Identification and characterization of a novel aldehyde metabolite of WIN18,446 and associated WIN18,446-ALDH1A2 protein adducts using mass spectrometry

Abstract

ALDH1A family enzymes (ALDH1A1, ALDH1A2, and ALDH1A3) catalyze retinoic acid synthesis, and their dysregulation is linked to disease. Selective inhibitors of these enzymes have been tested in drug discovery programs and one such compound, WIN18,446, was found to irreversibly inhibit ALDH1A2. WIN18,446 is a reversible male contraceptive in humans and animals. The inhibition of spermatogenesis by WIN18,446 is thought to be due to inhibition of ALDH1A. However, the mechanism of irreversible inhibition of ALDH1A2 by WIN18,446 is not known. A crystal structure obtained after incubating ALDH1A2 with WIN18,446 revealed a WIN18,446-derived metabolite covalently adducted to the catalytic cysteine, C320. Inspection of this structure suggested that the observed adduct is unstable and may be a metabolic intermediate stabilized under crystallographic conditions. In the current work, we tested this hypothesis. We identified and characterized an aldehyde metabolite of WIN18,446, which we designated M-54. M-54 is likely the metabolite of the intermediate observed in the crystal structure. Using a range of proteomics techniques, we identified a WIN18,446-derived ALDH1A2 protein adduct of mass 292.07 Da on C319 of ALDH1A2. This adduct may result from reaction of the crystal structure metabolic intermediate. Using the identified mass, we probed human liver samples from multiple donors and found WIN18,446 specific adducts on cysteines within the ALDH1A1 and ALDH2 active site regions. The current study provides new insight into the metabolism of WIN18,446 and the mechanism of inhibition of ALDH1A2. We also demonstrate a proteomics workflow for identifying and validating drug–protein adducts of unknown mass.

Introduction

Retinoic acid is the active metabolite of vitamin A, an essential micronutrient. Retinoic acid signaling is necessary for regulation of cell cycle and differentiation in a variety of organs. The concentrations of retinoic acid in tissues are regulated by the availability of the precursor retinol, biosynthesis via alcohol and aldehyde dehydrogenases, and metabolism via cytochrome P450 enzymes (Blaner, 2019; Esposito et al., 2024; Gudas, 2022; Napoli, 2017). The aldehyde dehydrogenase enzymes in the 1A family, ALDH1A1, ALDH1A2 and ALDH1A3, are critical for the synthesis of retinoic acid in target tissues. Altered activity of the ALDH1A enzymes has been implicated in several disease states in conjunction with altered retinoic acid concentrations (Zhao et al., 2025). As such, ALDH1A enzymes are considered potential drug targets (Buchman & Hurley, 2017; Koppaka et al., 2012) and several selective inhibitors of ALDH1A1, ALDH1A2 and ALDH1A3 have been explored in preclinical drug discovery programs (Esposito et al., 2024; Hogarth et al., 2011; Kimble-Hill et al., 2014).

ALDH enzymes are a large family of enzymes that catalyze detoxification reactions of lipid peroxidation-derived aldehydes (Muñoz-Clares et al., 2017). Within the ALDH family, ALDH1A1, ALDH1A2 and ALDH1A3 are tetrameric enzymes (Castellví et al., 2022) that catalyze the oxidation of retinaldehyde to retinoic acid in a tissue and cell type specific manner (S. L. M. Arnold et al., 2015; Rubinow et al., 2022). A closely related member of the ALDH family, ALDH2, catalyzes the oxidation of acetaldehyde, the oxidation product of ethanol, to acetic acid (Zakhari, 2006). The catalytic mechanism of the ALDH enzymes, resulting in oxidation of aldehydes to carboxylic acids, involves a catalytic cysteine, nicotinamide adenine dinucleotide (NAD⁺) as a cofactor, and a glutamic acid within the active site (Koppaka et al., 2012; Shortall et al., 2021). The three ALDH1A enzymes have two cysteines within the active site while ALDH2 has three. These active site cysteines are conserved across the ALDH family (Koppaka et al., 2012). In each case a single cysteine is thought to be the catalytic cysteine and is critical for the dehydrogenase activity. For C319/C320 in ALDH1A2, C320 is the catalytic cysteine. For C302/C303 in ALDH1A1 C303 is the catalytic cysteine (Tsou et al., 2011). For C318/C319/C320 in ALDH2, C319 is the catalytic cysteine (Kim et al., 2024). The neighboring cysteines within the active site motif may protect the catalytic cysteine from irreversible oxidation (Muñoz-Clares et al., 2017).

WIN18,446 (Figure 1A) is an inhibitor of ALDH1A1, ALDH1A2, ALDH1A3 and ALDH2 (S. L. Arnold et al., 2015; S. L. M. Arnold et al., 2015; Haenisch et al., 2024). Following administration of WIN18,446 to mice, the tissue retinoic acid concentrations are decreased in liver and testis due to the inhibition of ALDH1A enzymes (S. L. M. Arnold et al., 2015). In addition, the tissue retinoic acid synthesis activity is decreased following WIN18,446 administration (S. L. M. Arnold et al., 2015), likely due to irreversible inhibition of ALDH1A by WIN18,446 (Hogarth et al., 2015; Paik et al., 2014, 2025). Notably, WIN18,446 is a reversible male contraceptive in humans and rabbits, although in humans WIN18,446 also causes a “disulfiram reaction” when it is coadministered with alcohol (Amory et al., 2011; Heller et al., 1961, 1963). The inhibition of spermatogenesis by WIN18,446 has been associated with its inhibition of ALDH1A and of testicular retinoic acid synthesis. The disulfiram reaction, on the other hand, is associated with inhibition of ALDH2 by WIN18,446 (Haenisch et al., 2024; Paik et al., 2014).

Figure 1.

WIN18,446 and related structures. (A) The chemical structure of WIN18,446. (B) The covalent adduct observed by Chen et al., 2017 in the co-crystal structure of ALDH1A2 with WIN18,446, which we hypothesize is an intermediate structure. (C) The proposed mechanism of the formation of an aldehyde metabolite from WIN18,446 by ALDH1A2. In panel C, the proposed mechanism is that the catalytic thiol (Cys-S⁻) from the cysteine in the active site of ALDH1A2 displaces Cl⁻ leading to formation of intermediate 1. This species is observed in the co-crystal structure in panel B. The next step is Cl⁻ loss to form a thiocarbonyl sulfonium intermediate, 2, which can be attacked by hydroxide to form a thiohemiacetal intermediate, 3, which subsequently yields an aldehyde metabolite, M-54.

The effective inhibition of the ALDH1A enzymes by WIN18,446 has made it an attractive tool for mechanistic studies on retinoic acid signaling (Beedle et al., 2019; Hogarth et al., 2015; Kent et al., 2016). For example, treatment of neonatal mice with WIN18,446 allows blockage of spermatogenesis that can be restarted by single injection of retinoic acid resulting in synchronized spermatogenesis across the testis (Hogarth et al., 2015). Due to its widespread use as a lead compound for development of ALDH1A inhibitors and as a tool for biochemical and physiological studies, there is great interest in better understanding the mechanism of inhibition of ALDH1As by WIN18,446.

The mechanism of ALDH1A2 inhibition by WIN18,446, has been explored previously. WIN18,446 is a time-dependent, irreversible inhibitor of ALDH1A2 and the inhibition cannot be reversed by DTT treatment (S. L. Arnold et al., 2015; S. L. M. Arnold et al., 2015; Paik et al., 2014). Yet, the mechanism of the irreversible inhibition has not been unequivocally determined. A crystal structure of ALDH1A2 incubated with NAD⁺ and WIN18,446 was solved (Chen et al., 2017). In the resulting structure, a WIN18,446-derived species formed during the incubation was observed covalently adducted to the active-site cysteine, C320 (Figure 1B and structure 1 in Figure 1C). The authors suggested this as a plausible structural basis for irreversible inhibition.

