When Cofactors Aren’t X Factors: Functional Groups That Are Labile in Human Liver Microsomes in the Absence of NADPH

Abstract

The metabolic stability of compounds is often assessed at an early stage in drug discovery programs by profiling with hepatic microsomes. Exclusion of the reduced form of nicotinamide adenine dinucleotide phosphate (NADPH) in these assays provides insight into non-cytochrome P450 (CYP)-mediated metabolism. This report uses a matched molecular pair (MMP) application to assess which chemical substituents are commonly susceptible to non-NADPH-mediated metabolism by microsomes. The analysis found the overall prevalence of metabolism in the absence of NADPH to be low, with esters, amides, aldehydes, and oxetanes being among the most commonly susceptible functional groups. Given that non-CYP enzymes, such as esterases, may be expressed extrahepatically and lead to lower confidence in predicted pharmacokinetic profiles, an awareness of the functional groups that commonly undergo non-NADPH-mediated metabolism—as well as options for their replacement based on experimental MMP data—may help researchers derisk metabolic stability issues at an earlier stage in drug discovery.

The in vitro quantification of the metabolic stability of compounds is integrated into drug discovery programs as a means to prioritize compounds with optimal in vivo clearance at an early stage.¹⁻³ Commonly, such analysis is conducted with human liver microsomes (HLM) in the presence of the reduced form of nicotinamide adenine dinucleotide phosphate (NADPH).⁴ A principal metabolic pathway in these preparations is oxidation mediated by cytochrome P450 enzymes (CYPs), which is an NADPH-dependent process.¹⁻⁴ Exclusion of NADPH in microsomal assays (HLM-N) is an important negative control which provides insight as to whether a compound is being metabolized in an NADPH-independent fashion.^5,6 Hepatic microsomes contain a variety of drug metabolizing enzymes aside from CYPs, including esterases and epoxide hydrolases.¹ In addition to these, microsomal preparations may be contaminated with cytosolic enzymes, such as aldehyde oxidase and carboxyl esterases.⁷ A positive readout in HLM-N assays may suggest that a compound is a substrate for one of these non-NADPH-dependent enzymes, or even that it is chemically unstable.^5,6

While CYPs have evolved to metabolize a variety of different structurally unrelated lipophilic xenobiotics, non-NADPH-dependent enzymes present in microsomes tend to be more specific in their ability to target particular chemical moieties (e.g., esterases).⁸ Consequently, it may be easier to identify structural motifs that are sensitive to HLM-N metabolism through cheminformatic analysis of an HLM-N database. Such a database is available at Genentech, as HLM-N is an assay that is routinely conducted for new compounds in discovery programs.

Although there is a broad appreciation of the enzymes that may be present and active in HLM-N preparations, there are not, to our knowledge, any reported analyses of large HLM-N data sets.⁹⁻¹² In this work, a previously described molecular matched pair (MMP) application is used to analyze Genentech’s HLM-N database.¹³ In order to avoid potentially inflating the signal emerging from a limited set of related compounds, a cutoff of at least 15 examples for each MMP transformation was established, resulting in ∼53,000 transformations. This study endeavored to answer three central questions: (1) what is the prevalence of metabolism by HLM-N, (2) which functional groups are consistently susceptible to HLM-N metabolism, and (3) which isosteres does the data set suggest that mitigate these pathways?

In order to interrogate the prevalence of metabolism by HLM-N, an MMP analysis of Genentech’s HLM-N data set was performed. Average fold change in intrinsic clearance (for convenience, abbreviated as ΔCL_int) of MMP transformations was used in this analysis, with ΔCL_int > 1 representing a destabilizing change in going from MMP_A to MMP_B and ΔCL_int < 1 representing a stabilizing change. HLM-N experiments were conducted in an analogous manner to previously reported HLM methods, albeit with the exclusion of NADPH.¹⁴ In this assay setup, CL_int is derived from a parent disappearance time-course experiment, which makes the quantification of the stability of low turnover compounds more challenging. To aid in data quality and interpretation, a value of CL_int = 4.8 mL/min/kg was adopted as lower limit in this study. The change in molecular size between MMPs was limited to the addition of six heavy atoms to restrict the analysis to the contribution of smaller functional groups. Using this criterion, the largest MMP molecular weight change included in this analysis was ΔMW = 126, corresponding to addition of iodine to its parent molecule (i.e., MMP_A = H, MMP_B = I).

