Detection and Quantification of Drug–Protein Adducts in Human Liver

Abstract

Covalent protein adducts formed by drugs or their reactive metabolites are risk factors for adverse reactions, and inactivation of cytochrome P450 (CYP) enzymes. Characterization of drug–protein adducts is limited due to lack of methods identifying and quantifying covalent adducts in complex matrices. This study presents a workflow that combines data-dependent and data-independent acquisition (DDA and DIA) based liquid chromatography with tandem mass spectrometry (LC-MS/MS) to detect very low abundance adducts resulting from CYP mediated drug metabolism in human liver microsomes (HLMs). HLMs were incubated with raloxifene as a model compound and adducts were detected in 78 proteins, including CYP3A and CYP2C family enzymes. Experiments with recombinant CYP3A and CYP2C enzymes confirmed adduct formation in all CYPs tested, including CYPs not subject to time-dependent inhibition by raloxifene. These data suggest adducts can be benign. DIA analysis showed variable adduct abundance in many peptides between livers, but no concomitant decrease of unadducted peptides. This study sets a new standard for adduct detection in complex samples, offering insights into the human adductome resulting from reactive metabolite exposure. The methodology presented will aid mechanistic studies to identify, quantify and differentiate between adducts that result in adverse drug reactions and those that are benign.

Introduction

Covalent protein adducts are associated with altered activity or function of their target enzymes.¹ Some drugs such as β-lactam antibiotics react directly with proteins to form adducts, and these adducts can cause hypersensitivity and adverse drug reactions.² Targeted covalent inhibitors form adducts with their pharmacological target to inhibit activity but may also form adducts to other proteins causing adverse drug reactions.^3,4 In some cases drug bioactivation via metabolic enzymes must occur before adduct formation.⁵⁻⁹ This bioactivation produces electrophilic metabolites that react with cellular nucleophiles like glutathione (GSH) or with nucleophiles in proteins resulting in adducted, inactivated or dysfunctional proteins.¹⁰⁻¹² For example, reactive metabolites can form adducts with cytochrome P450 (CYP) apo-proteins and cause time-dependent inhibition, which can lead to undesired drug–drug interactions.¹³

Protein adducts are typically very low abundance and often occur on peptides poorly suited to liquid chromatography with tandem mass spectrometry (LC-MS/MS) detection. For this reason, past MS based adductomic studies have often focused on purified proteins.^14,15 Recent work has begun to move to complex samples and has demonstrated several proteins and residues adducted by hepatotoxicants in the liver following drug exposure.¹⁶⁻¹⁸ However, these studies have been limited to compounds known to cause a high adduct burden, such as acetaminophen. The methods are labor intensive and involve elaborate experimental and data processing workflows to distinguish true from false identifications of adducted peptides.^18,19 These studies resulted in only a handful of confirmed adducted peptide identifications. In addition to providing valuable data, they highlight the difficulty of protein adduct detection–especially in complex biological matrices.

In the absence of streamlined proteomics methods, radiolabeled drugs are often used to show the global extent of protein adducts.^8,11,20 Such methods do not, however, identify the adducted proteins or peptides, and simple adduct burden does not differentiate hepatotoxicants from drugs not causing toxicities.²¹ Progress in identifying xenobiotic-protein adducts, characterizing the relationship between particular adducts and toxicological consequences, and in defining the global adductomic landscape for specific drugs in biological systems has therefore been limited. New methods and improved tools are needed to elucidate the link between the types of protein adducts formed, the specific proteins and residues adducted, and subsequent changes in protein function. These methods will help identify drugs that may cause adverse drug reactions.

Here, we report a workflow that leverages closed-mass search tools, yields high sensitivity, and can identify many adducts even in complex samples. Our LC-MS/MS proteomics-based method allows identification and quantification of the adductome formed by reactive metabolites in complex matrices. We applied this workflow to assess the relationship between adduct formation and altered protein activity using CYP enzymes and raloxifene. Raloxifene is metabolized by CYPs to a reactive metabolite, raloxifene diquinone methide, which forms adducts with glutathione, CYP3A4, and CYP2C8.^14,15,22,23 Raloxifene inactivates CYP3A4 and CYP2C8,^14,22 and results in adduct formation in human liver microsomes (HLMs) detected using radiolabeled raloxifene.²¹ It is thus an ideal compound to test the new adductomics workflow presented here.

