7.1 Introduction

Detection and measurement of drugs, drug metabolites, and steroid hormones in body fluids is commonly used in clinical medicine and forensic science [1–4]. For example, blood concentrations of steroid hormones such as cortisol, estradiol, progesterone, and testosterone assist in the evaluation of endocrinology and reproductive disorders. Detection of anabolic steroids that are potentially abused as performance-enhancing drugs is important in competitive athletics [5, 6]. Therapeutic drug monitoring (TDM) involves determination of serum/plasma concentrations of medications and/or metabolites to guide drug dosing and avoid toxicity [7]. Lastly, drug of abuse and toxicology (DOA/Tox) analyses are used widely in emergency medicine, management of patients on pain medications, competitive athletics, and forensics [8].

The two main technologies for clinical analysis of drug, drug metabolites, and steroid hormones are immunoassays (antibody-based assays) [2, 3, 9] and chromatography/mass spectrometry (MS) [10, 11]. Immunoassays use polyclonal or monoclonal antibodies, with an increasing trend toward monoclonal antibody-based assays [2]. A basic immunoassay approach is exemplified by enzyme-linked immunosorbent assay (ELISA) which uses antibodies bound to a solid support such as multiwell microplates. Modern clinical chemistry analyzers often employ “homogeneous” immunoassays where the entire analysis occurs in the liquid phase. The advantages of immunoassays include low technical complexity for laboratory staff, high throughput, and wide commercial availability of Food and Drug Administration (FDA)-cleared assays. However, one of the main challenges with immunoassays is cross-reactivity with compounds that are structurally related to the target molecule(s) of the assay [12]. For TDM or measurement of steroid hormones, cross-reactivity can lead to potentially misleading results [7]. On the other hand, immunoassay cross-reactivity can be useful for DOA/Tox analysis in allowing for detection of a class of drugs such as benzodiazepines or opiates [7]. The usefulness of the assay becomes a balance of how well the immunoassay detects the intended target(s) compared to the assay's cross-reactivity with unintended targets [13, 14].

Chromatography with or without MS represents the other common approach for drug and steroid hormone analysis in body fluids [10, 11, 15, 16]. Gas chromatography (GC) and high-performance liquid chromatography (HPLC or simply LC) are used either on their own or in combination with MS. GC/MS and LC/MS/MS (HPLC coupled with tandem mass spectrometry) can provide definitive and specific identification of drugs, drug metabolites, and steroid hormones. Chromatography/MS methods used in DOA/Tox analysis are often used to confirm (verify) positive screening results obtained using initial analysis by immunoassay or other similar technology [10, 11]. MS-based methods are also increasingly used for the analysis of compounds that are not detected by commercially available immunoassays. These include the broad and very diverse class of “designer drugs” such as the amphetamine-like stimulants (popularly referred to by names such as “bath salts” and “plant food”), synthetic cannabinoids, or novel anabolic steroids abused as performance-enhancing drugs [17, 18]. MS-based methods also allow for the differentiation of compounds that are very close in structure such as steroid hormones or vitamins and their metabolic intermediates or synthetic analogs [15, 16].

While a steadily increasing number of larger clinical laboratories are adopting MS-based techniques, relatively few hospital- or clinic-based clinical laboratories use this technology. The main barriers for adoption include high capital cost of instrumentation (e.g., LC/MS/MS analyzers typically have purchase prices of $200,000-300,000 USD), technical complexity of operation, and labor-intensive steps in sample preparation and results analysis [15, 16]. There are also relatively few FDA-cleared chromatography/MS assays available in the United States for the analysis of drugs, drug metabolites, and steroid hormones. This places the burden on clinical laboratories to validate their own assays, placing this testing in the category of “laboratory-developed tests” [19]. In terms of United States regulations under the Clinical Laboratory Improvement Amendments of 1988 (CLIA'88 or simply CLIA), laboratory-developed tests are placed in the highest complexity category for laboratory assays, which have the most stringent requirements for the qualifications of testing personnel and supervisory staff. In contrast, most marketed immunoassays for drug or steroid hormone analysis are in the CLIA moderate or waived complexity categories that have less stringent constraints than high-complexity tests. Thus, many hospitals and clinics refer chromatography/MS-based testing to off-site commercial reference laboratories, resulting in much slower turnaround time than automated immunoassays.

Given that immunoassays will likely continue to be used widely in clinical medicine and forensics, better understanding and prediction of immunoassay cross-reactivity has multiple useful applications [13, 14, 20, 21]. First, computational algorithms can help prioritize cross-reactivity testing studies. This becomes especially important given that some of the compounds of interest are controlled substances (potentially requiring special licenses for acquisition) or drug metabolites that may be difficult and/or costly to obtain. Second, computational methods can rationalize existing in vitro cross-reactivity data and help extrapolate whether compounds observed to cross-react in one immunoassay are likely to affect other immunoassays. Third, computational prediction of cross-reactivity can help formulate hypotheses generated from toxicology data. For instance, early toxicology case reports of designer drug abuse may have data such as otherwise unexplained positive immunoassay drug screens [14]. Computational methods provide a means to evaluate likelihood that a given compound will cross-react with a particular immunoassay. Lastly, computational modeling of cross-reactivity can provide insight into the complex interactions between antibodies and their targets. This chapter reviews the application of computational methods for understanding and predicting immunoassay cross-reactivity, using this as a model system for applying cheminformatics to a clinical application.

7.2 Similarity Analysis Applied to Drug of Abuse/Toxicology Immunoassays

DOA/Tox screens are widely used in a variety of clinical settings such as emergency departments, substance abuse treatment programs, and pain management clinics [8]. Although there are multiple methods that can provide rapid detection of drugs and drug metabolites, immunoassays currently represent the most widely used methodology [3, 22]. Immunoassays have mostly replaced alternative technologies such as thin-layer chromatography and chemical spot assays, although these older methods still have niche applications (e.g., spot assays for field testing of confiscated narcotics). DOA/Tox screens are available in a broad range of formats from point-of-care kits to assays performed on high-throughput clinical chemistry analyzers. The most common specimens for analysis are urine, saliva (oral fluid), and blood [8]. Emerging sample types include hair and fingernails, both of which require extensive specimen processing but have the benefit of a wider detection window than urine, oral fluid, or blood [23]. In the specialized realm of newborn drug testing, umbilical cord tissue and meconium (newborn's first stools) are the most common specimens to detect maternal drug use over the course of the pregnancy [24].

Immunoassays commonly are used as a positive/negative “screen” and as such can provide a rapid qualitative assessment whether a class of drugs is present in a patient sample [3, 8, 22]. Depending on the setting, samples showing positive screens may be further tested by confirmatory methods such as GC/MS or LC/MS/MS. Common targets of DOA/Tox screens are individual drugs or drug classes such as amphetamines (e.g., amphetamine, methamphetamine), benzodiazepines (e.g., alprazolam, clonazepam, diazepam, lorazepam, midazolam), cocaine, opiates (e.g., codeine, heroin, hydrocodone, morphine, oxycodone), and tetrahydrocannabinol (THC; the active component of marijuana/cannabis). DOA/Tox screening presents differing challenges based on the specific target(s) of the assay [14]. In particular, an assay intended to detect cocaine use can focus on the specific metabolite benzoylecgonine without need to cross-react with other molecules. In contrast, a clinically useful benzodiazepines screening assay ideally would detect the commonly used benzodiazepines (and/or metabolites) while not cross-reacting with other “off-target” compounds.

