The product of the corresponding Michael addition reaction, i.e. propionamide (Figure 1), was used as ligand. The respective 3D structure was obtained from PubChem (Sunghwan et al., 2019) (CID 6578).

The 3D structures of the human proteins in Supplementary Table S1 were taken from the Protein Data Bank (Berman et al., 2000; Burley et al., 2021). When more than one structure was available, the one at the highest resolution was chosen (see Supplementary Table S1). Protein structures with missing residues were retrieved from the SWISS-MODEL repository (Bienert et al., 2017) or generated with SWISS-MODEL (Waterhouse et al., 2018), by selecting templates structures with the same sequence as the targets. When experimental structures of the human protein were not available, we generated homology models (see Sections 3.2.4–3.2.7). Target-template sequence alignments were obtained with either BLAST (Camacho et al., 2009) or HHblits (Steinegger et al., 2019), as implemented in the SWISS-MODEL webserver. Templates with the highest sequence identity and the highest resolution were selected and models were generated with SWISS-MODEL (Waterhouse et al., 2018) (see Supplementary Table S1). Protein structures were processed with MolProbity (Chen et al., 2010; Williams et al., 2018) to add missing hydrogen atoms, assign histidine protonation states and perform His/Gln/Asn flips, if recommended. The reactive cysteines were modeled as already deprotonated (Foloppe et al., 2001), as expected for the Michael addition reaction to take place (see Figure 1, step 2). Therefore, our computational protocol does not take into account the energetic cost of Cys deprotonation, i.e. ΔG = ln(10) × kT × (pK _a − pH). Moreover, we have assumed a default pH of 7, even though the protein targets in our dataset exhibit different optimal pH ranges (see Supplementary Table S3) and the Michael addition reaction is favored at basic pH (Lutolf et al., 2001; LoPachin et al., 2007; Nair et al., 2014). However, even if the Cys pK_a (calculated here with the H++ webserver at a default pH of 7) may predict population of the thiolate state smaller than the thiol one, reaction with acrylamide is expected to shift the acid-base equilibrium (step 2 in Figure 1) towards the deprotonated form.

Covalent docking to the reactive cysteine(s) of each target protein was performed using Haddock (version 2.2.) (De Vries et al., 2010; van Zundert et al., 2016). We followed the standard covalent docking protocol of Haddock (HADDOCK developer team, 2018). Such protocol was initially tested for covalent inhibitors of cathepsin K (HADDOCK developer team, 2018) and here we have validated it using experimental protein structures containing Cys-ACR covalent adducts (see Section 1.2 in Supplementary Material S1). The covalent bond between Cys and the ligand is modeled by scaling down the van der Waals radius of the Cys sulfur atom 10-fold and introducing two distance restraints: (i) between the sulfur atom of the targeted cysteine and the reactive carbon atom of the ligand, set to 1.8 ± 0.1 Å (i.e. the average length of a single C-S bond) and (ii) between the cysteine C _β atom and the ligand carbon atom adjacent to the reactive carbon, set to 2.8 ± 0.1 Å (i.e. the same as between the C _γ and C _ϵ atoms of methionine, to model the proper angular geometry). The docking procedure (Kurkcuoglu et al., 2018; Koukos et al., 2019) consisted in the following three different stages: (1) A rigid body docking was performed with all geometrical parameters treated as fixed and allowing 180° rotations to generate 1,000 initial poses. After minimization, the best scored 200 poses were selected for further refinement. (2) A semi-flexible simulated annealing simulation (SA) in torsion angle space was applied to introduce gradually flexibility to the system. SA can be further divided into three steps. (2a) First, a rigid body simulated annealing was performed to optimize orientations of the interacting partners. (2b) Then, the system underwent 1000 molecular dynamics (MD) steps from 500K to 50K, with a 2 fs timestep, in which ligand and protein side chain movement was allowed. (2c) Finally, flexibility was introduced to both protein side chains and backbone, besides the ligand. 1000 MD steps (with a 2 fs timestep) were performed with a stepwise temperature decrement from 300K to 50K. It should be noted that flexibility was only applied to the ligand and protein residues within a range of 5 Å. (3) The final stage of the docking protocol was a refinement in explicit water. Namely, three MD-based steps (with a 2 fs timestep) were carried out: (3a) A heating phase of 100 MD steps from 100K to 200K and to 300K, (3b) 1250 MD steps at a constant temperature of 300K, and (3c) a cooling down phase of 500 MD steps to a final temperature of 100K. In stage (3), both ligand and protein were fully flexible, with the exception of protein backbone atoms. The HADDOCK score settings recommended for small molecule docking were used across the whole protocol (Kurkcuoglu et al., 2018; Koukos et al., 2019). The obtained 200 docking poses were clustered based on their positional protein-ligand interface root-mean-square deviation (iL-RMSD) with a cutoff of 1.0 Å. The Haddock score of each cluster was calculated as the average of the top four structures, as done by the Haddock webserver (De Vries et al., 2010; van Zundert et al., 2016), using the equation: HADDOCK score = 1.0 Evdw + 0.1 Eelec + 1.0 Edesol + 0.1 Eair, where Evdw and Eelec are the van der Waals and electrostatic intermolecular energies, respectively, Edesolv is the desolvation energy and Eair is the distance restraints energy; the weights of the different terms were parameterized for scoring of protein-ligand complexes (Kurkcuoglu et al., 2018; Koukos et al., 2019). Further analysis was performed for the top cluster (i.e. the one with the best average Haddock score) and, if present, also for other clusters with Haddock scores within the standard deviation of the top cluster. For proteins with more than one reactive cysteine, we performed independent dockings for each of the Cys residues. This approximation is valid provided that these cysteines are far enough apart that they (or their ACR covalent adducts) cannot interact with each other. However, in the case of one of the target proteins, creatine kinase (Supplementary Table S1), the two Cys are within 7.1 Å (C _α -C _α distance) and thus we also considered the possibility that the two Cys could be targeted simultaneously by ACR (see Section 3.2.2). In this case, binding of two ligand molecules at the same time was modeled using the multibody docking approach (Karaca et al., 2010) implemented in HADDOCK. Namely, a so-called molecule interaction matrix is used to define partners that interact with each other. In particular, we defined the subsequent interacting pairs: protein-ACR molecule 1, protein-ACR molecule two and ACR molecule 1-ACR molecule 2.