Materials and Methods

Reagents and Drug Incubations

Details regarding chemicals, reagents, and drug incubations are described in the Supporting Methods of the Supporting Information (SI).

Inhibition of CYP3A4 and CYP2C9 by Raloxifene

Inhibition and potential inactivation of CYP2C9 by raloxifene was measured via an IC₅₀ shift assay using diclofenac and tolbutamide as substrates. CYP3A4 time-dependent inhibition kinetics by raloxifene was measured via midazolam hydroxylation. Details are described in Supporrting Methods.

Sample Preparation

HLM and supersome incubations were diluted with 50 mM ammonium bicarbonate or 50 mM tris(hydroxymethyl)aminomethane pH 8.5 to a final protein concentration of 0.45–0.66 μg/μL. Yeast enolase was spiked in as a process control at 16 ng enolase/μg total sample protein. Samples were reduced, alkylated and digested with trypsin for 6 h at 37 °C at an enzyme to substrate ratio of 1:15. Digestion was stopped by acidification. Pierce Peptide Retention Time Calibration Mixture (PRTC) was added as a system suitability standard. Samples were centrifuged at maximum speed in a benchtop microfuge for 10 min and supernatants were transferred to autosampler vials and stored at −80 °C until analysis. Full details can be found in Supporting Methods.

Mass Spectrometry

Sample digest (1.5 μg) was loaded onto a 150 μm Kasil fritted trap packed with 2 cm of ReproSilPur C18AQ (3 μm bead diameter, Dr. Maisch). Separation was done on a self-packed 75 μm ID 30 cm column at 0.25 μL/min using an acetonitrile gradient. An Orbitrap Exploris 480 (Thermo Fisher Scientific) was used to perform MS as described in detail in Supporting Methods. Two injections were done per sample, one in data dependent acquisition (DDA) mode and one in data independent acquisition (DIA) mode. Experiments specifically optimized to detect CYP3A4 active site peptide (C239) were done using a Bruker PepSep C18 15 cm × 150 μm, 1.9 μm column (part number 1893471) connected to an Orbitrap Eclipse Tribrid (Thermo Fisher Scientific) mass spectrometer. Further details are in the Supporting Methods.

Data Processing

Acquired spectra were converted to mzML format using ProteoWizard’s msConvert.²⁴ Adducts were identified from DDA data using the Comet²⁵ search algorithm with postprocessing using PeptideProphet²⁶ and PTMProphet²⁷ and visualization using the Limelight web application²³ as described in the results and Supporting Methods. All data were filtered at a false discovery rate (FDR) of 1% at the PSM level unless otherwise stated.

Peptide Quantification

Limelight’s “Download Blib Spectral Library” was used to create a spectral library of all peptides identified (adducted and unadducted) from the DDA data. This library was used to pick transitions for quantification of DIA data using Skyline.²⁸ Full details can be found in Supporting Methods.

Software Availability

A containerized Nextflow workflow to perform all steps of the adduct discovery pipeline is freely available at https://nf-ms-dda-tpp.readthedocs.io. The Limelight web application and associated documentation is available at https://limelight-ms.org/. Software to compare retention times and charge states of adducted peptides to their nonadducted counterparts is available at https://github.com/mriffle/raloxifene-paper-peptide-figure-notebook.

Data Availability

DDA data, including configuration files, can be downloaded or interactively viewed via Limelight at https://limelight.yeastrc.org/limelight/p/raloxifene-adducts. All raw data, including Skyline quantification, were deposited to the ProteomeXchange Consortium via Panorama Public²⁹ and are available under the data set identifier PXD054246 at: https://panoramaweb.org/raloxifene-adducts.url.

Results and Discussion

Adduct Discovery and Quantification Workflow

We previously developed an open-mass search based adduct discovery pipeline and software which compares untreated with treated samples to allow discovery of adduct modification masses resulting from exposure to xenobiotics.²³ Open-search workflows are capable of identifying drug–protein adducts regardless of whether the modification mass or residues modified are known or not. In the current work we present a new method leveraging closed-mass search tools that can be used once the adduct mass is known and which offers greater sensitivity than open-search methods. This workflow combines both data dependent aquisition³⁰ (DDA) and data independent aquisition³¹ (DIA) allowing comprehensive identification, quantification, and verification of low abundance adducted peptides from complex biological samples such as human liver microsomes (SI Figure S1, Supporting Note 1).