Based on the chemistry of the adduct observed in the crystal structure, the adduct is expected to be unstable. We hypothesized that the ALDH1A2–WIN18,446 derivative complex (Figure 1B) represents a transient metabolic intermediate that was stabilized under crystallographic conditions. We further hypothesized that this intermediate is unstable during incubation of ALDH1A2 with WIN18,446 in solution and that it subsequently converts to an aldehyde metabolite, to which we designate the name M-54 to reflect the 54 Da loss in molecular weight (Figure 1C). If this hypothesis is true, the adduct observed in the crystal structure is unlikely to cause the irreversible inhibition of ALDH1A2. The goal of the current work was to test these hypotheses and to further investigate the inactivation mechanism of ALDH1A2 by WIN18,446 using novel mass spectrometry based adductomics methods.

Materials and Methods

Expression and purification of ALDH1A2

The ALDH1A2-His expression vector containing the full-length human ALDHA1A2 in a pET28a backbone (Supplementary Figure S1) (S. L. Arnold et al., 2015) was transformed into Rosetta2 (DE3) pLysS bacteria. A 1L culture of LB-kanamycin-chloramphenicol was inoculated from a 5 mL overnight starter culture and grown at 37°C with 220 rpm shaking to OD 0.6. ALDH1A2-his expression was induced with 0.1 mM IPTG for 18 hours at 20°C with 220 rpm shaking. Bacteria were harvested by centrifugation at 7,000 g for 15 mins at 4°C in Avanti-JE centrifuge (Beckman Coulter). The supernatant was decanted, and the bacterial pellet was washed with ice-cold PBS (phosphate buffered saline) and pelleted at 4,000 g for 20 mins at 4°C in 5910Ri centrifuge (Eppendorf). Final bacterial pellets were stored at −80°C prior to purification.

Bacterial pellets were resuspended in lysis buffer [20 mM Tris pH 7.9, 500 mM NaCl, 30 mM imidazole, 2 mg/mL lysozyme (Roche cat no. 10837059001), cOmplete Mini EDTA-free protease inhibitor cocktail (Roche cat no. 11836170001), 1 mM phenylmethylsulfonyl fluoride (PMSF) (Sigma cat no. 1P-7626), and 25U benzonase (Sigma cat no. E8263)] at a 1:20 ratio of cells-to-buffer on ice and stirred at 4°C for 1 hour. Cells were disrupted by sonication in 8 rounds of 15, 1-second-long pulses. Power was set at 80%. The lysate was cleared by centrifugation for 60 mins at 142,000 g on an Optima XE-90 ultracentrifuge (Beckman Coulter) and the supernatant was collected for protein purification.

Protein purification was performed using an NGC FPLC chromatography system (Bio-Rad Laboratories, CA, USA) equipped with a UV detector and an automated fraction collector. The cleared lysate was loaded at 0.5 mL/min on a 1 mL Ni column (HisTrap HP, Cytivia, cat no. 29051021) which had been equilibrated with Ni Wash Buffer (20 mM Tris, 500 mM NaCl, 30 mM imidazole, pH 7.9) (Supplementary Figure 2A). Once loaded, the column was washed at 1 mL/min with 20 column volumes of Ni Wash Buffer. The ALDH1A2 protein was eluted at 1 mL/min with 10 column volumes of elution buffer (20 mM Tris with 500 mM NaCl and 1M imidazole pH 7.9) and 0.5 mL-fractions were collected. The peak fractions for the eluted protein peak were pooled and injected on a size exclusion column (Superdex200 HiLoad16/600, Cytiva, cat no. 28989335) which was equilibrated with SEC buffer (20mM Hepes pH 8.5, 150mM KCl, 1mM EDTA) at 0.5 mL/min (Supplementary Figure S2B). Fractions of 1.5 mL were collected and the purity of the eluted protein assessed by gel electrophoresis using a 10% SDS gel (BioRad) (Supplementary Figure S2C). The eluted ALDH1A2-His-containing fractions were pooled, and Tris(2-carboxyethyl)phosphine hydrochloride (TCEP) was added to a final concentration of 1mM. Final protein concentration was determined using the BCA assay (Pierce^™ BCA Protein Assay Kit - Reducing Agent Compatible, cat no. 23250). Purified protein was stored at 4°C, and all activity assays were performed within 2 months.

The activity of the recombinant ALDH1A2 was confirmed as described previously (S. L. Arnold et al., 2015) with minor modifications. In brief, samples containing 10 nM ALDH1A2 and 50 nM all-trans-retinal (Sigma cat no. R2500) in 20 mM Hepes KOH, 150 mM KCl, pH 8.0 were preincubated for 4 mins at 37°C prior to initiation of the reaction with 2 mM NAD⁺ cofactor (Sigma cat no. N6522). After incubation at 37°C for 2 mins, the reaction was terminated by adding an equal volume of pre-chilled acetonitrile with 200 nM all-trans-retinoic acid-d₅ (LGC Standards cat no. TRC-R250202), followed by centrifugation at 15,000 g for 10 mins at 4°C. The cleared samples were analyzed by LC-MS/MS using a Sciex 6500 QTRAP as previously described (S. L. Arnold et al., 2015). For the quantification, all-trans-retinoic acid (Sigma cat no. PHR1187) and the internal standard all-trans-retinoic acid-d₅ were used.

Identification of a WIN18,446 metabolite produced by ALDH1A2

To identify the metabolite of WIN18,446 produced by human ALDH1A2, WIN18,446 (5 μM) (Cayman Chemical cat no. 14018) was incubated with 10 μM purified recombinant ALDH1A2. The incubations were conducted in 150 mM KCl, 50 mM Hepes-NaOH, pH 8.0, at 37°C for 20 mins in the presence or absence of 2 mM cofactor NAD⁺. The total volume of the incubation was 200 μL. The reaction was initiated by adding WIN18,446 and stopped by adding 200 μL acetonitrile with 0.1% formic acid. The samples were centrifuged at 16,000 g at 4°C for 5 mins. The supernatant was used for metabolite identification (metID) using liquid chromatography-tandem mass spectrometry (LC-MS/MS). Negative control samples containing only WIN18,446 and no recombinant enzymes were also prepared. All samples were analyzed on a Thermo Fisher Scientific Orbitrap Fusion Tribrid MS equipped with a heated electrospray ionization (HESI) source and coupled to an Agilent 1290 Infinity II LC system. The LC separation was achieved with a Phenomenex Kinetex XB-C18 column (2.6 mm; 2.1 × 50mm; 100Å) with a flow rate of 0.4 mL/min. The column temperature was set at 40°C and the injection volume was 10 μL. The mobile phases A and B were water with 0.1% formic acid and acetonitrile with 0.1% formic acid, respectively. The LC gradient started at 5% B for 1 min, increased to 95% B over the next 5.5 mins, was then held at 95% B for 1 min before decreasing back to 5% B over 30 s, and re-equilibrating at 5% B for another 1 min.

Full-scan MS spectra were acquired in the Orbitrap mass analyzer at a resolution of 60,000 over an m/z range of 120–1,200. Source parameters were as follows: spray voltage 3.5 kV (positive), ion transfer tube temperature 350°C, vaporizer temperature 400°C, sheath gas 60, auxiliary gas 15, and sweep gas 2 (arbitrary units). Data-dependent MS/MS was performed using a Top-N method selecting precursors above a minimum intensity threshold of 2e⁵. Precursor ions were isolated in the quadrupole with an isolation window of 1.0 m/z and fragmented using higher-energy collisional dissociation (HCD) at stepped HCD at 35/45/55%. MS/MS spectra were acquired in Orbitrap at a resolution of 15,000. Dynamic exclusion was enabled with a repeat count of 1 and an exclusion duration of 2 s to maximize coverage of lower-abundance features. The MS data were analyzed using MassMetaSite 4.0, Xcalibur and Mass Frontier 8.1 to identify the metabolite(s) of WIN18,446 generated by ALDH1A2.