Albumin is a plasma protein able to bind chemically diverse ligands, from hemin and fatty acids to drugs, acting as their plasma carrier/transporter (Fasano et al., 2005). Liquid chromatography–tandem mass spectrometry (LC-tandem MS) experiments have shown that C34 binds covalently acrylamide (Noort and Hulst, 2003; Tong et al., 2004). HSA contains 35 cysteine residues and all form disulfide bridges except C34 (Carter and Ho, 1994). This single free Cys is solvent exposed, with a SASA value of 8.6Ų, and has a calculated pK_a value of 10.2 (see Supplementary Table S1). This is in line with spectroscopic measurements showing HSA Cys34 to be more acidic than a normal Cys, with a pK_a around 7 (Pedersen and Jacobsen, 1980). The difference between the computational and experimental pK_a values can be ascribed to the known limitations of computational pK_a predictors when dealing with Cys residues (Awoonor-Williams and Rowley, 2016), as well as uncertainties in the experimental estimation of pK_a values using spectroscopic methods. For instance, the pK_a of Cys34 changes by 1.5 pH units depending on the ionic strength of the buffer used (Pedersen and Jacobsen, 1980). Covalent docking of ACR to C34 resulted in two similar clusters in terms of both score (−31.3 and −32.3 a.u., respectively) and cluster size (69 and 63, see Table 1). Mapping of C34 onto the HSA structure also revealed that this cysteine has two putative proton acceptors in the vicinity (H39 and D38) that can deprotonate the thiol group, as well as a positively charged residue (K41) that could stabilize the transition state and/or product of that reaction (see Figure 4A). Comparison with available functional information (Sampath et al., 2001; Fasano et al., 2005) suggests that acrylamide covalent binding to Cys34 might affect the drug binding properties of albumin. In particular, infrared spectroscopy has shown that Cys34 is linked allosterically with Sudlow’s site I for anesthetics such as halothane, propofol and chloroform (Sampath et al., 2001). Hence, formation of a covalent adduct at Cys34 can be transmitted to this site and modulate anesthetic binding.

