Interpretation of in Vitro Metabolic Stability Studies for Racemic Mixtures

Abstract

In early drug discovery, where chiral syntheses may not yet have been elucidated or enantiomeric separation is not feasible, screening of racemates in metabolic stability assays may offer a pragmatic approach. To assess the risk of incorrectly deprioritizing enantiomers due to misclassification of apparent in vitro intrinsic clearance (CLint_app), we evaluated (1) theoretical simulations; (2) literature on enantiomeric CLint_app differences; (3) historic MSD data; and (4) new data on enantiomers with high eudysmic ratios and low predicted three-dimensional similarity. Overall, the results suggested minimal risk of not progressing an enantiomer due to an appreciably different (>3-fold) racemate CLint_app.

During drug discovery, a key early measured parameter is often metabolic stability, typically using hepatocytes (heps) or hepatic microsomes (mics) supplemented with NADPH. Metabolic stability (or apparent intrinsic clearance, CLint_app) is used, along with binding estimates, in the prediction of in vivo metabolic intrinsic clearance.¹ This, in turn, when contextualized with other key parameters, is used in predictions of oral bioavailability, exposure, half-life, and hence efficacious dose. In this Letter, CLint_app refers to the apparent (measured) CLint without consideration of incubation binding.

Compound specifications for initial screening assays are often based on both chemical and enantiomeric purity (typically >90%), particularly for in vitro and in vivo drug metabolism and pharmacokinetic (DMPK) studies. This is based on concerns over the nature of contaminants (e.g., possible CYP inhibitors) as well as the known potential for enantiomer-specific metabolism.²

Occasionally, due to synthetic limitations for chiral chemistry and/or analytical challenges with chiral separations, drug discovery project teams may look to reduce the cycle time for in vitro potency, selectivity, and DMPK assays by generating early data on racemic mixtures (composed of enantiomers E1 and E2). Such data may be used to inform on the value of progressing any given compound, therefore warranting subsequent chiral separation or chiral resynthesis. In addition, racemate CLint_app could help assess the potential for metabolic vulnerability of the required pharmacophore. Assuming a 50:50 enantiomeric mix (rather than enantiomeric excess >0 or contamination by intermediates or byproducts), the risk of missing active enantiomers by testing racemates in potency assays is low; if only a single enantiomer is active, that activity will be diluted only 2-fold by the inactive enantiomer. In this investigation, we attempted to understand whether similar assumptions can be made when measuring CLint_app. Throughout the simulations, we have assumed a 50:50 mix of enantiomers in the racemic mixtures, as would be formed by chemical syntheses without the use of chiral solvents or chiral catalysts.

Assays to measure CLint_app almost invariably have LC–MS/MS end points that do not distinguish enantiomers. Diastereomers were not considered in this analysis because they are generally separable under standard purification conditions and submitted to in vitro metabolic stability assays as distinct entities. However, when assays are performed with racemic mixtures, data will reflect a combination of the two different enantiomers (Figure 1). When the enantiomer CLint_app values are different, the resulting racemate plot will be biphasic. The initial slope (k_e) is used to determine the CLint_app (where CLint_app = −k_e/protein or cell concentration).

Figure 1.

Simulated ln(% remaining) vs time plots for enantiomer 1 (E1, squares), enantiomer 2 (E2, triangles) (CLint_app of 100 and 1000 μL/min/mg microsomal protein, respectively), and corresponding racemate (diamonds) in a 0.25 mg/mL microsomal metabolic stability assay.

The MSD metabolic stability database was interrogated to find compounds (irrespective of chirality) that had previously been tested more than once. This data (Table 1) was used as a baseline to define assay variability. Over 400 compounds had been repeat tested, with 88–95% showing CLint_app values within 3-fold. This level of reproducibility (2–3-fold) is consistent with both in-house (data not shown) and literature³ values for positive control compounds that are run in each assay.

Table 1. Replicate CLint_app Data for Identical Compounds Tested at Least Twice in MSD Metabolic Stability Assays.

	WH rat				human
assay	n	>3-fold (%)	>10-fold (%)	fold error	n	>3-fold (%)	>10-fold (%)	fold error
Heps	471	7	1.5	2.23	594	5	0.2	1.65
Mics	427	12	0.9	1.96	449	10	0.7	1.93

To understand the risk of misclassification of CLint_app, [i.e., racemate CLint_app (R) > 3-fold higher than the lower CLint_app enantiomer (LCE)], we performed simulations of the ratio of the racemate CLint_app (R) to that of each enantiomer across a wide range of E1/E2 CLint_app values and ratios These simulations (not shown) revealed that the CLint_app value for the more stable enantiomer tended to have greater differences from the racemate values, compared to the enantiomer with higher CLint_app (i.e., R/LCE [lower CLint_app enantiomer] > R/HCE [higher CLint_app enantiomer]). Subsequent simulations focused on the R/LCE CLint_app ratio for different E1/E2 CLint_app scenarios.