Characterization of ALDH1A2 catalyzed M-54 formation and kinetics

All incubations were conducted in reaction buffer consisting of 20 mM Hepes KOH buffer pH 8.0 with 150 mM KCl at 37°C unless otherwise stated. The formation of the aldehyde metabolite (M-54) from WIN18,446 (2 μM) [Acros Organics (Geel, Belgium)] in the presence and absence of NAD⁺ (2 mM) was first confirmed. Triplicate 100 μL reactions were initiated by adding 4 μL of ALDH1A2 (final concentration 0.5 μM) or buffer (negative control) and the mixtures incubated at 37°C for 15 mins. Reactions were terminated by adding 100 μL ice-cold acetonitrile. The samples were centrifuged at 15,000 g at 4°C for 10 mins and supernatants were transferred into MS vials for LC-MS/MS analysis.

The M-54 metabolite formation was measured using a Sciex QTRAP 6500 MS coupled to an Agilent Infinity II HPLC equipped with a Phenomenex Kinetex XB-C18 column (2.6 mm; 2.1 × 50mm; 100Å). The column temperature was set at 40°C and the injection volume was 5 μL. The LC mobile phase flow was 0.35 mL/min with gradient elution from 10% B for 1 min, increased to 100% B over the next 5 mins, kept at 100% B for 2 mins before decreasing back to 10% B over 30 s, and re-equilibrating at 10% B for another 2 mins. The metabolite was detected in positive ion APCI mode and the MRM transitions monitored included m/z 311>238, 311>128, and 311>69. Declustering potential (DP) was 40 eV and collision energy (CE) was 20 eV. The MS peak area of the m/z 311>238 transition was used to quantify formation of M-54 by ALDH1A2.

To characterize the M-54 formation kinetics, WIN18,446 (eight concentrations ranging from 50 nM to 3 μM in DMSO) was preincubated for 4 mins in the presence and absence of 2 mM NAD⁺. The reaction was then initiated by adding 20 μL of 1.8 μM ALDH1A2 (final concentration 180 nM) to a total reaction volume of 200 μL. The reaction was stopped after 2 mins with 200 μL ice-cold acetonitrile. Samples were centrifuged at 20,000 g at 4°C for 10 mins and the supernatant was analyzed by LC-MS/MS. Experiments were performed in triplicate and repeated on 3 different days.

The Michaelis-Menten equation was fit to the data using GraphPad Prism 10.6.1 (GraphPad software, LaJolla, CA). The model fits between +NAD⁺ and −NAD⁺ groups were compared using F test in GraphPad Prism.

Kinetics of ALDH1A2 inhibition by WIN18,446

To determine the kinetics of ALDH1A2 inactivation by WIN18,446, seven concentrations of WIN18,446 (ranging from 0.25 μM to 5 μM, plus vehicle control) were preincubated at 37°C in the presence and absence of 2 mM NAD⁺. Reactions were initiated by adding 20 μL of 1.8 μM ALDH1A2 (final concentration 180 nM) in a total reaction volume of 200 μL. Experiments were performed in triplicate in 96-wells plates and were repeated on 4 different days. At different timepoints (30 secs, 2 mins, and 5 mins for the inhibition in presence of NAD⁺ and 1 min, 5 mins, and 10 mins for the inhibition in the absence of NAD⁺), 10 μL aliquots of the inhibition reactions were transferred into 100 μL reaction buffer containing 1 μM all-trans-retinaldehyde and 2 mM NAD⁺. These reactions were incubated for 2 mins at 37°C to measure retinoic acid formation velocity. The reactions were terminated by adding 100 μL of pre-chilled acetonitrile with 200 nM all-trans-retinoic acid-d₅ as an internal standard. The samples were centrifuged at 15,000 g for 10 mins at 4°C, and supernatant was transferred to LC-MS/MS vials. Samples were analyzed by LC-MS/MS for retinoic acid concentrations as previously described (S. L. Arnold et al., 2015). Percent inhibition was quantified by comparing the retinoic acid formation velocity in the presence of WIN18,446 with or without NAD⁺ to that in the absence of WIN18,446.

Incubation method with purified ALDH1A2 for adduct identification

To identify WIN18,446 exposure related protein adducts on ALDH1A2, 20 μM WIN18,446 or vehicle control (DMSO), was preincubated in 20 mM Hepes KOH, 150 mM KCl, pH 8.0 at 37°C with or without 2 mM NAD⁺. Triplicate reactions for each condition were initiated by adding ALDH1A2 (final concentration 3 μM) and the samples were incubated at 37°C for 45 mins. To measure ALDH1A2 activity at the end of the incubation, a 2 μL aliquot was collected and added into 200 μL buffer containing 1 μM retinaldehyde and 2 mM NAD⁺. The retinoic acid formation reaction was allowed to proceed for 2 mins and retinoic acid formation was measured as described above. To detect protein adducts, the remaining sample was flash-frozen in liquid N₂ and stored at −80°C until sample preparation for LC-MS/MS analysis.

Incubation method with HLM for adduct identification

To identify proteins adducted by WIN18,446 in human liver samples, human liver S9 fractions (1 mg/mL) containing cytosolic and microsomal proteins were incubated with 1 mM WIN18,446 or vehicle (DMSO) at 37°C for 1h in 100 mM potassium phosphate pH7.4. The concentration of WIN18,446 in these adduct identification experiments was supraphysiological and was selected to avoid depletion of WIN18,446 given the high total protein concentration of the human liver S9 incubations. Excess unbound WIN18,446 was removed by desalting before proteomics as described below. For kinetics studies (see kinetics methods section above) physiological WIN18,446 concentrations were used. The S9 fractions were prepared from ten individual liver donors obtained from the University of Washington human liver bank (Lin et al., 2002; Paine et al., 1997). At the end of the incubation, samples were flash-frozen in liquid N₂ and stored at −80°C until analysis.

Proteomics

Sample Preparation

For purified ALDH1A2, samples were thawed and unbound WIN18,446 was removed by buffer exchange into ice cold 50 mM Tris pH 8.5 using protein desalting spin columns (ThermoFisher cat no. 89849) according to manufacturer’s instructions. Each digestion consisted of 50 μL buffer exchanged protein containing 8.5 μg protein to which 128 ng of yeast enolase (Sigma, cat no. E6126–500UN) was added as a process control (Tsantilas et al., 2024). Samples were brought to 10 mM TCEP and 0.02% n-dodecyl β-D-maltoside (DDM) (Sigma cat no. D4641) vortexed and heated at 95°C for 5 mins and then transferred to an Eppendorf Thermomixer with shaking (1000 rpm) at 37°C for 1 hour. Samples were allowed to cool to room temperature and brought up to 10 mM iodoacetamide. Alkylation was performed in the dark for 30 mins. DTT was added (10 mM final concentration) prior to digestion at a 1:15 (enzyme:substrate ratio) using trypsin (Promega Platinum, cat no. VA9000) for 12 hours at 37°C in an Eppendorf Thermomixer with shaking (1000 rpm). After digestion, samples were acidified with 250 mM HCl (final concentration) and supplemented with 15 fmol/μg final concentration Pierce Peptide Retention Time Calibration Mixture (PRTC) (Thermo Scientific, cat no. PI-88321) dissolved in 0.1% trifluoroacetic acid as a system suitability control (Tsantilas et al., 2024). The final protein concentration in the digested samples was 0.13 μg/μL. Samples were centrifuges at maximum speed in a benchtop microfuge for 10 mins and supernatant transferred to autosampler vials and stored at −80°C until use.

For human liver S9 fractions, unbound WIN18,446 was removed by buffer exchange as described above. Each digestion consisted of 30 μL buffer exchanged protein (30 μg total protein) plus 10 mM final concentration of TCEP, 480 ng enolase and 0.02% DDM (final concentration) in 63.2 μL volume. Samples were heated, reduced and alkylated, digested, quenched and had PRTCs spiked in as described above. The final digested protein concentration was 0.4 μg/μL and contained 15 fmol of PRTCs per μg of digested protein.