Creatine kinase is an enzyme responsible for converting creatine to phosphocreatine reversibly using adenosine triphosphate (ATP). CK is inhibited by acrylamide and such inhibition exhibits a biphasic behavior with respect to acrylamide concentration (Sheng et al., 2009), suggesting than more than one Cys residue within CK might be modified. C283 has been proposed as the primary site of ACR modification in CK. Based on site-directed mutagenesis, C283 was shown to be essential for enzymatic activity (Lin et al., 1994). Furthermore, experimental, studies indicated that C283 has a pK_a around 5.7 and thus this cysteine can be present as thiolate. This is probably required to constrain the position of the guanidinium group of the creatine substrate (Wang et al., 2006). The low pKa of C283 and its role in CK enzymatic activity makes it a good candidate for the main reactive Cys targeted by acrylamide. In contrast, the secondary site of ACR modification is unclear. A combined experimental and computational study (Sheng et al., 2009) suggested that acrylamide can bind to C283, as well as the nearby C74, but the results were not conclusive. Therefore, we performed docking for all five solvent exposed cysteines in CK (C74, C141, C146, C254 and C283; see Table 1). C283 is the most solvent exposed cysteine (37.2Ų) and with the lowest predicted pK_a value (around 9, see Supplementary Table S1). Although the calculated pK_a of C283 (∼ 9) differs from the experimentally measured value (5.7), we ascribed such difference to the known limitations of computational (implicit solvent-based) predictors when estimating the pK_a values of Cys residues, with RMSDs between 3.41 and 4.72 pKa units (Awoonor-Williams and Rowley, 2016). Given this uncertainty, we decided to use the calculated pK_a values only to rank by relative acidity the Cys residues within the same protein, i.e. C283 is the most acidic Cys in CK. In addition, the C283 docking yields the best score (−32.8 a.u.), as shown in Table 1. Furthermore, C283 is located in the catalytic site of CK, whereby nearby residues, such as R96 (distance of sulfur atom to ζ-carbon atom of 7.48Å), R132 (10.71Å) and R236 (10.70Å), can electrostatically stabilize the enolate intermediate of the Michael addition reaction (see Supplementary Table S3). Instead, docking at C74, previously proposed as acrylamide binding site (Sheng et al., 2009), gives a less favorable docking score (−21.3 a.u.). Together with most of the C74 poses showing distance values between the ligand C _β atom and sulfur atom outside the defined covalent bond range, this suggests that modification of C283 is preferred over binding to C74. Due to the proximity of C74 to C283, we also explored the possibility of two acrylamide molecules binding simultaneously to both C283 and C74, using a multibody docking approach (Karaca et al., 2010). The resulting docking poses indicate that adduct formation with one acrylamide molecule already occupies fully the pocket lined by C283 and C74 and thus will preclude binding of a second acrylamide molecule (see Supplementary Figure S12). Thus, C74 is unlikely to be modified by ACR, either alone or in combination with C283.

In contrast, docking to other cysteine residues revealed more suitable candidate for the secondary site of ACR modification in CK. The results for C141 and C146 yielded docking values closer to those of C283 (see Table 1), suggesting that modification of these two cysteines by acrylamide might be possible. Out of these two cysteines, C141 has a slightly more favorable docking score (−31.7 a.u.) than C146 (−28.2 a.u.), as well as a higher solvent exposed surface area (11.1 compared to 2.6Ų), suggesting a slightly higher preference of ACR for C141 over C146. Additionally, C141 has two nearby residues (H145 and E150) that could facilitate thiolate and/or adduct formation (see Supplementary Figures S6 and S7), whereas C146 is hydrogen bonded to P143 (see Supplementary Figures S6–S11). Based on our covalent docking results and the biphasic time dependent inactivation of CK observed in enzymatic assays (Sheng et al., 2009), we propose a molecular model in which ACR modification of C283 (Figure 4B) occurs first and is the primary site responsible for enzyme inactivation. Adduct formation at C283, located in the enzyme active site (Wang et al., 2006), will hinder creatine binding. At longer times, C141 might also be modified by ACR, further contributing to enzyme inactivation by thiol depletion (Meng et al., 2001; Lü et al., 2009).