The closed search portion of this workflow uses Comet³² to search DDA data. False discovery rates are calculated based on PeptideProphet²⁶ probability scores and PTMProphet²⁷ is used for rescoring adduct localizations. These tools were built into a containerized Nextflow workflow which performs all steps of the analysis, including generation of Limelight XML files for import of results into Limelight.²³ The full software pipeline is freely available (https://nf-ms-dda-tpp.readthedocs.io/) and can be run on Windows, Mac, or Linux using a single command.

Our data viewing software, Limelight, allows for easy viewing of MS proteomics data.²³ We added features to streamline our hybrid DDA/DIA workflow. Limelight allows comparison of treated and untreated samples and highlights evidence such as search scores, annotated spectra, peptide retention times, charge states, etc., allowing further verification of adduct identifications. Some drugs produce adducts that fragment predictably in the MS producing specific ions that are present in spectra of adducted peptides. Limelight allows users to filter their data on these specific ions. This helps screen data for adduct identifications in cases where adducts are labile. Limelight also allows automatic generation of spectral libraries from DDA data including adducted peptide spectra that can be used directly by Skyline²⁸ to produce and quantify extracted ion chromatograms (XICs) from DIA data (SI Figure S1).

Identification of Proteins Adducted by Raloxifene in Human Liver Microsomes

Human liver microsomes (HLMs) from three individual liver donors were incubated with vehicle, raloxifene-d₀ or raloxifene-d₄ in the presence of NADPH. Samples were proteolytically cleaved using trypsin and the resulting raloxifene diquinone methide adducts were identified using our Nextflow adduct discovery workflow (adduct mass 471.15 or 475.18 Da for raloxifene-d₀ or raloxifene-d₄, respectively).

In the entire data set 1,984,026 peptide spectrum matches (PSMs) were observed at a 1% false discovery rate (FDR) of which only 495 (0.025%) PSMs represented peptides containing a raloxifene adduct. This highlights the challenge of detecting adducted peptides in complex matrixes.

Raloxifene adducts were found in 78 proteins (127 unique peptides) in raloxifene treated samples. These included CYPs, several ER metabolic enzymes (e.g., retinol dehydrogenase, alcohol dehydrogenase, microsomal glutathione S-transferase and carboxylesterase), ribosomal components, and various binding proteins (Figure 1a, and SI Table S1). The discovery of 127 raloxifene-adducted peptides in various HLM proteins marks a significant success compared to the 5–6 unique adducted peptides identified in prior acetaminophen studies^18,33 and the single raloxifene-CYP adduct previously identified in a native membrane environment.²²

Figure 1.

Raloxifene adducts identified in HLM incubations from 3 individual donors. (a) Molecular function Gene Ontology (GO) terms for proteins containing raloxifene modifications are presented as a hierarchical graph. Each term is labeled with its accession number, name, and the number of PSMs and distinct proteins containing raloxifene modifications. Terms with a PSM count under 125 were removed and the graph was limited to 4 levels. More intense shading represents more PSMs. (b) Number of peptides containing raloxifene modification mass detections in untreated (green) and raloxifene treated samples (purple) classified by the amino acid modified. One PSM in untreated sample had 2 modifications. Thus, 5 PSMs result in 6 locations for untreated modifications. (c) Raloxifene modified peptides found in CYPs are listed along with the number of PSMs in untreated versus raloxifene treated samples. Residues adducted by raloxifene are represented by [*]. Alkylation and oxidation are indicated by [57] and [16], respectively. Results represent comet searches allowing a variable modification of 471.15 Da (d₀-raloxifene) or 475.18 Da (d₄-raloxifene) postprocessed with TPP and filtered at 1% FDR. Raw data are available on Limelight at: https://limelight.yeastrc.org/limelight/go/LzWY1QCWYR.