Mass Spectrometry

For purified ALDH1A2, 10 μL of digested sample, corresponding to 1.3 μg total protein, was loaded using a Thermo Vanquish Neo UHPLC System in “trap and elute” mode. The trap was a Thermo Scientific PepMap Neo Trap Cartridge (cat no. 174500) and the column was a Bruker PepSep C18 15 cm × 150 μm, 1.9 μm column (cat no. 1893471) connected to a 5 cm × 20 μm ID Sharp Singularity Fossil Ion Tech tapered tip. The column was enclosed in a custom constructed microspray source heated to 45°C.

Peptides were eluted from the column at 0.8 μL/min using the following acetonitrile gradient: (1) 0–0.7 mins; 4–6% B; flow 1.3 μL/min; (2) 0.7–1 min; 6% B; flow 1.3 μL/min; (3) 1–86.7 mins; 6–50% B; flow 0.8 μL/min; (4) 86.7–87.2 mins; 50–55% B; flow 1.3 μL/min; (5) 87.2–87.5 mins; 55–99% B; flow 1.3 μL/min; (6) 87.5–88 mins; 99% B; flow 1.3 μL/min; (7) 88–90 mins; 99% B; flow 1.3 μL/min. Buffer A was 0.1% formic acid in water. Buffer B was 0.1% formic acid, 80% acetonitrile and 20% water.

Mass spectrometry data were collected using a Thermo Fisher Scientific Orbitrap Eclipse Tribrid Mass Spectrometer. All samples were analyzed by both data-dependent acquisition (DDA) and by data-independent acquisition (DIA) based mass spectrometry methods. For adduct discovery experiments, DDA was performed at a MS and MS/MS resolution of 60,000 and 15,000, respectively. Cycle time was 3 seconds, MS scan range was 400–1,200 m/z. The automatic gain control targets for MS and MS/MS were set to a normalized value of 200% and 100%, and the maximum injection times were 118 ms and 22 ms, respectively. MS/MS spectra were acquired using an isolation width of 2.4 m/z and a normalized collision energy of 27. Precursors of +2 to +5 charge were included for MS/MS acquisitions and undefined charge states were excluded. Dynamic exclusion was set to 2 s and all spectra were collected in centroid mode.

For verification of treatment specific signals, spectra were acquired using DIA. The resolution for MS and MS/MS was 30,000. Cycle time was 3 seconds between MS scans. MS scan range was 395–1,205 m/z. The automatic gain control targets for MS and MS/MS were set to a normalized value of 100% and 800%, respectively, and the maximum injection time was set to 54 ms for both MS and MS/MS scans. MS/MS spectra were acquired using overlapping isolation windows (Amodei et al., 2019; Egertson et al., 2013) of 12 m/z with 67 MS/MS scans per cycle and a normalized collision energy of 27 assuming charge state 3. Cycle 1 spanned 406.4347 through 1198.7948 m/z in 12 m/z isolation windows. Cycle 2 spanned 400.4319 through 1204.7975 m/z in 12 m/z windows. All spectra were collected in centroid mode.

For human liver S9 fractions, 3 μL of sample, corresponding to 1.2 μg total protein, was loaded as described above for purified ALDH1A2, except the column used was a Bruker PepSep C18 8 cm × 150 μm, 1.5 μm column (cat no. 1893470). Peptides were eluted from the column at 1 μL/min using the following acetonitrile gradient: (1) 0–0.7 mins; 3–6% B; flow 1.3 μL/min; (2) 0.7–1 mins; 6–6.5% B; flow 1.3 μL/min; (3) 1–33 mins; 6.5–50% B; flow 1 μL/min; (4) 33–33.5 mins; 50% B; flow 1 μL/min; (5) 33.5–34 mins; 50–99% B; flow 1.3 μL/min; (6) 34–35 mins; 99% B; flow 1.3 μL/min. Buffer A and B were as described above.

MS data were collected using a Thermo Fisher Scientific Astral Zoom. Samples were analyzed by both DDA and DIA. DDA was performed at a resolution of 240,000 for MS scans in the orbitrap. MS/MS scans were performed in the Astral analyzer. Top-N mode was used with 100 scans between MS scans. MS scan range was 400–1,200 m/z. The automatic gain control target for MS set to a normalized value of 200% and for MS/MS was set to standard. Maximum injection times were 118 ms and 10 ms, respectively. MS/MS spectra were acquired using an isolation width of m/z 2 and a normalized collision energy of 27. Precursors of +2 to +5 charge were included for MS/MS acquisitions and undefined charge states were excluded. Dynamic exclusion was set to 4 s and all spectra were collected in centroid mode. DIA was performed at a resolution of 240,000 for MS scans in the orbitrap. MS/MS scans were performed in the Astral analyzer. MS scan range was 375–985 m/z. The automatic gain control target for MS was set to standard and to a normalized value of 200% for MS/MS. Maximum injection time was set to 50 ms and 7 ms for MS and MS/MS, respectively. MS/MS spectra were acquired using non-overlapping isolation windows of 5 m/z with 160 MS/MS scans per cycle and a normalized collision energy of 27 assuming charge state 2. All spectra were collected in centroid mode.

Data Processing

Open-mass searches of DDA data were performed using Magnum as previously described (Riffle et al., 2022). Any modification mass between 60 and 400 Da was allowed on cysteine, and post-processing for FDR control was done using Percolator (Kall et al., 2007). The entire open-mass search pipeline was performed using a fully containerized Nextflow workflow available at https://nf-openmod-dda.readthedocs.io. Closed-mass searches of DDA data were done as previously described (Zelter et al., 2024) using Comet (Eng et al., 2015) with post-processing for FDR control by PeptideProphet (Shteynberg et al., 2011) followed by rescoring of adduct localization using PTMProphet (Shteynberg et al., 2019). The comet isotope_error parameter was set to 3, allowing Comet to consider possible misassignments of the precursor mass due to selecting a higher isotope peak (e.g. the atypically intense M+2 caused by ³⁷Cl) thereby accommodating potential monoisotopic mass assignment errors. The complete closed-mass search workflow (Comet/TPP) is available at https://nf-ms-dda-tpp.readthedocs.io/.

All search parameters, pipeline configuration and FASTA search database files along with complete raw data and interactive search results are available on Limelight (Riffle et al., 2025): https://limelight.yeastrc.org/limelight/p/win18446-adducts. The Limelight web application and associated documentation is available at https://limelight-ms.org/.

Peptide Quantification

Spectral libraries were constructed from all peptides identified in closed Comet searches of the DDA data, including both adducted and unadducted forms. Libraries were generated using Limelight’s Download Blib Spectral Library function. These libraries were used to pick transitions for quantification of DIA data using Skyline (Pino et al., 2017). Full Skyline sessions are available on Panorama at https://panoramaweb.org/win18446-adducts.url.

Results and Discussion

Identification of an aldehyde metabolite of WIN18,446 produced by ALDH1A2

To test our hypothesis that an aldehyde metabolite of WIN18,446 is formed by ALDH1A2, WIN18,446 was incubated with purified recombinant ALDH1A2 and metID was conducted. The formation of a metabolite with a monoisotopic mass of 310 Da corresponding to a loss of 54 Da from WIN was detected in the incubations and named as M-54 (Figure 2). Based on the observed high-resolution mass and the peak abundance ratio of monoisotopic mass to isotopic mass (Table 1 and Figure 2B), the chemical formula of M-54 was proposed to be C₁₂H₂₀O₃N₂Cl₂. The MS/MS spectrum of m/z 311 indicated that M-54 is an aldehyde metabolite (Figure 2C) and the fragmentation supported our hypothesized structure (Figure 1C).