Data on DOA/Tox immunoassay cross-reactivity may be found in the assay package inserts provided by the manufacturer [14]. Assay manufacturers typically test a variety of drugs and metabolites within the targeted class along with frequently used medications (e.g., acetaminophen, diphenhydramine, salicylates) that would also be commonly found in patient samples. For some immunoassays, there is also data in the published literature related to cross-reactivity that has been discovered after assay marketing. Classic examples of post-marketing published reports of immunoassay cross-reactivity include fluoroquinolone antibiotic (e.g., ciprofloxacin) cross-reactivity with opiates assays [25, 26], fentanyl cross-reactivity with lysergic acid diethylamide (LSD) immunoassays [27], and sertraline (antidepressant) cross-reactivity with benzodiazepines assays [28]. Overall, the amount of cross-reactivity data reported in the package inserts for marketed DOA/Tox immunoassays varies considerably, with extensive data reported in some package inserts and minimal data reported in others [14]. Likewise, the published literature on immunoassay cross-reactivity is highly variable, with more attention on cross-reactivities likely to have higher clinical or forensic impact [13, 14].

There is thus an opportunity to utilize computational methods to provide a more systematic conceptual approach to DOA/Tox immunoassay cross-reactivity. As proof of concept, we used the in silico method of molecular similarity analysis, which determines the similarity between molecules independent of any in vitro or in vivo data [13, 14]. Molecular similarity can be assessed at the one-dimensional (1D), two-dimensional (2D), and three-dimensional (3D) levels [29–34]. Common 2D methods use fragment bit strings compared by use of the Tanimoto coefficient which scales the results from 0 (maximally dissimilar) to 1 (maximally similar). 3D-similarity methods typically involve the determination of a pharmacophore pattern that models how the arrangement of chemical features and distances between them are associated with biological activity [35].

For our initial studies using similarity approaches, we compiled cross-reactivity data for 84 marketed DOA/Tox immunoassays covering 18 classes of drugs [14]. Cross-reactivity studies for DOA/Tox immunoassays generally report the concentration of a compound that produces reactivity equivalent to a specified concentration of the target molecule of the assay [13, 14]. For example, opiate assays customarily use morphine as the target molecule, with a cutoff of either 300 or 2,000 ng/mL as the reference for positivity (the latter higher cutoff used in employment drug testing). For each DOA/Tox immunoassay, compounds were assigned to the following categories: strong true positives (high degree of cross-reactivity for an intended target of the assay), weak true positive (low degree of cross-reactivity for an intended target of the assay), strong false positive (high degree of cross-reactivity for an off-target compound), weak false positive (low degree of cross-reactivity for an off-target compound), true negative (no cross-reactivity for an off-target compound), and false negative (no cross-reactivity for an intended target of the assay). To this data, we applied two different 2D similarity methods (MDL public keys; long-range functional class fingerprint – FCFP - description six keys) along with 3-point and 4-point pharmacophore-based fingerprints [14]. Figure 7.1 shows an example of 2D similarity applied to the drug of abuse phencyclidine (PCP, also known as “angel dust”) as the target compound. Of the five other compounds shown, 2D similarity is highest to 4-phenyl-4-piperidino-cyclohexanol (PCP metabolite) and two compounds (dextromethorphan and meperidine) reported to cross-react with PCP immunoassays [14, 36]. PCP has low similarity to ketamine (a drug with similar pharmacologic properties) and essentially no 2D similarity to ibuprofen (a common over-the-counter medication). Neither ketamine nor ibuprofen have been reported to cross-react with PCP immunoassays.

At a broad level, MDL public keys similarity measures were significantly associated with cross-reactivity for DOA/Tox immunoassays [14]. For the MDL public keys data, approximately 46% of the strongly cross-reactive compounds (either false or true positives) had Tanimoto similarity coefficients of 0.8 or higher to the target compound of the assay. In contrast, only 1.4% of true negatives have similarity coefficients in this range. Thus, a cutoff of 0.8 has a positive predictive value of approximately 78% in identifying compounds capable of strong cross-reactivity versus the true negatives. Conversely, 66% of true negatives have similarity coefficients of less than 0.4 to the target compound of the assay compared to only 0.2% (3 of 1,681) for strongly cross-reactive compounds. Interestingly, the three compounds with strong cross-reactivity but low 2D similarity to the target molecule of the assay were all amphetamine derivatives cross-reacting with amphetamines immunoassays - 3,4-methylenedioxymethamphetamine (MDMA/Ecstasy); 3,4-methylenedioxy-α-ethyl-N-methylphenethylamine (MDBD); and 3,4-methylenedioxy-N-ethylamphetamine (MDEA). Adopting a lower cutoff of 0.4 would have a negative predictive value of 99.8% in distinguishing true negatives from strongly cross-reactive compounds.

Similarity comparisons help illustrate the challenges of broadly specific DOA/Tox immunoassays intended to detect multiple drugs within a class [14]. Figure 7.2 shows similarity data using MDL public keys for cross-reactivity of marketed amphetamines, barbiturates, benzodiazepines, and tricyclic antidepressant immunoassays. Amphetamine immunoassays generally have as their intended detection target the most clinically relevant amphetamines, namely, amphetamine, methamphetamine, and MDMA/Ecstasy. Amphetamine immunoassays often have either amphetamine or methamphetamine as the target hapten for immunoassay design [14]. However, there are other compounds such as phentermine that have equal or greater similarity to amphetamine or methamphetamine than illicit amphetamines or amphetamine-like drugs such as MDEA or MDBD (Figure 7.2a). Barbiturate immunoassays typically have a short-to-intermediate acting barbiturate such as secobarbital as the target hapten [14]. The cross-reactivity profile of barbiturates immunoassays is generally good, with the exception that certain barbiturates such as methohexital are not detected (Figure 7.2b).

Benzodiazepine immunoassays have classically used diazepam or related metabolites such as nordiazepam as the target hapten for immunoassay design [14]. One of the main challenges for benzodiazepine immunoassays is that there are currently a number of benzodiazepines used clinically or illicitly [37]. The 2D similarity of some benzodiazepines to diazepam is relatively low and overlaps with some out-of-class compounds (Figure 7.2c). Tricyclic antidepressants (TCAs) represent a class of medications used for the treatment of major depression, obsessive-compulsive disorder, chronic pain, insomnia, and a variety of other conditions [38, 39]. As illustrated in Figure 7.2(d), TCAs share structural similarity to a variety of other clinically relevant compounds, resulting in a fairly large number of off-target compounds capable of producing positive screening results in TCA immunoassays. The propensity for false positives significantly limits the clinical utility of TCA immunoassays, as a high percentage of positive screens may be attributable to off-target compounds (e.g., cyclobenzaprine) in the clinical setting.

Cannabinoid immunoassays generally use δ-9-tetrahydrocannabinol (δ-9-THC) or related THC metabolites as the target [14]. The 2D similarity of THC to its own metabolites is much higher than to out-of-class compounds (Figure 7.3a). Most marketed cocaine immunoassays target the primary metabolite benzoylecgonine [13, 14]. With the exception of the related drug atropine, benzoylecgonine has little 2D similarity to other clinically relevant drugs (Figure 7.3b).