Initial analysis of the HLM-N data set showed 130 MMP transformations with ΔCL_int > 3 (Figure 1A). Cursory inspection of these transformations revealed that, in many cases, the pairs of molecules underlying the overall MMP transformation had undefined stereochemistry. As each isomer of MMP_A forms an MMP with all other isomers of MMP_B, pairs of molecules with undefined stereochemistry gain multiplicative representation in the MMP data set (Figure 1B). While these isomeric MMPs do represent real MMPs, the quadratic growth in transforms can produce an outsized influence on the results, muddying the waters of any resultant analysis. Hence, undefined stereoisomers were removed from the data set.

Figure 1.

Preparation of the initial HLM-N data set through stereoisomer filtering. Each point represents an individual MMP transformation. (A) A graph depicting ΔCL_int vs MMP transformation count for the initial, stereoisomer-containing HLM-N data set. (B) Schematic illustrating the potential for quadratic growth of undefined stereoisomers within an MMP analysis. (C) Graph depicting ΔCL_int vs MMP transformation count for the HLM-N data set after removal of stereoisomers. (D) Comparison of ΔCL_int for the stereoisomer-containing and stereoisomer-removed HLM-N data sets. The blue lines represent a 1.5-fold change in ΔCL_int. Vortex version used to make the plots was Vortex v2020.1.117430-s.

In order to remove the influence of these undefined stereoisomers, the following filtering method was applied to the data set. Briefly, stereoisomer characters were removed from the SMILES strings, and then, the MACCS key fingerprint was taken of each of the SMILES strings and duplicates were removed (for full details, see SI).¹⁵ Removal of stereoisomers reduced the overall number of molecules in the data set from 118,110 to 100,651 (or, by ∼15%), reducing the number of MMP transformations with at least 15 examples to ∼9,700 (or, by ∼82%), and the number of transformations with ΔCL_int > 3 to just 38 (Figure 1C). Generally, removal of stereoisomers did not have a large effect on ΔCL_int, with ΔCL_int changing by less than 1.5-fold for 99.8% of transformations (Figure 1D).

Having removed stereoisomers, analysis of the resulting data set was conducted. A majority of MMP transformations had a small effect on HLM-N clearance, with 83% of transformations having ΔCL_int < 1.2 (Figure 2A). MMP transformations with larger changes in HLM-N clearance were less common: 13% had ΔCL_int = 1.2–1.5, 3% had ΔCL_int = 1.5–3.0, and only 1% had ΔCL_int > 3.0. The insensitivity of MMP transformations to HLM-N is largely due to the stability of most compounds under these conditions, with 68% of the examples in the HLM-N database having a CL_int under the stated cutoff (defined earlier as CL_int < 4.8 mL/min/kg) and 82% of examples in the “stable category” (defined as CL_int < 6.2 mL/min/kg). This stands in contrast to results from the HLM assay with NADPH (HLM+N), in which only 14% of examples are under the limit of detection and 23% of examples are in the “stable category”. These results suggest that the overall prevalence of metabolism by HLM-N is relatively low.

Figure 2.

(A) Distribution of MMP transformations by ΔCL_int categories: <1.2, 1.2–1.5, 1.5–3.0, >3.0. (B) Distribution of MMP transformations with ΔCL_int > 3.0 binned by the functional group contained within MMP_B. Vortex version used to make the plots was Vortex v2020.1.117430-s.