Figure 2.

Characterization of the aldehyde metabolite, M-54, of WIN18,446. (A) Chemical structure of the aldehyde metabolite M-54. (B) MS1 scan of M-54. (C) The MS/MS spectrum of M-54 with a proposed fragmentation pathway. (D) Extracted ion chromatograms of m/z 311.0929 (M-54) and m/z 365.0375 (WIN18,446). In panel D, the black solid line and dash line represent the selected ion chromatograms of ALDH1A2 plus WIN18,446 in the presence and absence of NAD⁺, respectively. The red line represents the negative control group which contained WIN18,446 but no ALDH1A2.

Table 1.

Metabolite profile of WIN18,446 in ALDH1A2 incubations

	Retention Time (min)	Theoretical m/z [M+H]+	Observed m/z [M+H]+	Chemical Formula	Mass error for monoisotopic mass (ppm)
WIN18,446	4.2	365.0357 367.0327 369.0300	365.0375 367.0346 369.0287	C12H20O2N2Cl4	4.93
M-54	3.1	311.0929 313.0900 315.0878	311.0948 313.0919 315.0893	C12H20O3N2Cl2	6.11

Note: z=1. The bolded m/z values were derived from the monoisotopic mass with ³⁵Cl while m/z values in italics were derived from the isotopic mass. Other theoretically possible WIN18,446 metabolites (Supplementary Figure 3) were evaluated by extracting their corresponding m/z values in Xcalibur. No signals consistent with the proposed metabolites were detected in the ALDH1A2 incubations with WIN18,446.

The proposed mechanism of M-54 formation (Figure 1) suggests that NAD⁺ cofactor is not needed for M-54 formation. We therefore tested whether M-54 formation could be detected in the absence of NAD⁺. M-54 formation was detected in the incubations with ALDH1A2 both in the absence and in the presence of NAD⁺ but not in the absence of ALDH1A2 (Figure 2D and Figure 3A). The formation of M-54 is therefore not NAD⁺-dependent but does require ALDH1A2. These results support the mechanism of M-54 formation proposed in Figure 1.

Figure 3:

Formation of the M-54 metabolite of WIN18,446 is dependent on ALDH1A2 and the kinetics are impacted by NAD⁺. (A) M-54 peak area was measured after 15 mins incubations containing 2 μM WIN18,446 with or without 0.5 μM ALDH1A2 in the presence of absence of 2 mM NAD⁺. (B) The kinetics of M-54 formation in presence (purple) and absence (teal) of 2 mM NAD⁺. Paired experiments, using a final concentration of 180 nM ALDH1A2, were performed in duplicate or triplicate with measurements on three separate days. The Michaelis-Menten equation was fitted to the data (solid lines).

Kinetics of the M-54 metabolite formation

The kinetics of M-54 formation were characterized in the presence and absence of NAD⁺. The formation of M-54 was WIN18,446-concentration dependent with or without cofactor (Figure 3B). The K_m of M-54 formation in the absence of NAD⁺ was significantly higher than in the presence of NAD⁺. In the presence of NAD⁺ the K_m for M-54 formation by ALDH1A2 was 0.35 ± 0.07 μM. In the absence of NAD⁺ no saturation of the Michaelis Menten curve was observed and the K_m was estimated as >3 μM (Figure 3B). The model that fit the M-54 formation in the presence of NAD⁺ was significantly different than that in the absence of NAD⁺ based on the F-test. Due to the absence of a reference standard for M-54, the k_cat for M-54 formation could not be determined. Although no saturation of product formation was observed in the absence of NAD⁺, the data suggests the k_cat for M-54 formation is not considerably lower in the presence of NAD⁺ than in its absence. These findings are consistent with the reaction mechanism proposed. Overall, these data suggest that presence of NAD⁺ changes the affinity or binding of WIN18,446 to ALDH1A2.

Kinetics of ALDH1A2 inactivation by WIN18,446

As M-54 was also formed in the absence of NAD⁺, we next tested whether inactivation of ALDH1A2 requires cofactor (Figure 4A). When 3 μM ALDH1A2 was incubated with saturating concentrations of WIN18,446 (20 μM) for 45 mins, no difference was observed in the percent activity remaining between the incubations done in the presence versus the absence of NAD⁺. In contrast, in the absence of WIN18,446, no inactivation of ALDH1A2 was observed (Figure 4A). Taken together these data show that like M-54 formation, the inactivation of ALDH1A2 by WIN18,446 does not require NAD⁺.

Figure 4:

The effect of WIN18,446 on ALDH1A2 activity as measured by retinoic acid formation. (A) ALDH1A2 (3 μM) mediated retinoic acid formation is inhibited by WIN18,446 (20 μM) in the presence and absence of NAD⁺ after 45 mins. (B) The rate of inactivation of ALDH1A2 (180 nM) was determined with increasing concentrations of WIN18,446 in the presence (red) and absence (blue) of NAD⁺. Lines show the fit of the inactivation kinetics model to the data from four different days. Panels C and D show the loss of ALDH1A2 activity as a function of time at different WIN18,446 concentrations in the presence (C) and absence (D) of NAD⁺. Panels C and D show representative results from a single experiment in (B). Kinetic experiments were performed as triplicates on 4 different days.

To characterize the impact of NAD⁺ on the inactivation kinetics of ALDH1A2, we determined the K_I and k_inact of WIN18,446 for ALDH1A2 in the presence and absence of NAD⁺ using retinoic acid formation as a marker of ALDH1A2 activity. The inactivation rate of ALDH1A2 by WIN18,446 was significantly greater in the presence of NAD⁺ (k_inact = 1.5 ± 0.2 min⁻¹) than in the absence of NAD⁺ (Figure 4B–D). Like the observations of M-54 formation, no saturation of inactivation rate was observed in the absence of NAD⁺ and the apparent K_I of WIN18,446 was higher in the absence versus in the presence of NAD⁺ (K_I = 0.9 ± 0.4 μM). The impact of NAD⁺ on the K_I of WIN18,446 was similar to the effect of NAD⁺ on the K_m of M-54 formation. This is consistent with an effect of NAD⁺ on WIN18,446 binding affinity to ALDH1A2.

The oxidation reactions of aldehydes to their corresponding carboxylic acids, catalyzed by ALDH1A enzymes, are known to be cofactor dependent as NAD⁺ is required for the hydride abstraction step (Heider & Hege, 2025). However, for the metabolism of WIN18,446, the data presented here supports our hypothesis that NAD⁺ is not a required cofactor for the formation of M-54 and is not essential for time dependent inactivation of the enzyme. Together our findings showing the impact of NAD⁺ on the M-54 formation and on the inactivation kinetics of ALDH1A2 suggest that NAD⁺ functions also as an allosteric modulator of ALDH1A2 changing the binding affinity of WIN18,446 to ALDH1A2.

An LC-MS/MS-based approach to identify unknown drug-protein adducts

The inactivation of ALDH1A2 by WIN18,446 was previously proposed to be consistent with a protein adduct on the catalytic cysteine, C320 (Chen et al., 2017). The results presented above, however, suggest that the previously observed adduct may be a transient intermediate species that was stabilized under crystallographic conditions. To allow the identification and characterization of protein adducts resulting from exposure of ALDH1A2 to WIN18,446, regardless of their mass or chemical composition, we selected an LC-MS/MS-based adductomics approach that uses both open- and closed-mass searching of data dependent acquisition (DDA) data.

Open-mass searching allows observed peptide masses to differ from identified peptide masses. Mass differences are returned as modifications of the identified peptides. This search process enables detection of undefined (unknown) modification masses such as those resulting from reactions with unknown or uncharacterized metabolites. However, search results from open-mass searches also contain many modification masses not related to drug exposure (Riffle et al., 2022). To allow identification of treatment specific modifications, data were collected from both WIN18,446 treated and untreated samples and compared to find masses enriched specifically in the WIN18,446-treated samples.