Adducts were predominantly in cysteine residues (Figure 1b) and were treatment specific: only 5 out of the 500 PSMs with a 471.15 or 475.18 Da modification were identified in vehicle treated HLMs. Raloxifene modification of cysteine was found to protect the cysteine residue from alkylation yielding evidence for covalent modification of cysteine. The majority of the raloxifene modifications observed, regardless of whether they were localized to a cysteine, tryptophan or tyrosine, contained MS/MS ions incorporating the adduct mass providing further evidence for covalent modification of these residues. The detection of 1% of total adduct identifications falling in untreated samples corresponds with the 1% FDR cutoff used for this analysis. This indicates the workflow developed here was performing well as all identifications in untreated samples are false discoveries.

Previous adductomics studies performed multiple steps to remove incorrect adduct identifications.¹⁸ The method presented here incorporates the recently introduced Variable Modification Count (VMC) model. The VMC model assists PeptideProphet in classifying PSMs containing variable modifications defined by the search engine parameters by modeling the distribution of VMCs, for each type of variable modification used in the search, among correct and incorrect search results. This Bayesian model takes advantage of the observation that higher variable modification counts are more likely to occur among incorrect (random) PSMs, than among correct PSMs. Incorporating the VMC model when calculating PeptideProphet probabilities allows PeptideProphet to better control error rates on peptides containing rare modifications, while preserving rare modification containing PSMs with strong spectral and other complementary evidence.

Analysis of the same data without the VMC model resulted in 1617 total PSMs with 471.15 or 475.18 Da modification but 509 of these (31%) were from untreated samples. These data highlight the effectiveness of the VMC model for the current application (SI Table S2). An additional target/decoy database was used as an “entrapment” database to allow independent assessment of FDR.^34,35 Despite a ratio of 1.5:1 entrapment proteins:target proteins in the search database, no adduct identifications were to entrapment peptides, providing further confidence in the results.

Most CYPs have been shown to form raloxifene diquinone methide.²² During in vitro incubations, the diquinone methide metabolite leaves the active site of the CYP and reacts with GSH in solution forming glutathione adducts.^22,36 The data presented here confirm the metabolite is reacting with proteins and residues in HLMs beyond the CYPs that form the metabolite. These data are the first to identify the specific proteins and peptides adducted by raloxifene in human livers.

In previous work, 4 proteins were identified as adducted by acetaminophen in rat liver microsomes.¹⁸ Two of these proteins were also found here as adducted by raloxifene in the equivalent residues of the human homologues. The rat microsomal glutathione S-transferase 1 peptide VFANPEDC*AGFGK was adducted by acetaminophen at C50, the active site cysteine.¹⁸ We observed both heavy (475.18 Da) and light (471.15 Da) raloxifene adducts in the same cysteine of the human equivalent peptide in HLMs, VFANPEDC*VAFGK, in MGST1 (SI Table S3). The rat microsomal CYP2C6 protein was also observed as adducted by acetaminophen at residue C372 in peptide FIDLIPTNLPHAVTC*DIK. Likewise, we observed C372 (peptide YIDLLPTSLPHAVTC*DIK) in the human homologue CYP2C9, to be adducted by both heavy and light raloxifene in human liver microsomes (SI Table S3).

These observations are significant as they illustrate that the same residues in liver microsome proteins from different organisms are adducted by different drugs. The findings of adducts in these proteins by acetaminophen and raloxifene may suggest that these proteins and residues are particularly susceptible to adduction due to reactivity, protein abundance or exposed residues in the protein structure.

Comparison of Raloxifene-Adducted Peptides to Their Nonadducted Counterparts

Like most pharmaceutical drugs, raloxifene is lipophilic. Raloxifene-adducted peptides should therefore be more lipophilic than their unadducted counterparts and elute later during reversed phase liquid chromatographic separation of peptides. Likewise, modifications can affect peptide charge during electrospray ionization. We built tools to track and compare retention times and peptide precursor charge states between PSMs corresponding to adducted peptides and PSMs corresponding to their unadducted counterparts. These tools can be applied to any drug-peptide adducts.