With an understanding of the overall prevalence of HLM-N in hand, an exploration of the more highly metabolized MMP transformations was undertaken. Inspection of highly destabilizing MMP transformations for which ΔCL_int > 3 revealed ester, amide, and aldehyde functionality to be prevalent among this group (Figure 2B). Namely, 66% of MMP transformations in this highly metabolized category resulted in ester functionality, 18% in amides, and 5% in aldehydes (Figure 2B). These results support the notion that esterases and amidases are primary metabolic components of HLM-N preparations. As these enzymes are quite ubiquitously expressed in many body tissues and fluids, even relatively low metabolic turnover in HLM-N experiments may signal a broader liability that could limit systemic exposure.¹⁶⁻¹⁸ The activity of esterases in preclinical species can additionally be quite different than that in humans, complicating pharmacokinetic predictions for molecules which undergo this type of metabolism.¹⁹

Esters were broadly unstable functional groups in the HLM-N assay: every example of an MMP that resulted in ester functionality (i.e., MMP_B = ester) increased HLM-N metabolism. Among esters, even the transform with the smallest change in ΔCL_int, MMP_A = methyl group to MMP_B = ethyl ester, increased HLM-N metabolism by 2.9-fold. Analyzing the full HLM-N data set showed that 96% of MMPs resulting in ester functionality had ΔCL_int > 3, with the remainder of examples falling in the ΔCL_int = 1.5–3.0 category (Figure 3A). A set of four MMPs starting from MMP_A = H provide insight into relative ester stability: methyl esters attached by a methylene unit were least stable (ΔCL_int = 32, count = 15); methyl esters were slightly less stable than ethyl esters (ΔCL_int = 12, count = 99 vs ΔCL_int = 7, count = 46, respectively); and esters were broadly less stable than aldehydes (Figure 3B). A search of the entire Genentech HLM-N database for ester-containing compounds revealed a modest correlation between sterics and HLM-N stability: of all tert-butyl-ester-containing molecules, 67% exhibited some turnover (i.e., were above the lower limit of detection) in the HLM-N assay, while 95, 76, and 89% of isopropyl-, ethyl-, and methyl-esters, respectively, showed turnover by HLM-N. These data demonstrate that esters are metabolically capricious functional groups within the context of HLM-N assays.

Figure 3.

(A) Distribution of MMP transformations that terminate in ester functionality (i.e., MMP_B = ester) by ΔCL_int categories: 1.5–3.0 and >3.0. (B) Ester- and aldehyde-containing MMPs. The first number below each transformation denotes ΔCL_int for that transformation, and the number in parentheses denotes the MMP count. Vortex version used to make the plots was Vortex v2020.1.117430-s.

Despite being among the most highly metabolized MMPs, amides were generally stable to HLM-N. Specifically, 82% of MMPs resulting in amide functionality had ΔCL_int < 1.2 (Figure 4). Of the amide-containing MMPs with ΔCL_int > 3, the majority contained acetanilide-derived structures, perhaps suggesting that these more-labile amides may be prone to hydrolysis under HLM-N conditions (Figure 5A). Within the ΔCL_int = 1.5–3 category, acetanilides—as well as secondary acetamides more generally—comprised a large portion of MMPs. This again suggests that these chemotypes may be more hydrolytically unstable than other amides in this context (Figure 5A).^20,21 Although the MMP data set under evaluation does not allow general hydrolytic instability to be distinguished from enzymatic hydrolysis, prior reports comparing catalyzed and noncatalyzed stability of esters and amides suggest that enzymatic hydrolysis may be a more competent pathway in many cases.²²⁻²⁴

Figure 4.

Distribution of MMP transformations that terminate in amide functionality (i.e., MMP_B = amide) by ΔCL_int categories: <1.2, 1.2–1.5, 1.5–3.0, and >3.0. Vortex version used to make the plots was Vortex v2020.1.117430-s.

Figure 5.

(A) Select MMP transformations that terminate with amides (i.e., MMP_B = amide) within the ΔCL_int = 1.5–3.0 and >3.0 categories. Acetanilide and acetamide motifs are common within these categories. (B) MMP transformations which terminate with cyclopropyl amides (i.e., MMP_B = cyclopropyl amide) within the ΔCL_int = 1.5–3.0 and >3.0 categories. The first number below each transformation denotes ΔCL_int for that transformation, and the number in parentheses denotes the MMP count.