After identifying the accurate mass of the WIN18,446 adduct by open-mass searching, closed-mass searches were performed with Comet (Eng et al., 2015) specifying the adduct mass discovered in the open-mass search. Closed searching provides a more targeted and conservative analysis because only explicitly specified modifications are considered (Zelter et al., 2024).

In addition to DDA, which is useful for adduct discovery via open- and closed-mass searching, data independent acquisition (DIA) experiments were performed on the same samples in parallel. DIA is an unbiased acquisition method which enables untargeted and quantitative assessment of extracted precursor and product ion chromatograms (XICs) for all detected peptides in the sample (Egertson et al., 2015; Gillet et al., 2012; Venable et al., 2004). These data were used to further confirm treatment specific signals for each of the peptides identified as adducted using the DDA-based search methods and evaluate elution profiles of the identified adducted and unadducted peptides.

Identification of WIN-protein adducts in purified ALDH1A2

Proteomics data were acquired from purified ALDH1A2 incubated with or without WIN18,446. Near-complete loss of ALDH1A2 enzymatic activity was observed for WIN18,446 treated samples regardless of the presence of cofactor (Figure 5A). LC-MS/MS was performed on these samples, and open-mass searching was done using Magnum (Riffle et al., 2022). A two-tailed test of proportions done in Limelight (Riffle et al., 2025) comparing untreated ALDH1A2 with WIN18,446 treated ALDH1A2 in the presence of NAD⁺ highlighted 292 Da and 294 Da as the top 2 masses enriched in the WIN18,446 treated samples (Figure 5B and C). For simplicity, the results below focus on samples generated in the presence of NAD⁺, however the results from samples treated with WIN18,446 in the absence of cofactor were near-identical (Supplemental Figure S4) confirming the adduct formation in both the presence and absence of NAD⁺.

Figure 5:

Open-search analysis reveals 292 Da and 294 Da modifications as the dominant masses enriched in WIN18,446 treated ALDH1A2 samples in the presence of NAD⁺. (A) ALDH1A2 (3 μM) mediated retinoic acid formation is inhibited by WIN18,446 (20 μM) in the presence and absence of NAD⁺ after 45 mins. (B) The top 2 ranked masses enriched in WIN18,446 +NAD⁺ treated samples based on Z-score. (C) A volcano plot of Z-score versus log₂-fold change of peptide spectrum match (PSM) ratios highlighting 292 Da and 294 Da as being the most significantly enriched masses in the WIN18,446 +NAD⁺ treated samples. Comparisons were performed between data from triplicate incubations of ALDH1A2 treated for 45 minutes with WIN18,446 plus cofactor versus vehicle plus cofactor-only controls. Results were filtered to have a Percolator calculated PSM level q ≤ 0.01 and for localized modifications only. Exact modification masses were binned into 1 Da bins by Limelight prior to analysis. A complete list of the Z-scores and p-values for all masses can be found in Supplementary File 2. (D) A histogram of PSM counts for treated samples within the 292 Da mass bin shows a distribution with an apex at 292.07 Da. An interactive view of these data is available on Limelight here: https://limelight.yeastrc.org/limelight/go/GM834MTK83. Near-identical results were observed for ALDH1A2 samples treated with WIN18,446 in the absence of NAD⁺ (Supplementary Figure S4).

In total, 295 peptide spectrum matches (PSMs) containing a localized modification of 292 Da were observed in the WIN18,446-treated samples versus 15 PSMs in the untreated samples (Figure 5 and Table 2). In the treated samples, 99.7% of the PSMs containing mass 292 Da were in ALDH1A2 and 97.3% were to the active site peptide (SPNIIFADADLDYAVEQAHQGVFFNQGQCCTAGSR). Of these, 72% had the modification localized to cysteine 319 (C319) (Table 2).

Table 2:

The number of peptide spectrum matches (PSMs) observed in WIN18,446 treated versus untreated samples of purified ALDH1A2 with an open modification mass of 292 or 294 Da after open searching with Magnum. Data shown are filtered to have a Percolator calculated PSM level q ≤ 0.01 and for localized modifications only. An interactive view of these data can be found on Limelight here: https://limelight.yeastrc.org/limelight/go/GM834MTK83

Treatment (obs. mod mass Da)	All PSMs with mod	PSMs with mod in ALDH1A2	PSMs with mod in other proteins	PSMs with mod in ALDH1A2 C319	PSMs with mod in ALDH1A2 C320
untreated (292)	15	14	1	14	0
untreated (294)	7	6	1	4	1
WIN treated (292)	295	294	1	207	80
WIN treated (294)	349	346	3	227	63

In the same WIN18,446 treated samples 349 PSMs containing a localized modification of 294 Da were also observed, versus 7 PSMs in the untreated samples. Of the PSMs containing mass 294 Da, 99% were in ALDH1A2 and 83% were to the active site peptide. Of these, 78% were localized to C319 (Table 2). An annotated spectrum of a C319 WIN18,446-modified ALDH1A2 active site peptide is shown in Figure 6. This peptide and modification localization represents almost 80% of PSMs identified containing a 292 or 294 Da modification resulting from exposure to WIN18,446.

Figure 6:

Representative annotated spectrum of peptide SPNIIFADADLDYAVEQAHQGVFFNQGQC(292)CTAGSR, with a 292 Da modification on C319 and a charge state of +4. The top table (green) shows the peptide and properties of the representative PSM. Below, left is an annotated MS2 spectrum. On the right are zoomed regions representing unmodified (y5+ and y6+) and WIN18,446-modified (y7+, y8+ and y9+) fragment ions. Note that, as expected given the correct adduct localization, only WIN-containing fragment ions have characteristic chlorine signatures. An interactive view of this spectrum is available here: https://limelight.yeastrc.org/limelight/d/pg/spectrum-viewer/ps/9333/psm/759595863?openmod-position=29

Spectra of PSMs matching the ALDH1A2 active site peptide containing 292 Da and 294 Da modifications were inspected manually. This inspection showed that spectra annotated as having a 294 Da modification actually represented a 292 Da modification. The +2 Da discrepancy results from monoisotopic mass misassignment rather than a distinct chemical species (Supplementary Note 1). The explanation for the prevalence of 294 Da assignments is that chlorine, a component of WIN18,446, contains two major stable isotopes, ³⁵Cl (~75%) and ³⁷Cl (~25%), separated by ~2 Da. Protein adducts containing chlorine therefore exhibit an atypically intense M+2 isotope peak. This can confound automated monoisotopic peak selection causing incorrect assignment of the ³⁷Cl isotope peak as the monoisotopic peak. This leads to an apparent +2 Da shift. Our manual analysis confirmed this as the explanation for the 294 Da mass assignments presented in Table 2 and in Supplementary Note 1.

For the analysis above, all masses were binned into 1 Da mass bins. A histogram of the exact masses in the 292 Da mass bin showed a gaussian distribution around 292.07 Da (Figure 5D, Supplemental Figure S4). Following identification of the WIN18,446 adduct mass by open-mass searching, closed-mass searches were performed using this defined mass to provide a more targeted and conservative analysis.

The closed searches resulted in 456 PSMs to the ALDH1A2 active site peptide containing the 292 Da modification in the WIN18,446 plus cofactor treated samples and 22 PSMs in the untreated samples. 89% of the 292 Da adducts were to C319 with just 11% localized to C320 (Table 3). Notably, only 5% of peptides containing a WIN-adducted C319 had an alkylated C320 while 100% of peptides containing a WIN-adducted C320 had an alkylated C319. Samples were alkylated prior to digestion under conditions expected to result in the majority of cysteines being alkylated. Majority alkylation was confirmed in vehicle only controls where 93% of observed ALDH1A2 active site peptides had one or more alkylations. The lack of alkylation in the WIN18,446 adducted peptides was surprising, suggesting that the WIN18,446 adduct on C319 sterically hinders the cysteine adjacent to it.