The developed software (https://github.com/mriffle/raloxifene-paper-peptide-figure-notebook) was used to compare HLM PSMs corresponding to peptides identified as raloxifene-adducted against PSMs corresponding to their unadducted peptide counterparts. In this analysis, PSMs were collated for all peptides containing either a 471.15 (raloxifene-d₀) or 475.18 (raloxifene-d₄) modification and for the unadducted counterparts of those same peptides. The number of PSMs, retention times and charge states were then compared for each peptide type (unadducted, 471-adducted and 475-adducted).

Raloxifene-adducted peptides had a mean retention time 31.5 min later than their unadducted counterparts, and a typical charge state +1 or +2 higher than their unadducted equivalent peptides (Figures 2 and SI Figure S2). Overall, 86 out of 94 peptides had a later retention time in raloxifene-adduced form compared to their unadducted counterparts and 90 out of 94 peptides had a higher charge state in raloxifene-adduced form (SI Figure S2). Fewer total raloxifene-d₄ (475-adducted) PSMs were observed versus 471-adducted PSMs, as fewer LC-MS/MS runs were performed using raloxifene-d₄ labeled samples. The total number of PSMs corresponding to 471 and 475-adducted peptides combined was 495, 3% of the number of PSMs corresponding to the exact same peptides with no adduct modification (14,650). The low number of adducted peptide PSMs relative to their unadducted counterparts further highlights the challenge of detecting adducted peptides in complex matrixes.

Figure 2.

Retention time and charge state distributions for PSMs of raloxifene adducted peptides relative to PSMs of their unadducted peptide counterparts from HLMs. (a) The mean retention time of the PSMs for each unadducted peptide was defined as 0 and all individual PSM retention times for corresponding unadducted and adducted peptides were calculated as a delta in minutes relative to that value. (b) The lowest observed charge state in the PSMs for each unadducted peptide counterpart was defined as 0 and all individual PSM charge states were calculated as a delta to that value. Raw data are the same as for Figure 1.

These data demonstrate that raloxifene-adducted peptides generally have a later retention time than their unadducted counterparts in the reversed phase chromatographic system used here. This is likely due to the hydrophobic nature of raloxifene. They also suggest that the raloxifene moiety typically carries a charge in addition to that carried by the peptide, resulting in increased charge states for raloxifene-modified peptides. This observation holds true even for late eluting peptides where the raloxifene adduct has less or no effect on retention time compared to earlier eluting peptides (SI Figure S2). This evidence is independent of the mass spectrum and therefore useful as it can provide additional confidence in identifications of raloxifene-adducted peptides.

Relative Quantification of Raloxifene-Adducted Peptides in Human Liver Microsomes

Previous proteomics-based adduct studies have focused primarily on discovery.^14,17,18,22 The few adducted peptides found have been generally assumed to impact protein activity. The current work identified many adducts, and thus questions regarding the relationship between the specific adducts formed, the proteins and peptides adducted, the relative abundance of individual adducts, and the target protein’s activity become tenable. The workflow presented here uses DIA data to allow verification of adduct identifications and quantification of the relative abundance of raloxifene-adducted peptides (Supporting Note 1).

Relative quantification of peptides allows a given peptide to be compared across different samples to evaluate changes in abundance of that specific peptide across the different samples. The signal generated by different peptides cannot be used to evaluate differences in abundance between different peptides nor to measure the absolute amount of a given peptide. This is because different peptides have different properties that affect signal in electrospray MS, for example ionization and desolvation. To answer such questions isotope labeled internal standards must be used. The production of such standards in known quantities is challenging for adducted peptides.

The advantage of the workflow presented here is that it does not require labeled standards and yet can measure how the abundance of specific adducted and unadducted peptides change across different samples and between donors.

DIA data from the three HLM donors was analyzed. Signals for raloxifene-adducted peptide transitions were detected in all three donors only in the raloxifene treated samples and no adducted peptide signal was detected in the vehicle treated samples (Figure 3, purple panel). This yields confidence in the assignment of these peptides as raloxifene-adducted. Furthermore, like the DDA data presented in Figure 2, the DIA data show the impact of raloxifene modification on peptide retention time. Raloxifene adducted peptides eluted 30 to 50 min later than their unadducted peptide counterparts (Figure 3, purple shaded data versus green shaded data). This shift in retention time, resulting from increased peptide lipophilicity in adducted peptides, could result in loss of detection of some adducted peptides due to extensive retention on column. This can make evaluation of adducted peptides challenging. An example of this is the tryptic CYP3A4 active site peptide, FDFLDPFFLSITVFPFLIPILEVLNICVFPR, discussed below. Adducted peptide signal was typically ∼3 orders of magnitude less than that of unadducted peptides, further highlighting the difficulty of detecting and quantifying adducted peptides.