Secondary cyclopropyl amides were a common substructure found within the category of amides with ΔCL_int = 1.5–3.0. Compared against both simple functionality (e.g., −H, −NH₂, and −OH) and amide isosteres (e.g., carbamate, urea, sulfamide, and −NH(heteroaryl)), introduction of a cyclopropyl amide increased turnover by HLM-N (Figure 5B). As cyclopropyl amides are commonly found within drug discovery programs, assessment of non-NADPH-dependent metabolism of these motifs at an early stage may help inform target progression, prediction of in vivo clearance, and potential mitigation strategies.^25,26 Although the specific metabolic fate of these cyclopropyl amides is outside the scope of this work, it should be noted that hydrolytic cleavage of these amides would produce cyclopropanecarboxylic acid, which can lead to carnitine depletion risks.²⁷

Within a drug discovery program, structurally isosteric amides may be used to satisfy similar binding interactions within a protein pocket and tune physicochemical properties while mitigating metabolic risks.²⁸⁻³⁰ Given the observed range of stabilities that structurally different amides can have under HLM-N conditions, efforts were made to identify more metabolically stable isosteric replacements for these motifs. Additionally, as esters and aldehydes–which may themselves be considered amide isosteres–were among the most metabolically labile groups observed in this analysis, they were included in this isosteric analysis for context.

In comparing isosteres, HLM+N data (i.e., HLM experiments conducted in the presence of NADPH) was included in addition to HLM-N to provide insight into the overall metabolic consequences of functional group interconversion. The HLM+N data set was subjected to the previously applied stereochemistry filter for consistency. As observed when using this as a processing step for the HLM-N data set, there was minimal change in HLM+N values for MMPs between the original and stereoisomer-removed data sets (Figure 6).

Figure 6.

Comparison of ΔCL_int for the stereoisomer-containing and stereoisomer-removed HLM+N data sets. The blue lines represent a 1.5-fold change in ΔCL_int. Vortex version used to make the plots was Vortex v2020.1.117430-s.

Comparison of each ester/amide isostere to its parent compound (i.e., MMP_A = H) was first conducted to establish the baseline contribution of each isostere to microsomal stability. Introduction of methyl ester functionality increased HLM-N turnover by 12-fold and HLM+N turnover by 3-fold (Figure 7, top line). This disparity between HLM-N and HLM+N results may be ascribed to a variety of causes, including the interplay between oxidative and hydrolytic metabolism: esters introduce a hydrolytic liability, while their polar, electron-withdrawing nature may attenuate microsomal oxidation of the parent structure. If the gains in oxidative stability outweigh the losses incurred by hydrolysis, ester MMPs could appear more stable in HLM+N than HLM-N. MMPs between parent compounds (i.e., MMP_A = H) and ester isosteres broadly extinguished turnover by HLM-N, with the exception of acetamides, methylsulfamides, and dimethylsulfamides which all slightly increased turnover by HLM-N. However, as described above, the magnitude of HLM-N ΔCL_int for these isosteres is much less than that observed for esters (i.e., HLM-N ΔCL_int = 1.05–1.33 vs 12.3, respectively) With the exception of dimethylsulfamides, all ester isosteres examined reduced observed turnover by HLM+N. These results provide a roadmap for mitigating metabolism of esters by both HLM-N and HLM+N through isosteric replacement.

Figure 7.

Ester/amide isostere table with HLM-N ΔCL_int, HLM+N ΔCL_int, and count data for MMPs. The top line in each box is the HLM-N ΔCL_int value for that MMP followed by HLM-N MMP count in parentheses. The bottom line in each box is the HLM+N ΔCL_int value for that MMP followed by HLM+N MMP count in parentheses. MMP pairs are read starting from MMP_A in blue, reading across rightward, and then upward to MMP_B in green.

Cross comparison of ester isosteres to each other via MMP analysis shed further light on the relative stabilities of these functional groups (Figure 7). Generally, MMPs between most ester isosteres had HLM-N ΔCL_int within 2-fold of each other with the notable exception of acetamides. As previously observed during the analysis of relative amide stabilities, acetamides were broadly less stable than other amide isosteres. In fact, secondary acetamides were observed to be 2.6-fold less stable in HLM-N assays than isomeric N-methyl amides (Figure 7). The higher metabolic turnover of acetamides was also recapitulated in the HLM+N assay, with two exceptions (MMP_A = dimethylamide or H, MMP_B = acetamide). It is important to note the directionality of Figure 7, which is read starting from MMP_A, reading across rightward and then upward to MMP_B. To get data on reverse transformations (i.e., MMP_B to MMP_A), the reciprocal of each value must be taken.