Acrylamide exposure has been shown to result in decreasing dopamine concentrations by altering postsynaptic dopamine receptors (Erkekoglu and Baydar, 2014). Site-directed mutagenesis data showed that electrophilic compounds, such as NEM, blocked ligand binding to the dopamine D3 receptor (D3R) by modifying C114 (Alberts et al., 2009). Furthermore, functional data compiled in GPCRdb (Kooistra et al., 2020) indicates that C114 is involved in both ligand binding and receptor activation. Taken together the C114 reactivity and functional data, we considered C114 as the most likely candidate for acrylamide modification. Modeling of the covalent C114-ACR adduct further revealed how ACR modification can impair D3R signaling. The ligand interacts with D110 (Figure 4C); this aspartate is essential for ligand binding in aminergic GPCRs (Michino et al., 2015), such as D3R. Together with the aforementioned functional roles of C114 (Kooistra et al., 2020), this indicates that formation of the ACR covalent adduct will hinder ligand binding and/or impair receptor activation. Moreover, Cys at this position (3.36, following the Ballesteros-Weinstein generalized numbering for class A GPCRs) is conserved across dopamine receptors D2, D3 and D4. Given the role of these receptors in dopaminergic neurotransmission, ACR modification of Cys(3.36) might be one of the molecular mechanisms by which ACR intoxication mimics Parkinsonian symptoms.

Dopamine transporters are integral membrane proteins responsible for regulating dopamine neurotransmitter concentrations at the synaptic cleft (Giros and Caron, 1993). Chemicals such as peroxynitrite and 2-aminoethyl methanethiosulfonate (MTSEA), which have in common the potential to modify cysteine sulfhydryls, are known to inhibit DAT (Park et al., 2002). Mutagenesis data has also shown that oxidation of C342 causes a decrease in DAT activity (Park et al., 2002). Furthermore, Cys modification is enhanced if the transporter is in the inward-facing state (Chen et al., 2000). To understand this differential reactivity of the two conformational states of DAT, we performed two covalent dockings for C342, using DAT structures in either outward- and inward-facing conformations (hereafter, OF and IF). Since experimental structural information for human DAT is missing, we generated homology models of the two transporter conformations. The templates used for the OF and IF models were the Drosophila melanogaster DAT (PDB code 6M2R) (Pidathala et al., 2021) and the human serotonin transporter (PDB code 6DZZ) (Coleman et al., 2019), respectively. The target-template sequence identities are 56.2% (OF) and 52.4% (IF); thus, the models are expected to be medium-to-high quality (Chothia and Lesk, 1986; Olivella et al., 2013; Piccoli et al., 2013). We further assessed the quality of the models by calculating their Ramachandran plots (Supplementary Figure S1) and QMEANbrane local quality values (Supplementary Figure S2). The percentage of residues in favored/allowed regions is 93.3%/98.7% (OF) and 95.6%/99.4% (IF), whereas the predicted local quality scores are above 0.7 (except for loop regions or not resolved in the template structures). Thus these two quality assessments support the reliability of the DAT homology models used here. SASA calculations show that C342 is more solvent exposed in the IF model, with SASA values four-fold larger than the OF model (see Supplementary Table S1). Therefore, cysteine accessibility seems to play a role in the observed higher reactivity of ACR with the IF state (Chen et al., 2000). Our covalent docking results (see Table 1) further support the enhanced ACR modification in the IF state. The top cluster for the IF model (Figure 4D) had a more favorable score of −15.1 a.u. than the one (−2.0 a.u.) for the OF model (Figure 4E). Such preferential binding of acrylamide to C342 in the IF state could alter the conformational transition between the two states crucial for dopamine transport, resulting in altered neurotransmitter concentrations. Considering the link between DAT and Parkinson’s disease, it is tempting to suggest that this might be responsible, at least in part, for the PD-like symptoms of acrylamide neurotoxicity (McHugh and Buckley, 2015; Jayaramayya et al., 2020).

Besides C342, we also performed covalent docking for C315. Mutagenesis experiments and transport assays upon treatment with sulfhydryl reagents have shown that C342 is the main modification site responsible for transport inhibition in wild-type DAT (Chen et al., 2000). However, C315 can also be modified and have a minor contribution to transport inhibition in a DAT C90A/C306A/C319F/C342A mutant construct lacking C342 (Whitehead et al., 2001), with higher C135 accessibility in the IF state (Chen et al., 2000; Whitehead et al., 2001). The docking score for this alternative C135 site (−11.9 a.u.) in the IF state is less favorable than for the main C342 site (−15.1 a.u.), further supporting our proposal that the docking score can help discriminate the most reactive Cys within a given protein target.