Classic opiates are compounds such as codeine and morphine that are derived from the opium poppy [40]. Semisynthetic derivatives of opiates include buprenorphine, heroin (3,6-diacetylmorphine), hydrocodone, oxycodone, and oxymorphone. Opioids is a term that encompasses opiates as well as synthetic compounds such as fentanyl that do not contain the core opiate structure and are thus of a different structural class of drugs [41]. Other examples of non-opiate opioids include meperidine, methadone, and propoxyphene. Opioids are among the most heavily prescribed medications in the United States. Abuse of prescription opioids is a major public health concern, with over 20,000 deaths in the United States attributed to overdose of these drugs [42]. Figure 7.3c shows 2D similarity data for opiate immunoassays that use morphine as the molecular target. The strong true positives have higher similarity to morphine than compounds causing false positives. Several fluoroquinolone antibiotics that can cause false positives on some opiate assays [42] have similarity coefficients between 0.5 and 0.6 using MDL public keys. In contrast, buprenorphine does not generally cross-react with standard opiate assays despite a similarity of approximately 0.8 with morphine. Likewise, oxycodone (one of the most heavily prescribed opiates) [43], along with its metabolite oxymorphone, cross-reacts only weakly with most of the marketed opiates immunoassays. Thus, designers of opiate immunoassays face a difficult challenge in cross-reacting with clinical opiates without unintended cross-reactivity with off-target compounds such as fluoroquinolone antibiotics [13, 14].

PCP is a drug of abuse that has faded considerably in use since the 1980s and 1990s [44]. As illustrated in Figure 7.3d, the intended targets of PCP immunoassays (namely, the parent drug and unique metabolites) have higher 2D similarity to PCP than compounds that are false positives or true negatives, consistent with generally good clinical performance of PCP immunoassays. Nevertheless, there are some compounds that can produce PCP immunoassay false positives under certain conditions. A good example is dextromethorphan, an opioid-like drug that is a common component of over-the-counter cough suppressant and cold/flu medications. At recommended doses, urine concentrations of dextromethorphan do not reach levels that will cross-react with PCP immunoassays. However, dextromethorphan is sometimes abused, often by teenagers and young adults who rapidly consume very large doses of dextromethorphan (typically obtained over-the-counter or via the Internet) to achieve intoxication [36, 45–47]. At these very high doses, urine concentrations of dextromethorphan will cross-react with PCP immunoassays. Similarly, meperidine also has weak cross-reactivity with PCP immunoassays but may cause immunoassay positivity in individuals who consume very high doses of meperidine [14].

Overall, 2D similarity studies of DOA/Tox immunoassays performed better using the MDL public keys than with FCFP [14]. The main limitation of using FCFP was that the similarity coefficients for true negatives overlapped more substantially with those for true positives, limiting the positive predictive value. FCFP performs particularly poorly with immunoassays for benzodiazepines, opiates, propoxyphene, and TCAs, with many true positives having lower similarity coefficients than the average similarity for true negatives. In our proof-of-concept studies, we also explored 3D similarity classification approaches by using 3- or 4-point pharmacophore fingerprints. However, we found that even with extensive exploration, these algorithms were too restrictive and missed true positives, including many with known strong cross-reactivity.

We utilized similarity analyses to predict additional cross-reactive compounds for 10 immunoassays (amphetamines, barbiturates, benzodiazepines, cocaine, cannabinoids, methadone, opiates, PCP, propoxyphene, and TCAs) [14]. We tested 46 such compounds and identified 14 previously unreported cross-reactivities for two marketed immunoassays. Eight of the cross-reactivities had not previously been reported for any marketed immunoassay for the respective class of drugs. The results of these studies indicate that computational predictions can be used to enhance cross-reactivity experimental testing.

7.3 Similarity Analysis Applied to Therapeutic Drug Monitoring Immunoassays

Analysis of drugs and drug metabolites in body fluids for TDM frequently uses immunoassays [7]. In contrast to DOA/Tox assays, which are frequently utilized as qualitative screens for classes of drugs, TDM assays generally need to provide an accurate quantitative concentration of a single compound, typically either a parent drug or an active metabolite [21]. The most common specimen for TDM is blood (often serum or plasma but sometimes whole blood), although other body fluids such as urine or cerebrospinal fluid may be analyzed in certain situations [7]. Immunoassays used for TDM ideally have high specificity for the target of the assay, with minimal or no cross-reactivity with structurally related compounds [21]. Achieving high assay specificity becomes especially challenging when there are multiple metabolites or other structurally similar drugs that may be present. The quantitative nature of TDM means that assay interference can potentially produce misleading values that can lead to inappropriate clinical decisions [12].

Cross-reactivity data for TDM immunoassays is usually generated by either of two main methods [21]. In the first approach, the potential cross-reactive compound is tested alone and its signal compared to that of a defined concentration of the target compound. In the other common approach, the potential cross-reactive compound is tested together with a specified concentration of the target compound and its signal expressed as an equivalent concentration of the target compound. The two approaches have subtle strengths and weaknesses. For example, the first approach can better define whether a cross-reactive compound can produce an erroneous signal when the target drug is not present at all. The second approach replicates the scenario when a patient is taking the target drug but also another drug that cross-reacts with the TDM immunoassay. Occasionally, package inserts simply state that a compound “cross-reacts” (or similar language) without providing quantitative data.

For proof of concept in the application of computational studies to TDM immunoassay cross-reactivity, 96 marketed versions of 28 TDM immunoassays were analyzed, compiling data from assay package inserts and published literature [21]. For each assay, compound cross-reactivity was broadly classified into Strong Cross-Reactives, Weak Cross-Reactives, and Non-Cross-Reactives. Similar to the DOA/Tox assays discussed in the previous section, similarity measures using MDL public keys performed better than FCFP coefficients in separating out Strong and Weak Cross-Reactives from Non-Cross-Reactives. Using MDL public keys, approximately 60% of the Strong Cross-Reactives had similarity coefficients of at least 0.8 relative to the target molecule of the immunoassay, while only 3.2% of Non-Cross-Reactives had similarity coefficients of 0.8 or higher. Conversely, nearly 50% of Non-Cross-Reactives had similarity coefficients of less than 0.4 to the target molecules, while only 5% of Strong Cross-Reactives fit in this category. There were only five TDM immunoassays (carbamazepine, gentamicin, salicylates, tobramycin, and topiramate) that had examples of Strong Cross-Reactives with similarity coefficients of 0.5 or lower. In all cases, the Strong Cross-Reactives with these low similarity coefficients were reported for only a single marketed assay, suggesting that unique features of the specific assay may contribute to interference by these compounds that have low structural similarity to the target compound of the immunoassay.