Figure 3.

Quantification of peptides in untreated and raloxifene treated human liver microsome (HLM) preparations from 3 individual donors. Representative extracted ion chromatograms (XICs) and resulting peak areas are shown for fragment ion transitions from a selection of peptides. Yellow (top) panels: Control peptides. Yeast enolase (ENO1) was spiked into all samples as a process control. GP3 (GAPDH) is an endogenous housekeeping protein. Quantification of representative unadducted peptides from proteins found to be modified by raloxifene are also shown. Green (left) panel: Quantification of unadducted peptides corresponding to the raloxifene-adducted peptides shown in the purple (right) panel. Protein name, peptide sequence and precursor transition charge state are shown for each peptide. Example XICs are shown for treated and untreated HLM donor 1 samples. Peak areas are shown for all three donors. Horizontal green bars indicate untreated samples, and horizontal purple bars indicate raloxifene treated samples. Raw Skyline quantification data are available at https://panoramaweb.org/raloxifene-adducts.url.

Yeast enolase, which was spiked into all samples before digestion as a process control,³⁷ showed excellent reproducibility between runs with a coefficient of variation (CV) of 3.9% (Figure 3). Measurement of the housekeeping protein, glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gave a mean CV of 12.6% between donors indicating total protein was reasonably controlled between donors. As expected, the expression of CYP3A, CYP2C and other nonhousekeeping proteins varied substantially between donors (mean CV of 37%). CYP3A4 expression in the individual donors was assessed based on a previously reported quantitative peptide³⁸ (LSLGGLLQPEKPVVLK) and showed a mean CV of 43.9% between donors. However, comparing untreated to treated samples within a donor, the mean CV of all surrogate peptides from endogenous proteins presented in Figure 3 (yellow panel excluding enolase) was 7.7% indicating that raloxifene treatment did not cause high variation between treated and untreated samples when looking at unadducted peptides.

The variation of unadducted peptide counterparts of adducted peptides between donors (Figure 3, green panel) was similar to the surrogate peptides in the yellow panel (mean CV 40.9%). For unadducted peptide counterparts, however, the untreated to treated variation within a donor was about 2-fold higher (CV 18.7%) than for this comparison done on the unadducted surrogate peptides. This is expected given that the unadducted counterpart peptide pool will be depleted when those peptides are adducted during treatment. Despite this increase in variability, in many cases a clear relationship between adduction of a given peptide and depletion of its unadducted counterpart could not be observed.

The variation in the abundance of raloxifene-adducted peptides was the highest among all the peptides measured (mean CV of 57.8%) with the MGST1 adducted peptide appearing to be a notable exception. Since CYP3A4 has been proposed to be the predominant enzyme forming the diquinone methide metabolite,³⁶ one may expect that adduct formation would be greater with higher CYP3A4 expression. However, the abundance of specific adduct formation did not show an obvious relationship with CYP3A4 abundance. These data suggest that adduct formation in different proteins in the liver cannot be easily predicted by the expression of a single enzyme forming a reactive metabolite.

A Recombinant System Allows Verification of CYP-Raloxifene Adducts and Their Effects on CYP Activity

In the current study, 15 adducted peptides were identified in 9 CYP enzymes from raloxifene treated HLMs (Figure 1c). Ten of the CYP-raloxifene peptides detected in HLMs were unique to specific CYP enzymes (Figure 1c). CYPs from a given subfamily share at least 40% sequence identity and thus share many tryptic peptides (SI Figures S3 and S4). Five of the identified raloxifene-adducted peptides were common to more than one CYP.

Previous studies in supersomes showed that of all the CYPs tested, only CYP3A4 and CYP2C8 were subject to time-dependent inhibition by raloxifene.²² Furthermore, raloxifene adducts have previously only been specifically identified in these two CYPs.^14,15,22,23 It has thus far been plausible to hypothesize that raloxifene adducts typically cause time-dependent inhibition of CYP enzymes.