GAPDH is a housekeeping enzyme involved in both glycolysis, as well as apoptotic cell signaling. Multiple studies, both experimental and computational, have shown that acrylamide can covalently modify GAPDH, inhibiting enzymatic activity (Tanii and Hashimoto, 1985). Moreover, such enzymatic inactivation is concentration- and time-dependent, as well as pH sensitive Martyniuk et al. (2011). C152 has been identified as the most reactive cysteine compared to two other solvent exposed cysteine residues, C156 and C247 (Martyniuk et al., 2011). At small concentrations of acrylamide, almost only C152 is modified. As C152 is essential for GAPDH catalysis by acting as nucleophile, formation of the Michael adduct will result in enzyme inhibition (Martyniuk et al., 2011). However, at higher concentrations, ACR adducts with C156 and C247 are also formed and have been shown to further contribute to enzyme inhibition. To identify features that could explain this differential reactivity, we performed covalent docking for each of the aforementioned Cys residues (see Table 1; Figure 4F; Supplementary Figure S17–20). The top docking cluster for residue C152 had score of −12.3 a.u., significantly more favourable than those for C156 and C247 (46.1 and 9.9 a.u., respectively). Moreover, the last two dockings showed poses with C-S distances outside the covalent bond range. Therefore, modeling of the Michael adduct indicates that C152 is the primary binding site of ACR in GAPDH, in agreement with experiments (Martyniuk et al. 2011). In addition, both C156 and C247 had a calculated pK_a above 12, and thus are less likely to become deprotonated. Instead, the calculated pK_a value of 6.6 for C152 suggests that the sulfur atom can be present, at least partially, as a thiolate anion. Besides, C152 forms a hydrogen bond with residue H179 (Figure 4F); this could help deprotonate C152. Indeed, H179 activates the thiol group during enzymatic catalysis (Soukri et al., 1989). Moreover, the resulting doubly protonated H179 and its respective positive charge could electrostatically stabilize the enolate-type intermediate of the Michael addition reaction. Furthermore, the SASA values further support C152 as the most reactive residue, since its predicted accessibility is two orders of magnitude higher than C156 and C247 (see Table 1). Altogether, the computational results are in agreement with the experimental evidence that C152 is the primary site of acrylamide modification of GAPDH. Moreover, the higher reactivity of C152 with respect to C156 and C247 seems to correlate with the more favorable score obtained for the first cysteine.

Hemoglobin is a heme-containing protein responsible for oxygen transport from lungs to other tissues. Structurally, Hb is a heterotetramer (Ahmed et al., 2020) formed by two α and two β subunits that assemble as dimer of dimers (α ₁ β ₁ and α ₂ β ₂, respectively). Mass spectrometry showed that C93 within the β chains and C104 in the α chains are modified by acrylamide, with C93 being the most reactive site (Basile et al., 2008). Hence, measuring the levels of ACR-modified Hb in plasma can be used to monitor acrylamide exposure (Barber et al., 2001; Basile et al., 2008). The two aforementioned reactive cysteines have a predicted pK_a value above 12. However, C93 is more solvent-exposed compared to C104 (see Supplementary Table S1), suggesting that C93 is more accessible to acrylamide. This is in line with the covalent docking results obtained here (Table 1). The top docking cluster for C93 shows a significantly better docking score (−13.9 a.u.) compared to −3.3 a.u. for C104. Indeed, docking simulations with C104 did not result in a properly formed covalent bond between Cys and acrylamide, which could explain the less favorable docking scores compared to C93. Moreover, C93 has three nearby potential proton acceptors, H346, H97 and D294 (Figure 4G), belonging to the same β ₂ subunit. C93 is also located near K40 of the adjacent α ₁ chain, which stabilizes the adduct by forming a hydrogen bond with the ligand oxygen atom (see Supplementary Figures S21–S23). In addition, the nearby positively charged side chain could help stabilize the transient negative charge developed on the ligand oxygen atom during the nucleophilic attack. Instead, C104 only has a single nearby residue, H103, which could deprotonate the thiol group or stabilize the adduct. Taken together, our computational analysis suggests that C93 in the β subunit should be the primary site for acrylamide adduct formation in Hb, whereas binding to C104 in the α subunit is likely to occur only at higher acrylamide concentrations or longer times, in line with the experimentally observed reactivity (Basile et al., 2008). Moreover, the location of C93 at the interface between the α ₁ and β ₂ suggests that covalent modification of this cysteine by acrylamide could affect Hb function. Oxygen binding to Hb induces changes within this quartenary structure, i.e. a conformational transition from deoxyhemoglobin (T-state) to oxyhemoglobin (R-state). The largest movement occurs between the α ₁C-helix and the β ₂FG corner and a smaller change takes place between the α ₁FG corner and the β ₂C-helix (Perutz, 1979). Thus, the aforementioned α ₁ β ₂ intersubunit location of C93 might alter the transition from the T to R state triggered by oxygen binding to Hb and/or its cooperativity mechanism. We propose here that the effect of ACR modification on C93 could be tested experimentally, for instance, by measuring oxygen saturation at different oxygen partial pressures, after incubation with acrylamide. In the absence of such experimental validation, the functional impact of ACR modification is partially supported by a previous experimental study showing that covalent modification of C93 and C104 by other (larger) organic compounds prevented formation of the Hb tetramer (Hwang and Greer, 1980).