Similarity comparisons demonstrate the varying challenges of achieving specificity for individual drugs using TDM immunoassays [21]. Figure 7.4 shows similarity data using MDL public keys for cross-reactivity of marketed cyclosporine (immunosuppressant), lamotrigine (anticonvulsant), theophylline (used for treatment of asthma and newborn apnea), and valproic acid (anticonvulsant). In the case of cyclosporine, only metabolites of the parent drug show high specificity to the target compound (Figure 7.4a). In clinical practice, cyclosporine immunoassays consistently give higher blood concentrations on average than chromatographic assays, likely due primarily to cross-reactivity of the immunoassays with one or more cyclosporine metabolites [48]. Tacrolimus (another immunosuppressant) immunoassays contend with not only the high similarity of tacrolimus metabolites but also the closely related drug sirolimus, another immunosuppressant that might be used concomitantly with tacrolimus in organ transplant recipients. Similar to cyclosporine, tacrolimus immunoassays often give higher blood concentrations than chromatographic assays owing to metabolite cross-reactivity with the immunoassay [49]. Lamotrigine is a newer generation anticonvulsant drug that has little 2D similarity to other clinically relevant drugs and even to its own metabolites (Figure 7.4b). In contrast to cyclosporine and tacrolimus immunoassays, lamotrigine immunoassays perform very similarly to chromatographic assays, indicating that immunoassay cross-reactivity with metabolites is minimal [50].

Theophylline immunoassays encounter the challenge of differentiating theophylline from other methylxanthines such as caffeine and theobromine (found in chocolate and tea) [51, 52] (Figure 7.4c). This cross-reactivity can especially be an issue in adults, who commonly ingest caffeine and theobromine from beverages or foods. In contrast to the above assays, valproic acid has fairly low similarity to its own metabolites and also to other clinically relevant drugs, allowing for high assay specificity [53] (Figure 7.4d).

7.4 Similarity Analysis Applied to Steroid Hormone Immunoassays

Immunoassays are frequently used in the clinical setting for quantitation of blood and urine concentrations of steroid hormones such as cortisol, estradiol, and testosterone [1, 4]. Similar to DOA/Tox and TDM analysis, the most common alternative approach for measurement of steroid hormones are chromatographic/MS-based assays [15, 16]. Although the use of chromatographic assays has been increasing, many clinical laboratories continue to use immunoassays for routine steroid hormone analysis.

Cross-reactivity is one of the major limitations of steroid hormone immunoassays. Interfering compounds can be structurally related endogenous molecules (e.g., 17-hydroxyprogesterone for progesterone immunoassays), drugs (including herbal compounds and performance-enhancing drugs), or natural products [12, 54]. Cross-reactivity data is included in manufacturer package inserts and in the published literature. Using a similar approach to the TDM immunoassays described above, we classified steroid hormone immunoassay cross-reactivity data into Strong Cross-Reactives, Weak Cross-Reactives, Very Weak Cross-Reactives, and Non-Cross-Reactives [20]. For some of the potentially cross-reactive compounds, there are published reference ranges for concentrations in blood. This allows for an estimate of the apparent concentration in blood that will be caused by a given cross-reactive compound. This is only a rough estimate as there are multiple factors that influence cross-reactivity in actual patient samples.

We studied cross-reactivity for five steroid hormones: cortisol, dehydroepiandrosterone (DHEA) sulfate, estradiol, progesterone, and testosterone. DHEA sulfate and estradiol immunoassays did not have any compounds that showed strong cross-reactivity [20]. For DHEA sulfate immunoassays, only one compound (pregnenolone sulfate) was estimated to produce any clinically significant impact on apparent DHEA sulfate measurements, and this was only in the setting of the very highest pregnenolone sulfate concentrations in pregnancy. For estradiol immunoassays, no compound was estimated to produce clinically relevant cross-reactivity. Even estriol, which becomes the major estrogen during pregnancy, has weak enough cross-reactivity to not impact apparent estradiol concentrations.

For cortisol, progesterone, and testosterone immunoassays, all compounds with documented strong cross-reactivity had 2D similarity coefficients of 0.8 or higher using MDL public keys, with the majority in this group exceeding 0.9. 2D similarity was less successful at differentiating Weak Cross-Reactives, Very Weak Cross-Reactives, and Non-Cross-Reactives [20]. 2D similarity may be insufficient to resolve the subtle features involved in causing weak versus no cross-reactivity. Figures 7.5 and 7.6 demonstrate cross-reactivity and similarity predictions for cortisol and testosterone immunoassays, respectively. Figure 7.5 shows that two common cortisol analogs, prednisolone and 6-methylprednisolone, can produce cross-reactivity resulting in apparent cortisol concentrations that overlap or even exceed typical reference ranges for humans [55]. Similarly, Figure 7.6 shows that 6-methyl testosterone (an anabolic steroid) can produce clinically significant cross-reactivity in males and females, whereas nandrolone (another anabolic steroid) and norethindrone (a progestin commonly found in female oral contraceptives) can cause increases in apparent testosterone concentrations that would be clinically relevant in females.

Overall, the computational studies with steroid hormone immunoassays demonstrate that 2D similarity measurements can narrow the search for strongly cross-reactive compounds, all of which had high similarity coefficients (0.8 or higher) [20]. This feature can be useful in prioritizing compounds for cross-reactivity testing. For example, the list of anabolic steroids abused as performance-enhancing drugs continues to expand. Definitive identification of novel anabolic steroids requires sophisticated chromatographic/MS analysis, as may be done in international athletics competitions [5, 6]. Anabolic steroids and other performance-enhancing drugs may be also used in animals such as racehorses [56]. Indeed, for some of the anabolic steroids, there are much more detailed pharmacokinetic studies in horses than in humans [57].

7.5 Cheminformatics Applied to “Designer Drugs”

“Designer drugs” are a varied group of psychoactive compounds typically discovered by modifications of existing drug classes such as amphetamines, cannabinoids, and opioids [17, 18, 58–67]. Two current broad categories of designer drugs are the amphetamine-like stimulants [17, 59, 65] and the synthetic cannabinoids [58, 60, 67, 68], each of which encompasses a chemically diverse set of chemical structures. Figures 7.7 and 7.8 show representative chemical structures of these two classes of designer drugs. The amphetamine-like stimulants are related to amphetamine, methamphetamine, and MDMA/Ecstasy and are sometimes referred to as “bath salts,” stemming from a misleading sales name under which some of these products were sold [17, 59, 65]. Detailed descriptions of the chemical synthesis and psychoactive effects of these compounds is accessible in the published literature, including two books by Shulgin and Shulgin that describe hundreds of compounds in detail [69, 70]. The amphetamine-like drugs can be further broken down into subcategories such as 2C compounds, β-keto amphetamines, piperazines, and tryptamine (Figure 7.7). Methylone, methylenedioxypyrovalerone (MDPV), and mephedrone are three of the more common amphetamine-like designer drugs. The illegal distribution patterns of amphetamine-like drugs shift as law enforcement and regulatory agencies target specific drugs within this class. Within the United States, increasing numbers of amphetamine-like drugs are designated as Schedule I controlled substances (no valid medical use and illegal for use outside of restricted research protocols), which limits the ability to sell and distribute these compounds [71].

The synthetic cannabinoids are also chemically diverse (Figure 7.8) and were originally synthesized and evaluated for medical applications [58, 60, 67, 68]. Similar to THC from cannabis/marijuana, the synthetic cannabinoids bind to cannabinoid receptors in the nervous system, producing a range of psychoactive and neurological effects. However, clandestine laboratories began to produce synthetic cannabinoids for personal use, often in products marketed as “legal highs” (with misleading descriptions such as “incense,” “potpourri,” “K2,” or “spice,” or simply designated as “not for human consumption” or “for research use only”). Many of the synthetic cannabinoids were discovered by J.W. Huffman; hence, many are designated as part of the “JWH” series [72]. JWH-018 [1-naphthyl-(1-pentylindol-3-yl)methanone] and JWH-250 [2-(2-methoxyphenyl)-1-(1-pentylindol-3-yl)] are two of the most commonly abused synthetic cannabinoids. As shown in Figure 7.8, some synthetic cannabinoids such as JWH-018 (a naphthoylindole) differ considerably in chemical structure from THC, while other compounds such as HU-210 (1,1-dimethylheptyl-11-hydroxy-THC) are analogs of THC.