Table 3:

The ALDH1A2 active site peptides, modification locations and number of PSMs observed in WIN18,446 treated versus untreated samples of purified ALDH1A2 containing a defined modification of 292.0745 Da after closed searching with Comet. Data shown are for the active site peptide only and represent Comet searches allowing a variable modification of 292.0745 Da postprocessed with TPP and filtered at 1% FDR. An interactive view of these data can be found on Limelight here: https://limelight.yeastrc.org/limelight/go/8G3M8JDDWV

Peptide Sequence	Untreated			WIN treated			WIN treated
	no WIN +NAD+ 45 min			+WIN +NAD+ 45 min			+WIN no NAD+ 45 min
	PSMs Count (rep 1)	PSM Count (rep 2)	PSM Count (rep 3)	PSM Count (rep 1)	PSM Count (rep 2)	PSM Count (rep 3)	PSM Count (rep 1)	PSM Count (rep 2)	PSM Count (rep 3)
SPNIIFADADLDYAVEQAHQGVFFNQGQC[292.07]CTAGSR	9	6	5	126	125	135	120	133	137
SPNIIFADADLDYAVEQAHQGVFFNQGQC[57.02]C[292.07]TAGSR	0	1	1	19	15	18	23	33	27
SPNIIFADADLDYAVEQAHQGVFFNQGQC[292.07]C[57.02]TAGSR	0	0	0	3	8	7	5	2	0

The fact that we observed the adduct on C319 rather than C320 was also unexpected as the catalytic cysteine in ALDH1A2 is C320, and the WIN18,446 derivative adduct in the crystal structure was observed on C320. It is possible that differences in the local microenvironments of C319 and C320, such as differences in thiol pKa and nucleophilicity, bias the reaction outcome. Reaction of WIN18,446 at C320 may favor M-54 formation, whereas reaction at C319 may result in formation of a more stable adduct.

Extracted ion chromatograms (XICs) were produced from the DIA data for the peptides identified as modified by WIN18,446 in the DDA data shown in Table 3. Peak areas were quantified and signal was evaluated to confirm presence of the adduct in treated samples only. All 3 versions of the ALDH1A2 active site peptide identified as adducted by WIN18,446 had clear treatment specific signal (Figure 7, bottom left panels). In PSMs representing these peptides, fragment ions from regions of the peptide that did not contain the WIN18,446 modification showed no chlorine isotopic signature. In contrast, fragment ions from regions of the peptide containing the WIN18,446 modification exhibited the characteristic chlorine isotopic pattern, consistent with the presence of chlorine within the adduct (Figure 7, right panels). Additionally, the retention times of the WIN18,446 modified peptide were about 10 minutes later than their unadducted counterparts (Figure 7, bottom left panels compared to middle left panel).

Figure 7:

Quantification of peptides from DIA data of untreated and WIN18,446 treated purified ALDH1A2 samples. Representative extracted ion chromatograms (XICs) and resulting peak areas are shown for fragment ion transitions from WIN18,446 modified peptides and their unmodified counterparts. Yellow (top left) panel: Control peptides. Yeast enolase (ENO1) was spiked into all samples as a process control. Green (middle left) panel: Quantification of unadducted peptides corresponding to the WIN-adducted peptides shown in the purple (bottom left) panel. XICs are shown for treated and untreated samples. Peak areas are shown for all conditions tested. Horizontal green bars indicate untreated samples, and horizontal purple bars indicate WIN18,446 treated samples. A complete Skyline file for all peptides identified in all samples can be found on Panorama here: https://panoramaweb.org/win18446-adducts.url. Right hand panels show annotated spectra of representative PSMs from the DDA data corresponding to each of the 3 WIN18,446 modified peptides. Inset into each panel are zoomed regions showing fragment ions containing the modification, resulting in a chlorine signature, and before the modification, showing no chlorine signature. Live views of these spectra are available on Limelight for the top (https://limelight.yeastrc.org/limelight/d/pg/spectrum-viewer/ps/8942/psm/743565299), middle (https://limelight.yeastrc.org/limelight/d/pg/spectrum-viewer/ps/8942/psm/743555395) and bottom (https://limelight.yeastrc.org/limelight/d/pg/spectrum-viewer/ps/8924/psm/742941261) annotated spectra.

We also searched for the crystallized WIN18,446 derivative adduct previously observed by Chen et al., in 2017. The adduct they observed has the chemical formular C₁₂H₂₀N₂O₂Cl₃, which would result in a modification mass of 328.0512 Da after covalent adduct formation. Our open-mass searches found no evidence of a treatment-specific 328 Da modification. Comparing treated and untreated samples, this modification mass was assigned a Z-score of −2.48 and a p-value of 1, indicating no significant difference between treated and untreated samples. This was true regardless of the presence or absence of cofactor (Supplementary File 2). Closed-mass searches were also performed specifying 328.0512 Da as the defined modification mass, but no PSMs containing this mass were found in either the treated or untreated samples (full results available on Limelight here: https://limelight.yeastrc.org/limelight/go/NLKGX183GQ).

Possible mechanisms of WIN18,446 adduct formation

Based on the observed adduct modification mass of 292.07 Da, we suggest two possible mechanisms of adduct formation (Figure 8). The fact that the adduct was observed on both C319 and C320 suggests that both C319 and C320 react with WIN18,446. According to our proposed mechanisms the same intermediate that was observed in the crystal structure, and results M-54 formation, can also lead to the formation of a covalent protein adduct, but with a different WIN-derived structure. We hypothesize that when the crystal structure intermediate is formed with the catalytic cysteine C320 (Figure 1B or structure 1 in Figure 1C) the formation of the M-54 metabolite is favored. In contrast, if the crystal structure intermediate forms with C319, the creation of a covalent protein adduct of mass of 292.07 Da is favored. This hypothesis is supported by the observation that more 292 Da modifications were localized to C319 than to C320 (Table 2) and that a larger signal for C319 localization was observed using DIA (Figure 7).

Figure 8:

Possible mechanisms of WIN18,446-induced protein adduct formation. Cysteine 319 (Cys319) of ALDH1A2 is shown in blue, while the amide group of Cys320 is shown in magenta. Adducts 1 and 2 yield the same modification mass of 292 Da following reaction with the thiol (–SH) group of the modified cysteine. The proposed adduct has a chemical formula C₁₂H₁₉Cl₂N₂O₂, corresponding to a mass of 292.0745 Da after subtracting one hydrogen from the formula of the reactive metabolite (C₁₂H₁₈O₃N₂Cl₂). This is done to account for the fact that proteomics search algorithms assume an intact cysteine thiol, which is not the case after adduct formation.

Identification of proteins adducted by WIN18,446 in human liver S9 fractions

WIN18,446 inhibits ALDH1A1 in human liver (Zhong et al., 2021) and results in irreversible inhibition of ALDH2 (Haenisch et al., 2024). However, it is unknown whether WIN18,446 inhibits a broader complement of metabolic enzymes in the liver. Having identified the accurate modification mass of protein adducts resulting from exposure of purified ALDH1A2 to WIN18,446, we were then able to explore the overall proteome adducted by WIN18,446 in the human liver. To this end, human liver S9 fractions from 10 donors were exposed to WIN18,446 or to vehicle as a control. Samples were digested and analyzed by both DDA- and DIA-based LC-MS/MS and Comet searches were performed against the entire human proteome additionally searching for the WIN18,446 modification mass determined by open searching above.

Analysis of all 10 treated and untreated samples identified 1,356,354 PSMs representing 54,278 unique peptides and 17,163 protein groups across all samples at 5% FDR. The complete dataset is available on Limelight at: https://limelight.yeastrc.org/limelight/go/STNQQBJ8PF. Filtering for peptides containing a 292.07 Da modification identified by at least 2 PSMs resulted in 2 treatment specific peptides in WIN18,446 exposed samples (Table 4).