N-ethylmaleimide(NEM)-sensitive factor is a homohexameric ATPase (Hoyle et al., 1996). In the presynaptic neuron, NSF, together with SNARE proteins, is involved in fusion of neurotransmitter-loaded vesicles with the cell membrane and vesicle recycling and thus is key for synaptic neurotransmission (May et al., 2001). Previous studies indicated that thiol reagents (e.g., NEM or NO) inhibit NSF (Matsushita et al., 2003). Mass spectrometry data showed that acrylamide modifies cysteine sulfhydryls, thereby altering the ATPase activity of NSF (Barber and LoPachin, 2004). Moreover, experimental evidence indicates that C264 is a critical residue for NSF function (Zhao et al., 2007). This cysteine is located in a so-called Walker A motif, important to ATP binding and thus for NSF ATPase function.

Due to the lack of experimental structural information for human NSF, we generated a homology model based on a Cricetulus griseus template (PDB code 3J94) (Zhao et al., 2015), which has sequence identity of 98.4%. Therefore, the model is expected to be high quality (Chothia and Lesk, 1986; Olivella et al., 2013; Piccoli et al., 2013). Indeed, its Ramanchandran plot (Supplementary Figure S1) shows 90.4%/96.5% residues in favored/allowed regions (comparable to the 93.0%/98.0% values, respectively, for the template structure, solved at 4.20Šresolution). Additionally, the QMEANDisCo local quality values (Supplementary Figure S2) are mostly above 0.7 (except for loop regions or not resolved in the template structures), further supporting the quality of the model. Our computational analysis using this homology model showed that, among the potential attachment points of ACR, C264 is both the most susceptible to deprotonation and the most accessible to acrylamide, with a pK_a value of 7.7 and a SASA value of 85.2Ų. The docking results further support the modification of C264 by acrylamide (Figure 4H). Structural inspection of the docking poses revealed that ACR would partially block access to the ATP binding site, thus hindering ATP binding and decreasing NSF activity.

Filling of the synaptic vesicles with neurotransmitters relies on the proton gradient created by the vesicular proton ATPases (v-ATPases) using ATP. N-ethylmaleimide (NEM), a sulfhydryl reagent similar to ACR, reduces H⁺ uptake, as well as decreases v-ATPase activity (Shulamit and Sihra, 1989). The modification site responsible for v-ATPase inhibition by sulfhydryl reagents (Feng and Forgac, 1992; LoPachin and Barber, 2006) has been proposed to be C254, which is located in a loop segment of the v-ATPase catalytic subunit, corresponding to the so-called Walker A (GAFGCGKT) motif coordinating ATP binding and hydrolysis. Physicochemical characterization of C254 revealed that this cysteine has a pK_a value of 10.3 and a SASA value of 11.0 Å, further supporting this particular cysteine as ACR target site. Thus, we performed the corresponding Haddock calculation for C254. The covalent docking poses obtained here (Figure 4I) show that the ligand is placed at the entrance of the active site and thus can hinder ATP binding. Nonetheless, the loop where C254 is located exhibits large rearrangements during the conformational cycle of v-ATPase (see Supplementary Figure S28) and such structural changes cannot be modeled with covalent docking. Hence, we integrated additional experimental data for validation. Our hypothesis is indirectly supported by the experimental observation that NEM inactivation of v-ATPase is associated with exposure of a single cysteine residue that can be protected by incubation with nucleotides (Hunt and Sanders, 1996). Moreover, another experimental study showed that C254 modification, either through formation of a disulfide bridge with C532 or through adduct formation with NEM, causes inactivation of the v-ATPase (Feng and Forgac, 1994). Therefore, we surmise that modification of C254 by ACR might have a similar inhibitory effect.