Detection of designer amphetamine-like stimulants and synthetic cannabinoids presents a substantial technical and logistic challenge for clinical and forensic toxicology laboratories [73]. These compounds can generally be detected by chromatography/MS methods; however, maintaining methods to keep pace with shifting trends in designer drug use is difficult for all but the most dedicated laboratories with expertise in this area. DOA/Tox screening immunoassays based on amphetamine, methamphetamine, and/or MDMA/Ecstasy as the target molecule(s) will only detect a small subset of designer amphetamine-like drugs [74, 75]. THC immunoassays generally do not show cross-reactivity with synthetic cannabinoids that lack the classic cannabinoid chemical backbone found in THC, a finding consistent with similarity analysis. In general, the synthetic cannabinoids show very low 2D similarity to 9-carboxy-THC, with very few compounds having a Tanimoto coefficient greater than 0.5. Despite the technical complications, some manufacturers have marketed ELISA immunoassays for the amphetamine-like stimulants and synthetic cannabinoids [76–80].

Figure 7.9 shows 2D similarity using MDL public keys applied to the target molecules used in marketed synthetic cannabinoids immunoassays, namely, prominent urinary metabolites of JWH-018, JWH-073 [1-butyl-1H-indol-3-yl)-1-naphthenyl-methanone], and JWH-250 [74]. As can be seen, the 2D similarity of synthetic cannabinoids to these target molecules varies considerably, even within the JWH series. This is consistent with the observed cross-reactivity of these assays in detecting the more closely structurally related compounds and not other synthetic cannabinoids [77]. This limits the practicality of immunoassays for synthetic cannabinoids, especially given the assay development and regulatory costs involved in bringing in vitro assays to the market [73].

The amphetamine-like stimulants represent a somewhat different challenge from the synthetic cannabinoids [73]. Many of these stimulants are based on the same core structure as amphetamine and methamphetamine, resulting in some cross-reactivity with amphetamines immunoassays. To better predict the cross-reactivity of amphetamine-like stimulants for amphetamine screening immunoassays, we sorted 42 amphetamine-like compounds into categories based on whether the compounds caused a clinically positive result at 5,000 ng/mL, at 20,000 but not 5,000 ng/mL, or only at 100,000 ng/mL [75]. While the pharmacokinetics of these compounds are varied, urine concentrations can often reach 5,000–20,000 ng/mL in abusers but would exceed 20,000 ng/mL only in extreme overdoses. The dataset included three widely used, commercially available immunoassays: Abbott Diagnostics, CEDIA (cloned enzyme donor immunoassay), and EMIT (enzyme-multiplied immunoassay technique). Receiver operator characteristic (ROC) curve analysis compared various 2D similarity methods (including consensus methods that averaged similarity between the test molecule and target compounds such as amphetamine and MDMA/Ectasy) as well as 3D shape and pharmacophore. In general, 2D consensus methods performed well. However, there were differences among the data for the different immunoassay techniques. In particular, 2D consensus methods performed well for Abbott and EMIT data, while 3D methods performed well for CEDIA data. We utilized the best-performing modeling methods in predicting the likelihood that various other amphetamine-like compounds would cause cross-reactivity.

One way to rationalize these findings is that amphetamine screening immunoassays have been designed by manufacturers to cross-react well with amphetamine, methamphetamine, and sometimes MDMA/Ecstasy, as these have been traditionally the most clinically relevant amphetamines [74, 75]. An antibody cross-reacting only with amphetamine but not methamphetamine or vice versa would have limited clinical utility. However, manufacturers have adopted different methodologies and approaches for their immunoassays to achieve similar purposes. Interestingly, the CEDIA amphetamines assay uses three different monoclonal antibodies (using amphetamine, methamphetamine, and MDMA/Ectasy, respectively, as the haptens) [75]. The superiority of a 3D pharmacophore based on methamphetamine in predicting CEDIA cross-reactivity for amphetamine-like compounds may suggest that the pharmacophore for methamphetamine represents a “middle ground” between the smaller amphetamine molecule and the larger MDMA/Ectasy molecule that contains methamphetamine within its substructure.

The computational modeling methods applied to amphetamine-like stimulants and synthetic cannabinoids should also be applicable to new drugs of abuse such as designer opioids and benzodiazepines [81–83]. Reports of abuse and toxicity of these compounds have recently emerged. Similarly to the amphetamine-like stimulants and synthetic cannabinoids, the designer opioids and benzodiazepines often have history dating back decades, often discovered in medicinal chemistry efforts and then abandoned for therapeutic purposes [81]. The use of similarity methods could help predict whether existing opiates and benzodiazepine assays are likely to cross-react with the new designer drugs, thus helping to prioritize experimental cross-reactivity testing.

7.6 Relevance to Antibody-Ligand Interactions

The results of our studies raise interesting questions about the structural interactions between diagnostic antibodies and their target ligands. To our knowledge, an X-ray crystallographic or nuclear magnetic resonance (NMR) structure of a diagnostic antibody used for DOA/Tox screening or steroid hormone analysis has not been reported. However, there are crystallographic structures of antibodies evaluated for the treatment of addiction or overdoses of amphetamine [84], cocaine [85, 86], morphine [87], and PCP [88]. There have also been a number of studies looking at the 3D structure of antibodies that bind steroid hormones. These studies illustrate the complexity and heterogeneity inherent in interactions between antibodies and ligands. For example, a crystallographic study of two different estradiol antibodies revealed that antibodies that bound estradiol with equivalent specificity could have substantially different amino acid sequence, ligand orientations, and ligand-binding pockets [89]. Several studies of anti-testosterone antibodies revealed how directed mutagenesis could be used to obtain higher antibody specificity for ligands [90–92]. Crystallographic studies showed that potentially therapeutic anti-cocaine and anti-PCP antibodies interact with all portions of the ligand [85, 86, 88], whereas the study of an anti-morphine antibody demonstrated interaction more with the hydrophobic portion of the ligand, while the hydrophilic half of morphine was mostly exposed to the solvent [87]. The crystallographic structure of digoxin (a relatively large drug molecule used to treat congestive heart failure and arrhythmias) with a Fab antibody fragment showed the carbohydrate portions of the drug exposed to the solvent and unbound by the antibody [93]. This heterogeneity of antibody-ligand molecular interactions undoubtedly underlie some of the observed variation in cross-reactivity of marketed immunoassays for DOA/Tox and steroid hormones.

7.7 Conclusions and Future Directions

Cross-reactivity between structurally related compounds remains a challenge in the clinical use of immunoassays. Similarity analysis represents a computational tool to identify compounds likely to cross-react with immunoassays. Our studies have shown that 2D similarity using MDL public keys as molecular descriptors are especially useful for this purpose. FCFP and pharmacophore fingerprints have more limited use and are best suited for studying very close structural analogs (e.g., methamphetamine and related designer amphetamine-like drugs). Other molecular fingerprints could be evaluated in the future.