To gain deeper insight into the human CYP enzymes adducted by raloxifene and the effect of these adducts on CYP activity, supersomes (insect cell microsomes) were incubated with raloxifene-d₀ or raloxifene-d₄ in the presence of NADPH and the incubations analyzed using our analysis pipeline. In this study, supersomes expressing a single human CYP enzyme (CYP3A4, CYP3A5, CYP2C8, CYP2C9 or CYP2C19) along with human cytochrome P450 reductase (reductase) and Cytochrome b₅ were tested.

Raloxifene-CYP adducts were found in all raloxifene treated supersome samples tested (SI Figure S5 and Table S4). Notably, reductase adducts were not detected in HLMs but were abundant in incubations with CYP3A4, CYP3A5 and CYP2C9 supersomes. This may be due to the high expression of reductase in the supersome system. The reductase to CYP ratio in the supersomes greatly exceeds that present in native HLMs but does vary between supersome batches.

Adduct identifications were treatment specific with only 1 of the 164 PSMs identified in CYPs or reductase showing an identification in vehicle treated supersomes. Like raloxifene-adducted HLM peptides, raloxifene-adducted supersome peptides had a mean retention time 26.7 min later than their unadducted counterparts, and 43:48 peptides had a higher charge state than their unadducted equivalent peptides (SI Figures S6 and S7).

These data present a more complex picture than previous studies and suggest that while some adducts affect CYP activity, many CYP-raloxifene adducts do not directly alter activity. Specific data supporting this hypothesis is described below.

CYP3A4 is Both Adducted by and Inactivated by Raloxifene

Several publications have presented the active site cysteine in CYP3A4 (C239) as being the only residue identified as adducted by raloxifene.^14,22,39 This adduct was previously shown to be necessary for raloxifene mediated time-dependent inhibition of CYP3A4.^14,39 We previously showed that when purified CYP3A4 is incubated with raloxifene, at least 12 additional CYP3A4 residues are adducted.²³ The detection of CYP3A4 C58, C98, and C468 adducts in addition to C239 in the HLM results presented here supports the finding that multiple residues in a given CYP can be adducted. These data are important as HLMs represent CYP3A4 in its native membrane environment, making artifacts in protein folding and aggregation less likely than in past experiments using purified proteins.

Except for the adduct on C239, all the adducts identified in HLMs in CYP3A4 peptides, and in peptides shared between CYP3A4 and other CYPs, were recapitulated in the CYP3A4 supersome results (SI Figure S5 and Table S4). Initial experiments with CYP3A4 supersomes failed to detect raloxifene-adducted CYP3A4 C239 (SI Figure S5). This result is not surprising given the length and hydrophobicity of unadducted peptides spanning C239 ([LLR]FDFLDPFFLSITVFPFLIPILEVLNIC*VFPR). We developed an optimized chromatographic method to improve detection of these peptides. This method resulted in clear detection of the peptide containing C239 in both DDA and DIA data (SI Figure S8) and these data confirm that CYP3A4 C239 is adducted by raloxifene as previously observed.

In summary, these results show that many additional raloxifene-adducted peptides in CYP3A4 are formed by raloxifene exposure beyond just C239 and that not all adducted peptides are amenable to detection by standard methods.

CYP3A5, CYP2C9 and CYP2C19 are Adducted by, but Not Inactivated by Raloxifene

Treatment specific raloxifene adducts were detected in CYP3A5, CYP2C9 and CYP2C19 supersome incubations, yet according to past work, none of these enzymes are subject to time-dependent inhibition by raloxifene.²²

The peptide common to CYP2C8, CYP2C9, CYP2C18 and CYP2C19 in HLMs, SPC*MQDR was found to be raloxifene-adducted in all the CYP2C family supersome incubations tested (SI Figure S5). In HLMs, therefore, this peptide is probably derived from a combination of CYP2C proteins. This residue may be highly reactive or easily accessible for reactive metabolites or particularly amenable to LC-MS/MS as it was detected as adducted in all CYP2C supersome incubations.