Table 4:

WIN18.446 modified peptides identified in human liver S9 fractions. Peptides, proteins, modification locations and total number of PSMs observed in all untreated versus all treated donors are shown. Searches were done using Comet with a defined modification of 292.0745 Da and postprocessed with TPP. Results are shown at 5% FDR and filtered to require a 292.0745 Da modification and at least 2 PSMs. An interactive view is available on Limelight here: https://limelight.yeastrc.org/limelight/go/D3ZK9734Z7

Peptide Sequence	Protein	PSM Count Untreated	PSM Count WIN18,446 treated
SPC[57.02]IVLADADLDNAVEFAHHGVFYHQGQC[292.07]CIAASR	ALDH1A1	0	18
SPNIIMSDADMDWAVEQAHFALFFNQGQC[292.07]CC[57.02]AGSR	ALDH2	0	6

As before, DIA data was acquired in parallel. XICs were produced and peak areas were quantified from the DIA data corresponding to the peptides identified as adducted by WIN18,446 in the DDA search results (Figure 9). Evaluation of the DIA data for these peptides confirmed treatment-specific signal in both cases. Additionally, chlorine isotope signatures were observed in modified fragment ions from annotated spectra of the identified WIN18,446-adducted peptides in the DDA data (Supplementary File 3).

Figure 9:

Quantification of peptides from DIA data of untreated and WIN18,446 treated human liver S9 fractions from 10 donors. Representative extracted ion chromatograms (XICs) and resulting peak areas are shown for fragment ion transitions from WIN18,446 modified peptides in ALDH1A1 and ALDH2 along with unmodified peptides from the same proteins. Yellow (top) panel: Control peptides. Yeast enolase (ENO1) was spiked into all samples as a process control. Green (middle) panel: Quantification of unadducted peptides from ALDH1A1 and ALDH2 showing detection of those proteins in all samples. WIN18,446-adducted peptides identified in these samples are shown in the purple (bottom) panel. Protein name, peptide sequence and precursor transition charge state are shown for each peptide. Example XICs are shown for treated and untreated samples. Peak areas are shown for all samples tested. Horizontal green bars indicate untreated samples, and horizontal purple bars indicate WIN18,446 treated samples. A complete Skyline file for all peptides identified in all samples can be found on Panorama here: https://panoramaweb.org/win18446-adducts.url.

The only two proteins adducted by WIN18,446 in the human liver samples were ALDH1A1 and ALDH2. This is remarkable as ALDH2 and ALDH1A1 are also inhibited by WIN18,446 in vitro and in humans and mice in vivo. The adducts detected were exclusively within the active site cysteine motif in these two enzymes, supporting the role of the discovered adduct in inactivation of ALDH enzymes. In both cases the cysteine preceding the catalytic cysteine within the active site cysteine motif was the one modified by WIN18,446. In ALDH1A1 C302 was modified within the C302/C303* motif. In ALDH2 C318 was modified within the C318/C319*/C320 motif. In each case the * represents the catalytic cysteine. These residues are equivalent to C319 in ALDH1A2, which is also the cysteine preceding the catalytic cysteine within the dual cysteine active site motif of ALDH1A2 (C319/C320*) (Kim et al., 2024), and which was the cysteine we found to be modified by WIN18,446 in purified ALDH1A2. In the liver samples, no ALDH1A2 adducts were found. Although we were able to detect ALDH1A2 in all livers tested, only five ALDH1A2 peptides were detected in total (~9% total sequence coverage of the protein) and these peptides did not include the active site region. Low sequence coverage of ALDH1A2 is not surprising as previous studies have been unable to detect expression in most livers (Zhong et al. 2021). Based on these findings the lack of detection of ALDH1A2 adducts is likely due to the very low expression of ALDH1A2 in the human liver and lack of adequate sensitivity to detect the adducted peptide if it is present.

Taken together these data show the detection of WIN18,446 adducted ALDH2 and ALDH1A1 in human livers, and suggest that the formation of the adduct detected and identified with purified ALDH1A2 is not unique only to the purified enzyme. The adducted ALDH2 and ALDH1A1 peptides were detected in all 10 human livers in the DIA analysis although the abundance varied between donors. This result suggests future DIA based analyses could be used to evaluate interindividual variability in adduct abundance. These findings also support the use of LC-MS/MS based proteomics for identification of adducted peptides in complex samples such as the human liver. Such studies may aid in future evaluation of mechanisms and pathways of toxicity and identification of inactivation mechanisms of catalytic enzymes.

Conclusions

WIN18,446 is an irreversible inhibitor of ALDH1A family proteins and a useful compound to study retinoid signaling in vitro and in vivo. WIN18,446 is also a reversible male contraceptive in humans, however, when taken with alcohol it causes a “disulfiram reaction” due to the increase in acetaldehyde concentrations resulting from ALDH2 inhibition. Despite much interest in WIN18,446, the mechanism of the irreversible inhibition of ALDH family enzymes has not yet been determined.

A crystal structure of ALDH1A2 captured a WIN18,446 derivative covalently adducted to the active-site catalytic cysteine, which suggested a mechanism for irreversible inhibition. In the current work, we found that this derivative is unstable in solution and is converted to an aldehyde metabolite, which we designated M-54. We unequivocally showed that M-54 is formed as a metabolite of WIN18,446 both in the presence or absence of the cofactor NAD⁺. Our data also suggest that NAD⁺ acts as an allosteric modulator of WIN18,446 binding rather than as a cofactor for the M-54 formation. We determined the accurate mass of M-54, propose its chemical composition and describe the mechanism and kinetics of its formation. The formation of M-54 by ALDH1A2 indicates that M-54 may be an important in vivo metabolite of WIN18,446. Further studies of the potential pharmacological and toxicological activity of M-54 are warranted.

Our work utilized a comprehensive proteomics approach to identify and characterize protein adducts resulting from exposure of ALDH1A2 to WIN18,446 without making a priori assumptions of adduct masses or chemical composition. This workflow leveraged both open- and closed-mass searching of DDA data in combination with verification of treatment specific signal using the unbiased data acquisition properties of DIA. Our results showed that such exposure results primarily in a modification of 292.07 Da on cysteine 319 of ALDH1A2, the cysteine preceding the catalytic cysteine (C320) within the dual cysteine active site motif of ALDH1A2 (C319/C320). This modification and the proposed structure are consistent with the crystal structure intermediate forming in C319 followed by the formation of the stable adduct.

Finally, we performed a proteome-wide search to explore protein-WIN18,446 adducts formed in human liver samples and found adducts in the active site cysteine motifs of both ALDH1A1 and ALDH2. Equivalent to the ALDH1A2 C319 modification, the WIN18,446 modifications in ALDH1A1 and ALDH2 were localized to the cysteine preceding the catalytic cysteine within the active site cysteine motif. While we cannot exclude adduct formation on other peptides based on a lack of detections, these data do suggest that WIN18,446 preferentially adducts to cysteines within the active site motif of the ALDH1/2 family.

The work described here provides new insight into the metabolism of WIN18,446 by ALDH1A2 and its mechanism of inhibition. Future studies should evaluate the metabolism and inactivation of other ALDH enzymes by WIN18,446 to further evaluate the impact of the microenvironment around the cysteine on adduct formation and ALDH inactivation. This work also highlights a powerful and generally applicable proteomics workflow to identify and verify drug-protein adducts of unknown mass and to detect such adducts in complex human samples. All computational components of this workflow have been packaged and made freely available for use by the community.

Data Availability Statement

Raw DDA data, search configuration files and corresponding DDA search results can be downloaded or viewed interactively via Limelight at: https://limelight.yeastrc.org/limelight/p/win18446-adducts. All raw data, including DIA raw files along with Skyline quantification, were deposited to the ProteomeXchange Consortium via Panorama Public (Sharma et al., 2018) and are available under the data set identifier PXD073497 at: https://panoramaweb.org/win18446-adducts.url.