A limitation of similarity approaches is that these do not account for the complex 3D molecular interactions between ligand and antibody or cases where the antibody only interacts with portions of the ligand, as seen with crystallographic structures of antibodies against morphine and digoxin [87, 93]. For target compounds such as these, similarity searching using substructures may be useful. An additional limitation of similarity methods is that these do not account for the complicated and often nonlinear concentration dependence of cross-reactivity. Weak cross-reactivity for many compounds with a given immunoassay may be clinically insignificant if suitably high concentrations are not commonly present in specimens being analyzed. However, as seen with abuse of dextromethorphan resulting in false positive PCP screens, some compounds with weak cross-reactivity for an immunoassay may be clinically significant if very high concentrations can be found in body fluids in some circumstances. Furthermore, there are no currently good computational models for the interaction of multiple cross-reacting substances present at once in the same specimen. This type of situation can result from drugs such as benzodiazepines which may have multiple metabolites with varying degrees of cross-reactivity. Predicting the ability of mixtures to cross-react is certainly likely to require training sets with which to build and test models.

Overall, similarity analysis represents a computational tool for understanding and predicting cross-reactivity of immunoassays. Use of the Tanimoto coefficient provides an easily understood quantitative measure for structural similarity between compounds. Similarity analysis can guide investigations of potentially cross-reacting compounds. The computational ease of similarity calculations allows for screening of even large databases of therapeutic drugs, herbal compounds, endogenous molecules, and toxins. These results can help prioritize the much more costly and time-consuming process of experimental cross-reactivity testing. Similarity analysis also provides a logical framework for regulatory agencies and immunoassay manufacturers to focus cross-reactivity testing schemes and guidelines on those compounds most likely to interfere.

In future, as the datasets increase in size, we could expand beyond the molecular similarity approach to using machine learning methods. For example Bayesian, support vector machine, trees, and deep learning, which have all found use in ADME/Tox and other areas of drug discovery, could be readily applied to predicting cross-reactivity [94].

Acknowledgment

We are very grateful to our many collaborators for assisting us on the studies described.

