis of the mouse subset of the carcinogenic potency database (CPDB) which also included chemicals tested by the US national toxicology program (NTP). This database consisted of 627 chemicals tested in mice for carcinogenic activity with the tumorigenicity data being standardized and reported as TD50 values. In addition, MULTICASE software (www.multicase.com) was used to identify several structural features that are not explained by an electrophilic mechanism and which may be indicative of non-genotoxic chemicals or mechanisms involved in carcinogenesis other than mutations. The prediction capabilities of the system for identifying carcinogens and noncarcinogens were 70 % and 78 % for a modified validation set.

Tafazoli *et al.* (1998) used the micronucleus (MN) test and the alkaline single cell gel electrophoresis (Comet) assay for analyzing potential mutagenicity, genotoxicty, and cytotoxicity of five chlorinated hydrocarbons. Using the generated data as well as the data of another five related chemicals that were investigated previously, a QSAR analysis was performed and the results indicated that LBC\_C1 (longest carbon-chlorine bond length), MR

highly imperfect. Structural alerts are defined as molecular functionalities (structural features) that are known to cause toxicity, and their presence in a molecular structure alerts the investigator to the potential toxicities of the test chemical. Nevertheless, the assumption that any impurity with a structural alert is potentially DNA-reactive and thus subject to the default TTC limit may often lead to unnecessary restrictive limits. From a resource and time table viewpoint of a new drug production, the experimental determination of genotoxicity is not feasible for millions of drug candidates in the pharmaceutical industry. Thus, compounds identified as potential hazards by *in silico* methods would be high priority

*In silico* toxicology is the application of computer technologies to analyze existing data, model, and predict the toxicological activity of a substance. In sequence, toxicologically based QSARs are mathematical equations used as a predictive technique to estimate the toxicity of new chemicals based upon a model of a training set of chemicals with known

Ashby and Tennant (1991) reported some correlations of electrophilicity with DNA reactivity (assessed by Ames-testing data) for about 300 chemicals and elucidated the concept of structural alerts for genotoxic activity in the 1980s/1990s. Using a database of >4000 compounds, Sawatari *et al.* (2001) determined correlations between 44 substructures and bacterial mutagenicity data. A high proportion of genotoxic compounds were found for electrophilic reagents such as epoxides (63 %), aromatic nitro compounds (49 %), and primary alkyl monohalides (46 %). In a retrospective analysis of starting materials and intermediates involved in API syntheses, the most common structurally alerting groups were found to be aromatic amines, aromatic nitros, alkylating agents and Michael acceptors

One of the strengths of QSAR models is that they contribute to a mechanistic understanding of the activity, and, at the same time, they constitute practical tools to predict the activity of further, untested chemicals solely based on chemical structure (Benigni *et al.*, 2005). Another strength of QSAR models is that they are strictly data-driven, and are not based on a prior hypotheses. On the other hand, high-quality experimental data must be used to build the training data set. As error (e.g. incorrect molecular structure or erroneous data from toxicology studies of a chemical) is introduced into the model, amplification of that error is

Cunningham *et al.* (1998) investigated a SAR analysis of the mouse subset of the carcinogenic potency database (CPDB) which also included chemicals tested by the US national toxicology program (NTP). This database consisted of 627 chemicals tested in mice for carcinogenic activity with the tumorigenicity data being standardized and reported as TD50 values. In addition, MULTICASE software (www.multicase.com) was used to identify several structural features that are not explained by an electrophilic mechanism and which may be indicative of non-genotoxic chemicals or mechanisms involved in carcinogenesis other than mutations. The prediction capabilities of the system for identifying carcinogens

Tafazoli *et al.* (1998) used the micronucleus (MN) test and the alkaline single cell gel electrophoresis (Comet) assay for analyzing potential mutagenicity, genotoxicty, and cytotoxicity of five chlorinated hydrocarbons. Using the generated data as well as the data of another five related chemicals that were investigated previously, a QSAR analysis was performed and the results indicated that LBC\_C1 (longest carbon-chlorine bond length), MR

generated and represented in the prediction (Benigni *et al.*, 2005; Valerio, 2009).

and noncarcinogens were 70 % and 78 % for a modified validation set.

candidates for confirmatory laboratory testing (Kruhlak *et al.*, 2007; Snodin, 2010).

activity and a defined chemical space (Valerio, 2009).

(Snodin, 2010).

(molar refractivity), and ELUM0 (energy of the lowest unoccupied molecular orbital, indicating electrophilicity) were the most significant factors to be considered for discriminating between genotoxins and nongenotoxins.

Benigni *et al.* (2005) showed that the QSAR models could correctly predict–– based only on the knowledge of the chemical structure––the genotoxicity of simple and unsaturated aldehydes. The active and inactive compounds were separated based on the hydrophobicity (log P) and bulkiness (MR) properties.

Bercu *et al.* (2010) used *in silico* tools to predict the cancer potency (TD50) of a compound based on its structure. SAR models (classification/regression) were developed from the carcinogenicity potency database using MULTICASE and VISDOM (a Lilly Inc. in-house software).

It is commonly accepted that the carcinogenicity of chemicals is owing to their genotoxicity and, in fact, the mutation and carcinogenesis data are practically coincident. Thus, the two endpoints were collapsed into one ''genotoxicity'' classification, in which QSAR analysis was applied. Now the question remains as to how to predict non-genotoxic carcinogenicity. In fact, it cannot be well approached until some mechanistic understanding of nongenotoxic carcinogenesis is achieved. At this time, this approach is unable to grasp the structural features of non-genotoxic carcinogens (Ashby, 1990; Cunningham *et al.*, 1998; Benigni *et al.*, 2005).

The other limitation to currently available QSARs is the lack of models for organometallics, complex mixtures (e.g. herbal extracts), and high molecular weight compounds such as polymers (Valerio, 2009). However, the QSAR predictive software offers a rapid, reliable, and cost effective method of identifying the potential risk of chemicals that are well represented in QSAR training data sets, even when experimental data are limited or lacking (Kruhlak *et al.*, 2007). These models should be further developed/validated by employing new mechanistic findings and using newly reported experimental data.