Figure 2 shows that when the E1/E2 CLint_app ratio is high (e.g., 30- or 100-fold), the measured racemate CLint_app would overestimate the lower enantiomer CLint_app significantly (R/LCE CLint_app ratio = 12–15 and 42–45 respectively). The plot also shows that when E1/E2 CLint = 10, the R/LCE ratio can vary widely. To understand this further, we performed additional simulations for E1/E2 CLint_app ratios of 3 and 10. For example, for E1/E2 CLint_app ratio = 10, we simulated racemate data for enantiomer pairs having CLint_app values ranging from 1 and 10 to 1000 and 10000 μL/min/mg. The results (Figure 3) show that for E1/E2 CLint_app ratios of 3, the R/LCE would be 1.5–2.5. However, when the E1/E2 ratio was 10, at low racemate CLint_app, the R/LCE CLint_app ratio would be >5, but this decreased to 1 as measured racemate CLint_app increased.

Figure 2.

Simulated plots: ratio of the racemate CLint to that of the more stable enantiomer (R/LCE) versus the ratio of CLint values for the enantiomers (E1/E2).

Figure 3.

Simulated R/LCE plotted against racemate CLint_app for E1/E2 ratios of 3 (squares) and 10 (circles).

These simulations were performed assuming MSD in-house microsomal stability screening conditions with initial time-points of 0 and 5 min. To test the hypothesis that these time-points were impacting the data interpretation, we also simulated 0 and 1 min initial time-points for enantiomers with CLint_app of 300 and 3000 μL/min/mg (E1/E2 CLint_app ratio = 10). The simulated racemate CLint_app was 577 μL/min/mg with 0 and 5 min compared to 1232 μL/min/mg with 0 and 1 min, resulting in R/LCE CLint_app ratios of 2 and 4, respectively. Simulating with earlier time-points allows a more accurate determination of the initial slope resulting in a higher CLint_app. Under standard experimental conditions (e.g., 0 and 5 min time-point), overestimating lower enantiomer CLint_app using racemates may therefore be less likely.

Chiral inversion and racemization are not usually considered during metabolic stability screening since the analytical methods usually employed do not resolve enantiomers. Chiral inversion, the process whereby one enantiomer converts to the other, through either enzymatic (e.g., R-ibuprofen^4,5) or chemical (e.g., clopidogrel⁶) means, provides additional complexity when interpreting racemate data. If one enantiomer inverts, then the other enantiomer will have both formation and clearance rates. Under this scenario, initial slopes in theoretical models may be misleading, as shown in Table 2. In this example, racemate CLint_app provides a good approximation of actual CLint_app for both the enantiomers until there is a 10-fold difference in the sum of the CLint pathways, as also seen when chiral inversion does not occur.

Table 2. Simulation of the Effect of Chiral Inversion on Observed CLint_app^a.

sum of E1 CLintapp	E2 CLintapp (initial slope)	racemate CLintapp	R/LCE (initial slope)	R/LCE (actual; E2 CLintapp = 100)
101	10.9	53.5	4.9	0.5
110	11.8	58	4.9	0.6
200	20.1	100	5.0	1.0
1100	65	313	4.8	3.1

^aCLint_app values were fixed to 100 for both E1 to E2 (chiral inversion) and E2 metabolism. Noninversion metabolism CLint_app for E1 was varied from 1 to 1000.

The simulations demonstrated that if E1/E2 CLint_app ratios are >10, then racemate CLint_app data may significantly overestimate the lower CLint_app. However, for E1/E2 CLint_app ratios of ≤10, using our standard experimental time-points, racemate CLint_app data will be an acceptable surrogate for the individual enantiomers. Exceptions would occur where E1/E2 CLint_app ratio = 10 with a low racemate CLint_app (e.g., < 100 μL/min/mg). However, these compounds are unlikely to be deprioritized from progression to further assays if low CLint_app is the goal. Based on this, we subsequently explored whether E1/E2 CLint_app values >10 are indeed likely.

To explore actual enantiomeric differences in CLint_app, literature data were evaluated for compounds with known stereoselective metabolism. The overall E1/E2 CLint_app ratios, based on either substrate depletion or the sum of all metabolite formation CLint_app values, were found to be <3 for each compound (warfarin⁷ = 1.3, verapamil⁸ < 2, metoprolol⁹ < 1.2, and omeprazole¹⁰ = 2.9), despite some marked differences in the CLint_app for formation of individual metabolites.

A retrospective analysis of MSD screening data was also performed for matched pairs of enantiomers (n = 750; randomly assigned as E1 or E2) with CLint_app values determined in rat and human microsomes and/or hepatocytes. Briefly, experimental conditions were as follows: 0.3 μM substrate was incubated with 0.25 mg/mL pooled human liver microsomes (initiated with the addition of 1 mM NADPH) or 1 million cells/mL cryopreserved pooled human hepatocytes. Loss of parent compound was evaluated by LC–MS/MS after quenching with acetonitrile containing internal standard, protein precipitation, and centrifugation of time-point samples. Each compound had a single replicate incubation for these screening assays. Strict criteria were used to select enantiomer pairs requiring each isomer structure to be fully stereochemically defined and annotated as a pure single isomer by the chemist. Compounds came from many programs and represented broad small molecule chemical space. As shown in Figure 4, the CLint_app values for the enantiomeric pairs generally fell within 3-fold of each other with 10-fold differences being rare (<1% of CLint_app values were >10-fold different). The enantiomeric pair CLint_app data is remarkably similar to the assay reproducibility data for identical compounds summarized in Table 1 (>10-fold CLint_app differences of 0.3–0.9% for enantiomers and 0.2–1.5% for the same compound in replicate assays). This suggests that enantiomeric differences do not generally appear to be larger than would be expected from overall assay variability.