References

1. 1 Holder, G. (2006) Measurement of glucocorticoids in biological fluids. Methods Mol. Biol., 324, 141–157.
2. 2 Kricka, L.J. (2006) Principles of immunochemical techniques, in Tietz Textbook of Clinical Chemistry and Molecular Diagnostics, 4th edn (eds C.A. Burtis, E.R. Ashwood, and D.E. Bruns), Elsevier Saunders, St. Louis, MO, pp. 219–243.
3. 3 Melanson, S.E. (2012) The utility of immunoassays for urine drug testing. Clin. Lab. Med., 32, 429–447.
4. 4 Wheeler, M.J. (2006) Measurement of androgens. Methods Mol. Biol., 324, 197–211.
5. 5 Ayotte, C. (2010) Detecting the administration of endogenous anabolic androgenic steroids. Handb. Exp. Pharmacol., 77–98.
6. 6 Lippi, G., Franchini, M., and Banfi, G. (2011) Biochemistry and physiology of anabolic androgenic steroids doping. Mini. Rev. Med. Chem., 11, 362–373.
7. 7 Moyer, T.P., Burritt, M.F., and Butz, J. (2006) Therapeutic drugs and their management, in Tietz Textbook of Clinical Chemistry and Molecular Diagnostics, 4th edn (eds C.A. Burtis, E.R. Ashwood, and D.E. Bruns), Elsevier Saunders, St. Louis, MO, pp. 1237–1285.
8. 8 Porter, W.H. (2006) Clinical toxicology, in Tietz Textbook of Clinical Chemistry and Molecular Diagnostics (eds C.A. Burtis, E.R. Ashwood, and D.E. Bruns), Elsevier Saunders, St. Louis, MO, pp. 1287–1369.
9. 9 Moeller, K.E., Lee, K.C., and Kissack, J.C. (2008) Urine drug screening: practical guide for clinicians. Mayo Clin. Proc., 83, 66–76.
10. 10 Marquet, P. (2012) LC–MS vs. GC–MS, online extraction systems, advantages of technology for drug screening assays. Methods Mol. Biol., 902, 15–27.
11. 11 Wu, A.H. and French, D. (2013) Implementation of liquid chromatography/mass spectrometry into the clinical laboratory. Clin. Chim. Acta, 420, 4–10.
12. 12 Kricka, L.J. (2000) Interferences in immunoassays – still a threat. Clin. Chem., 46, 1037–1038.
13. 13 Krasowski MD, Pizon AF, Siam MG, Giannoutsos S, Iyer M, Ekins S. Using molecular similarity to highlight the challenges of routine immunoassay-based drug of abuse/toxicology screening in emergency medicine. BMC Emerg. Med. 2009;9(5); (online).
14. 14 Krasowski, M.D., Siam, M.G., Iyer, M. et al. (2009) Chemoinformatic methods for predicting interference in drug of abuse/toxicology immunoassays. Clin. Chem., 55, 1203–1213.
15. 15 Rauh, M. (2009) Steroid measurement with LC–MS/MS in pediatric endocrinology. Mol. Cell Endocrinol., 301, 272–281.
16. 16 Shackleton, C. (2010) Clinical steroid mass spectrometry: a 45-year history culminating in HPLC-MS/MS becoming an essential tool for patient diagnosis. J. Steroid Biochem. Mol. Biol., 121, 481–490.
17. 17 Christophersen, A.S. (2000) Amphetamine designer drugs – an overview and epidemiology. Toxicol. Lett., 112–113, 127–131.
18. 18 Cottencin, O., Rolland, B., and Karila, L. (2014) New designer drugs (synthetic cannabinoids and synthetic cathinones): review of literature. Curr. Pharm. Des., 20, 4106–4111.
19. 19 Weiss, R.L. (2012) The long and winding regulatory road for laboratory-developed tests. Am. J. Clin. Pathol., 138, 20–26.
20. 20 Krasowski, M.D., Drees, D., Morris, C.S. et al. (2014) Cross-reactivity of steroid hormone immunoassays: clinical significance and two-dimensional molecular similarity prediction. BMC Clin. Pathol., 14, 33.
21. 21 Krasowski, M.D., Siam, M.G., and Iyer, M. (2009) Ekins S. Molecular similarity methods for predicting cross-reactivity with therapeutic drug monitoring immunoassays therapeutic drug monitoring., 31, 337–344.
22. 22 Hammett-Stabler, C.A., Pesce, A.J., and Cannon, D.J. (2002) Urine drug screening in the medical setting. Clin. Chim. Acta, 315, 125–135.
23. 23 Caplan, Y.H. and Goldberger, B.A. (2001) Alternative specimens for workplace drug testing. J. Anal. Toxicol., 25, 396–399.
24. 24 Cotten, S.W. (2012) Drug testing in the neonate. Clin. Lab. Med., 32, 449–466.
25. 25 Baden, L.R., Horowitz, G., Jacoby, H., and Eliopoulos, G.M. (2001) Quinolones and false-positive urine screening for opiates by immunoassay technology. JAMA, 286, 3115–3119.
26. 26 Zacher, J.L. and Givone, D.M. (2004) False-positive urine opiate screening associated with fluoroquinolone use. Ann. Pharmacother., 38, 1525–1528.
27. 27 Gagajewski, A., Davis, G.K., Kloss, J. et al. (2002) False-positive lysergic acid diethylamine immunoassay screen associated with fentanyl medication. Clin. Chem., 48, 205–206.
28. 28 Fitzgerald, R.L. and Herold, D.A. (1997) Improved CEDIA benzodiazepine assay eliminates sertraline cross-reactivity. J. Anal. Toxicol., 21, 32–35.
29. 29 Bender, A. and Glen, R.C. (2004) Molecular similarity: a key technique in molecular informatics. Org. Biomol. Chem., 2, 3204–3218.
30. 30 Ekins, S., Mestres, J., and Testa, B. (2007) In silico pharmacology for drug discovery: applications to targets and beyond. Br. J. Pharmacol., 152, 21–37.
31. 31 Gillet, V.J., Willett, P., and Bradshaw, J. (1998) Identification of biological activity using substructural analysis and genetic algorithms. J. Chem. Inf. Comput. Sci., 38, 165–179.
32. 32 Hert, J., Willett, P., Wilton, D.J. et al. (2004) Comparison of fingerprint-based methods for virtual screening using multiple bioactive reference structures. J. Chem. Inf. Comput. Sci., 44, 1177–1185.
33. 33 Hert, J., Willett, P., Wilton, D.J. et al. (2004) Comparison of topological descriptors for similarity-based virtual screening using multiple bioactive reference structures. Org. Biomol. Chem., 2, 3256–3266.
34. 34 Willett, P. (2003) Similarity-based approaches to virtual screening. Biochem. Soc. Trans., 31, 603–606.
35. 35 Zhang, Q. and Muegge, I. (2006) Scaffold hopping through virtual screening using 2D and 3D similarity descriptors: ranking, voting, and consensus scoring. J. Med. Chem., 49, 1536–1548.
36. 36 Rengarajan, A. and Mullins, M.E. (2013) How often do false-positive phencyclidine urine screens occur with use of common medications? Clin. Toxicol. (Phila), 51, 493–496.
37. 37 O'Brien, C.P. (2005) Benzodiazepine use, abuse, and dependence. J. Clin. Psychiatry, 66 (Suppl. 2), 28–33.
38. 38 Perry, P.J., Zeilmann, C., and Arndt, S. (1994) Tricyclic antidepressant concentrations in plasma: an estimate of their sensitivity and specificity as a predictor of response. J. Clin. Psychopharmacol., 14, 230–240.
39. 39 Preskorn, S.H., Dorey, R.C., and Jerkovich, G.S. (1988) Therapeutic drug monitoring of tricyclic antidepressants. Clin. Chem., 34, 822–828.
40. 40 Ling, W., Mooney, L., and Hillhouse, M. (2011) Prescription opioid abuse, pain and addiction: clinical issues and implications. Drug Alcohol Rev., 30, 300–305.
41. 41 Waldhoer, M., Bartlett, S.E., and Whistler, J.L. (2004) Opioid receptors. Annu. Rev. Biochem., 73, 953–990.
42. 42 Martins, S.S., Sampson, L., Cerda, M., and Galea, S. (2015) Worldwide prevalence and trends in unintentional drug overdose: a systematic review of the literature. Am. J. Public Health, 105, 2373.
43. 43 Ordonez Gallego, A., Gonzalez Baron, M., and Espinosa, A.E. (2007) Oxycodone: a pharmacological and clinical review. Clin. Transl. Oncol., 9, 298–307.
44. 44 Morris, B.J., Cochran, S.M., and Pratt, J.A. (2005) PCP: from pharmacology to modelling schizophrenia. Curr. Opin. Pharmacol., 5, 101–106.
45. 45 Bryner, J.K., Wang, U.K., Hui, J.W. et al. (2006) Dextromethorphan abuse in adolescence: an increasing trend: 1999–2004. Arch. Pediatr. Adolesc. Med., 160, 1217–1222.
46. 46 Romanelli, F. and Smith, K.M. (2009) Dextromethorphan abuse: clinical effects and management. J. Am. Pharm. Assoc., 49, e20–e25; quiz e6–7.
47. 47 Schwartz, R.H. (2005) Adolescent abuse of dextromethorphan. Clin. Pediatr. (Phila), 44, 565–568.
48. 48 Soldin, S.J., Steele, B.W., Witte, D.L. et al. (2003) Lack of specificity of cyclosporine immunoassays. Results of a college of American pathologists study. Arch. Pathol. Lab. Med., 127, 19–22.
49. 49 Murthy, J.N., Davis, D.L., Yatscoff, R.W., and Soldin, S.J. (1998) Tacrolimus metabolite cross-reactivity in different tacrolimus assays. Clin. Biochem., 31, 613–617.