Assessment of the MMP transformations with HLM-N ΔCL_int = 1.5–3.0 revealed a diverse range of chemotypes, suggesting that outside of esters, amides, and aldehydes, functional groups that predictably undergo turnover by HLM-N become sparser. Among this less metabolized category, however, transformations with MMP_B = oxetane formed a sizable cluster. Within the HLM-N database, there were 113 MMPs that terminated with an oxetane substructure; of these, 70% had ΔCL_int > 1, which supports the notion that oxetanes have some susceptibility to metabolism by HLM-N. Although the precise mechanism of this metabolic liability is outside the purview of this study, prior reports have demonstrated that epoxide hydrolases are competent enzymes for oxetane hydrolysis.³¹

Introduction of oxetane-containing functional groups onto their parent molecules (i.e., from MMP_A = H) generally increased turnover by HLM-N (Figure 8, entries 1–4). An MMP terminating with simple oxetane increased HLM-N metabolism by 1.28-fold. Extension to a methylene-linked oxetane resulted in slightly higher turnover, as did inclusion of a methyl group or ether linkage (Figure 8, entries 2–4). To assess potential oxetane isosteres, a range of cyclic- and acyclic-ether-containing MMPs were examined (Figure 8, entries 5–9). Methoxyethyl substitution was roughly 2-fold more stable by HLM-N than the corresponding methylene-linked oxetane, and five- and six-membered cyclic ethers likewise were more stable by HLM-N when compared with oxetane-containing functionality. A direct comparison of oxetane with its methylene-linked variant again suggested that more exposed oxetanes may impart higher risk for HLM-N turnover (Figure 8, entry 6). Since departing from perhaps better-characterized CYP-mediated metabolism can introduce additional uncertainties in human PK predictions, consideration of oxetane-containing lead molecules in HLM-N assays may be a useful way to derisk these compounds at an early stage. These results provide insight into the non-CYP-mediated liabilities of oxetanes, and further contextualizes prior reports describing the metabolic stability of these motifs relative to other cyclic ethers.^32,33

Figure 8.

MMP transformations terminating in oxetane-containing functionality (i.e., MMP_B = oxetane-containing functionality). Data is provided below each MMP in the format HLM-N ΔCL_int followed by MMP count number in parentheses.

An MMP analysis of Genentech’s HLM-N database has been conducted and revealed esters, amides, aldehydes, and oxetanes as functional groups that are commonly metabolized under these conditions. Outside of these functional groups, most molecular motifs appear to be stable in the HLM-N assay, underlining the relatively low prevalence of non-CYP-mediated metabolism by microsomes. These results support the notion that hydrolases are active components in these preparations. Given that hydrolases may be expressed at high levels extrahepatically, small signals in HLM-N assays may portend in vivo clearance liabilities and limit accuracy in the prediction of human pharmacokinetic profiles. Awareness of the functional groups that increase this liability may help mitigate these downstream risks through bioisosteric replacement at an early stage, or through derisking with additional in vitro DMPK studies. It is our hope that this study serves this purpose.

Glossary

Abbreviations

NADPH

the reduced form of nicotinamide adenine dinucleotide phosphate

MMP

matched molecular pair

MMP_A

the initiating motif of a matched molecular pair

MMP_B

the ending motif of a matched molecular pair

CYP

cytochrome P450 enzyme

HLM

human liver microsomes

HLM-N

human liver microsomes in the absence of NADPH

HLM+N

human liver microsomes containing NADPH

ΔCL_int

average fold change in intrinsic clearance

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.2c00071.

• Assay and computational methods, along with additional statistics for MMP transformations (PDF)

Author Contributions

The manuscript was written through contributions of all authors.

The authors declare no competing financial interest.