Figure 4.

Retrospective analysis of MSD database comparing CLint_app values for matched pairs of enantiomers in Wistar Han rat and human liver microsomal (a and c, μL/min/mg) and hepatocyte stability assays (b and d, μL/min/10⁶ cells). Lines are unity (solid) and 3-fold (dashed) and 10-fold from unity (dotted).

Interestingly, compounds that had >10-fold enantiomeric CLint_app differences in one assay type did not consistently show this effect across species or matrices. It was also noted that similar overall data were obtained in both NADPH-dependent microsomal and hepatocyte stability assays, suggesting concordance for both cytochrome P450 (CYP)- and non-CYP-mediated metabolism pathways (although this was not specifically evaluated in this study).

Further analysis of the same MSD database showed that, of these 750 enantiomer pairs, approximately 100 had also been tested in metabolic stability assays as racemates. Figure 5 illustrates that, for these compounds, the CLint_app ratio of the racemate to LCE was generally within 2-fold, which was in agreement with the simulation in Figure 3 (i.e., E1/E2 CLint_app ratio ≤ 3 leads to racemate/LCE CLint_app ratio ≤ 2).

Figure 5.

Histogram showing frequency of measured CLint_app ratios for racemic mixtures compared to the lower CLint_app enantiomer (LCE) from MSD database.

Submission of compounds for metabolic stability assays is often dependent on primary potency criteria. Consequently, if both enantiomers of a compound were tested for metabolic stability, it is likely that they were both active and potentially adopt a similar three-dimensional shape in their target as well as metabolic enzymes. To ensure against bias in the retrospective analysis, pairs of enantiomers with largely different potency and predicted three-dimensional shapes were prospectively selected for measurement of CLint_app. Criteria for compound selection were as follows: eudysmic ratio > 100 in any potency assay and low 3D overlay score (<1.5 maximal TanimotoCombo score using ROCS¹¹ after generation of up to 200 conformers of each enantiomer with OMEGA¹²). The resulting selection was further refined based on a range of predicted CLint_app (in-house QSAR model) and structural diversity. CLint_app data was generated in rat and human liver microsomal stability assays for the 10 resulting enantiomer pairs. For this “extreme” compound set, enantiomeric differences in CLint_app were >3-fold for only one pair of enantiomers in human liver microsomes and three pairs in rat liver microsomes. In both rat and human liver microsomes, a single pair showed enantiomeric CLint_app differences >10-fold (data not shown).

In conclusion, using a variety of approaches, the risk of misinforming project teams through generation of metabolic stability data on racemic mixtures is low. Through theoretical considerations, enantiomeric CLint_app differences of >10-fold may cause concern. However, from literature evaluation, retrospective analysis and prospective studies with isomers of high eudysmic ratio and dissimilar shape, such large differences in CLint_app for enantiomers, seems to be a relatively rare event. With the improving throughput of chiral separation technologies¹³ as well as advances in chiral chemistry, program teams may only need to consider testing racemic mixtures for nonroutine workflows. Additionally, teams would need to assess the overall impact of timelines by adding an additional step for measuring racemate CLint_app as well as the subsequent testing of the separate enantiomers. However, understanding the risks should enable pragmatic screening study design decisions when warranted. These results also highlight that physicochemical properties are an important contributor to in vitro metabolic stability,¹⁴ as demonstrated by similar stabilities for enantiomers prospectively selected to be as biologically differentiated as possible.

Glossary

ABBREVIATIONS

CLint_app

apparent intrinsic clearance

CYP

cytochrome P450

DMPK

drug metabolism and pharmacokinetics

E1, E2

enantiomers 1 and 2

Heps

hepatocytes

LC–MS/MS

liquid chromatography with triple quadrupole mass spectrometry

Mics

microsomes

MSD

Merck Sharp & Dohme, a subsidiary of Merck & Co., Inc.

NADPH

nicotinamide adenine dinucleotide phosphate

R/LCE

racemate CLint_app divided by CLint_app of the most stable enantiomer (“lower CLint_app enantiomer”)

R/HCE

racemate CLint_app divided by CLint_app of the least stable enantiomer (“higher CLint_app enantiomer”)

QSAR

quantitative structure–activity relationship

WH rat

Wistar Han rat

3D

three-dimensional

Author Contributions

The manuscript was written through contributions of all authors.

The authors declare the following competing financial interest(s): All authors are all employees of Merck Co., Inc.