50. 50 Juenke, J.M., Miller, K.A., Ford, M.A. et al. (2011) A comparison of two FDA approved lamotrigine immunoassays with ultra-high performance liquid chromatography tandem mass spectrometry. Clin. Chim. Acta, 412, 1879–1882.
51. 51 Savory, D.J. (1995) Interference by caffeine in theophylline measurement. Ann. Clin. Biochem., 32 (Pt 3), 338.
52. 52 Tibi, L. and Burnett, D. (1995) Interference by caffeine in polarization fluoroimmunoassays for theophylline. Ann. Clin. Biochem., 32 (Pt 3), 339.
53. 53 Frank, E.L., Schwarz, E.L., Juenke, J. et al. (2002) Performance characteristics of four immunoassays for antiepileptic drugs on the IMMULITE 2000 automated analyzer. Am. J. Clin. Pathol., 118, 124–131.
54. 54 Klee, G.G. (2004) Interferences in hormone immunoassays. Clin. Lab. Med., 24, 1–18.
55. 55 Roberts, R.F. and Roberts, W.L. (2004) Performance characteristics of five automated serum cortisol immunoassays. Clin. Biochem., 37, 489–493.
56. 56 Hungerford, N.L., Sortais, B., Smart, C.G. et al. (2005) Analysis of anabolic steroids in the horse: development of a generic ELISA for the screening of 17alpha-alkyl anabolic steroid metabolites. J. Steroid Biochem. Mol. Biol., 96, 317–334.
57. 57 Soma, L.R., Uboh, C.E., Guan, F. et al. (2007) Pharmacokinetics of boldenone and stanozolol and the results of quantification of anabolic and androgenic steroids in race horses and nonrace horses. J. Vet. Pharmacol. Ther., 30, 101–108.
58. 58 Castaneto, M.S., Gorelick, D.A., Desrosiers, N.A. et al. (2014) Synthetic cannabinoids: epidemiology, pharmacodynamics, and clinical implications. Drug Alcohol Depend., 144, 12–41.
59. 59 Coppola, M. and Mondola, R. (2012) Synthetic cathinones: chemistry, pharmacology and toxicology of a new class of designer drugs of abuse marketed as “bath salts” or “plant food”. Toxicol. Lett., 211, 144–149.
60. 60 Gunderson, E.W. (2013) Synthetic cannabinoids: a new frontier of designer drugs. Ann. Intern. Med., 159, 563–564.
61. 61 Hill, S.L. and Thomas, S.H. (2011) Clinical toxicology of newer recreational drugs. Clin. Toxicol. (Phila), 49, 705–719.
62. 62 Kapka-Skrzypczak, L., Kulpa, P., Sawicki, K. et al. (2011) Legal highs – legal aspects and legislative solutions. Ann. Agric. Environ. Med., 18, 304–309.
63. 63 Musselman, M.E. and Hampton, J.P. (2014) “Not for human consumption”: a review of emerging designer drugs. Pharmacotherapy, 34, 745–757.
64. 64 Nelson, M.E., Bryant, S.M., and Aks, S.E. (2014) Emerging drugs of abuse. Emerg. Med. Clin. North Am., 32, 1–28.
65. 65 Thornton, M.D. and Baum, C.R. (2014) Bath salts and other emerging toxins. Pediatr. Emerg. Care, 30, 47–52; quiz 3–5.
66. 66 Zawilska, J.B. and Wojcieszak, J. (2013) Designer cathinones – an emerging class of novel recreational drugs. Forensic Sci. Int., 231, 42–53.
67. 67 Zawilska, J.B. and Wojcieszak, J. (2014) Spice/K2 drugs – more than innocent substitutes for marijuana. Int. J. Neuropsychopharmacol., 17, 509–525.
68. 68 Auwarter, V., Dresen, S., Weinmann, W. et al. (2009) ‘Spice’ and other herbal blends: harmless incense or cannabinoid designer drugs? J. Mass Spectrom., 44, 832–837.
69. 69 Shulgin, A. and Shulgin, A. (1991) PIHKAL: A Chemical Love Story, Transform Press, Berkeley, CA.
70. 70 Shulgin, A. and Shulgin, A. (1997) TIHKAL: The Continuation, Transform Press, Berkeley, CA.
71. 71 Drug Enforcement Administration, Department of Justice (2013) Establishment of drug codes for 26 substances. Final rule. Fed Regist., 78, 664–666.
72. 72 Huffman, J.W. and Dai, D. (1994) Design, synthesis and pharmacology of cannabimimetic indoles. Bioorg. Med. Chem., 4, 563–566.
73. 73 Peters, F.T. and Martinez-Ramirez, J.A. (2010) Analytical toxicology of emerging drugs of abuse. Ther. Drug Monit., 32, 532–539.
74. 74 Krasowski, M.D. and Ekins, S. (2014) Using cheminformatics to predict cross reactivity of “designer drugs” to their currently available immunoassays. J. Cheminform., 6, 22.
75. 75 Petrie, M., Lynch, K.L., Ekins, S. et al. (2013) Cross-reactivity studies and predictive modeling of “Bath Salts” and other amphetamine-type stimulants with amphetamine screening immunoassays. Clin. Toxicol. (Phila), 51, 83–91.
76. 76 Apollonio, L.G., Whittall, I.R., Pianca, D.J. et al. (2007) Matrix effect and cross-reactivity of select amphetamine-type substances, designer analogues, and putrefactive amines using the Bio-Quant direct ELISA presumptive assays for amphetamine and methamphetamine. J. Anal. Toxicol., 31, 208–213.
77. 77 Arntson, A., Ofsa, B., Lancaster, D. et al. (2013) Validation of a novel immunoassay for the detection of synthetic cannabinoids and metabolites in urine specimens. J. Anal. Toxicol., 37, 284–290.
78. 78 Kerrigan, S., Mellon, M.B., Banuelos, S., and Arndt, C. (2011) Evaluation of commercial enzyme-linked immunosorbent assays to identify psychedelic phenethylamines. J. Anal. Toxicol., 35, 444–451.
79. 79 Nakanishi, K., Miki, A., Zaitsu, K. et al. (2012) Cross-reactivities of various phenethylamine-type designer drugs to immunoassays for amphetamines, with special attention to the evaluation of the one-step urine drug test Instant-View, and the Emit(R) assays for use in drug enforcement. Forensic Sci. Int., 217, 174–181.
80. 80 Swortwood, M.J., Hearn, W.L., and Decaprio, A.P. (2014) Cross-reactivity of designer drugs, including cathinone derivatives, in commercial enzyme-linked immunosorbent assays. Drug Test Anal., 6, 716–727.
81. 81 Carroll, F.I., Lewin, A.H., Mascarella, S.W. et al. (2012) Designer drugs: a medicinal chemistry perspective. Ann. NY Acad. Sci., 1248, 18–38.
82. 82 Rech, M.A., Donahey, E., Cappiello Dziedzic, J.M. et al. (2015) New drugs of abuse. Pharmacotherapy, 35, 189–197.
83. 83 Pourmand, A., Armstrong, P., Mazer-Amirshahi, M., and Shokoohi, H. (2014) The evolving high: new designer drugs of abuse. Hum. Exp. Toxicol., 33, 993–999.
84. 84 Celikel, R., Peterson, E.C., Owens, S.M., and Varughese, K.I. (2009) Crystal structures of a therapeutic single chain antibody in complex with two drugs of abuse-Methamphetamine and 3,4-methylenedioxymethamphetamine. Protein Sci., 18, 2336–2345.
85. 85 Larsen, N.A., Zhou, B., Heine, A. et al. (2001) Crystal structure of a cocaine-binding antibody. J. Mol. Biol., 311, 9–15.
86. 86 Pozharski, E., Moulin, A., Hewagama, A. et al. (2005) Diversity in hapten recognition: structural study of an anti-cocaine M82G2. J. Mol. Biol., 349, 570–582.
87. 87 Pozharski, E., Wilson, M.A., Hewagama, A. et al. (2004) Anchoring a cationic ligand: the structure of the Fab fragment of the anti-morphine antibody 9B1 and its complex with morphine. J. Mol. Biol., 337, 691–697.
88. 88 Lim, K., Owens, S.M., Arnold, L. et al. (1998) Crystal structure of monoclonal 6B5 Fab complexed with phencyclidine. J. Biol. Chem., 273, 28576–28582.
89. 89 Monnet, C., Bettsworth, F., Stura, E.A. et al. (2002) Highly specific anti-estradiol antibodies: structural characterisation and binding diversity. J. Mol. Biol., 315, 699–712.
90. 90 Hemminki, A., Niemi, S., Hoffren, A.M. et al. (1998) Specificity improvement of a recombinant anti-testosterone Fab fragment by CDRIII mutagenesis and phage display selection. Protein Eng., 11, 311–319.
91. 91 Valjakka, J., Hemminki, A., Niemi, S. et al. (2002) Crystal structure of an in vitro affinity- and specificity-matured anti-testosterone Fab in complex with testosterone. Improved affinity results from small structural changes within the variable domains. J. Biol. Chem, 277, 44021–44027.
92. 92 Valjakka, J., Takkinenz, K., Teerinen, T. et al. (2002) Structural insights into steroid hormone binding: the crystal structure of a recombinant anti-testosterone Fab fragment in free and testosterone-bound forms. J. Biol. Chem., 277, 4183–4190.
93. 93 Jeffrey, P.D., Schildbach, J.F., Chang, C.Y. et al. (1995) Structure and specificity of the anti-digoxin antibody 40–50. J. Mol. Biol., 248, 344–360.
94. 94 Ekins, S. (2016) The next era: deep learning in pharmaceutical research. Pharm. Res., 33, 2594–2603.