CYP2C9 supersome incubations yielded the second highest number of raloxifene-adducted peptides after CYP3A4 (SI Figure S5). Of special significance was the identification, in CYP2C9 supersomes, of a raloxifene adduct on the active site cysteine, C216, ILSSPWIQIC*NNFSPIIDYFPGTHNK (SI Figure S5 and Figure 4). DIA data corresponding to this peptide was analyzed in untreated, raloxifene-d₀ and raloxifene-d₄ treated CYP2C9 samples. Extracted ion chromatograms showed signal for raloxifene-d₀ and raloxifene-d₄ adducts only in their respective raloxifene-d₀ and raloxifene-d₄ treated samples and adducted peptide signal was absent from the vehicle-only control (Figure 4). The unadducted peptide was present in all three samples (SI Figure S9). It was previously suggested that an adduct in the active site cysteine would likely cause time-dependent inhibition.²² This suggestion was based on structural alignments between CYPs showing that C216 and C225 in CYP2Cs are in equivalent locations to CYP3A4 C239, described above.^14,39 We confirmed that raloxifene does not inactivate CYP2C9 but does inactivate CYP3A4 (SI Figure S10). The fact that CYP2C9 C216, along with 5 other distinct CYP2C9 residues are adducted by raloxifene and yet raloxifene does not inactivate CYP2C9 indicates that raloxifene-adduction at multiple sites including C216 does not result in time-dependent inhibition.

Figure 4.

CYP2C9 active site C216 is modified by raloxifene. Annotated spectrum of raloxifene-d₀ adducted C216 peptide (ILSSPWIQIC*NNFSPIIDYFPGTHNK) plus extracted ion chromatograms (XICs) and integrated peak areas for this peptide from untreated, raloxifene-d₀ and raloxifene-d₄ treated recombinant CYP2C9 supersome samples. Precursor signal is specific to raloxifene-d₀ sample. Quantified product ions are shared between raloxifene-d₀ and d₄ samples, are in the same DIA isolation window, and therefore show up in both treated samples. Expanded figure is presented in SI Figure S9.

Taken together the data presented here suggest that many CYP-raloxifene adducts do not directly alter CYP activity. Yet, these adducts might still have biological consequences. The formation of these adducts in human liver may result in formation of adducted proteins that are recognized by the immune system leading to an immune response. The formation of adducts may also alter protein–protein or CYP-membrane interactions in the liver in vivo.

Conclusions

We recently described a novel proteomics workflow for identifying unknown protein adducts²³ that focused on identifying the specific modification masses resulting from exposure to reactive metabolites and intermediates. In this current work, we present methods to identify the proteins and peptides adducted by reactive metabolites once the modification mass is known, allowing for greater sensitivity. This work demonstrates that proteomics-based LC-MS/MS methods can be used to identify and characterize adducts formed by xenobiotics in complex biological matrixes such as human liver. The results show that reactive metabolites can leave the active site of the CYP and form many adducts in HLM proteins. They also suggest that in complex systems, adduct formation may depend on multiple factors. Such factors could include formation rate of the reactive metabolite, expression level of the potentially adducted proteins, reactivity and accessibility of specific residues and reactivity of the metabolite formed. We used CYP enzymes as a test case for the relationship between adduct formation and change in protein activity. The data presented here show that many adducts can be formed in CYP enzymes without an observable change in CYP activity. It is thus plausible that some adducts are benign, and points to the need for future mechanistic studies to define modification-activity relationships with specific proteins. The data analysis workflow presented here has been packaged and made freely available for execution via a single command.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.4c00663.

• Supplementary Methods; Supplementary Note 1: Adduct Discovery and Quantification Workflow; Supplementary Note 2: Human Liver Microsome Results; Supplementary Note 3: Recombinant CYP Insect Cell Supersome Results (PDF)

Author Contributions

A.Z., M.R., G.Z. and N.I. conceived the experiments. G.Z., N.I. and E.B.R. performed drug–protein incubations. A.Z. carried out the MS experiments. A.Z. performed the MS data analysis with help from M.R., D.D.S. and M.R.H. M.R. and D.J. developed Limelight. D.D.S. developed the PeptideProphet VMC model. The manuscript was written by A.Z. and N.I. with contributions from all authors. All authors discussed the results and commented on the manuscript. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.