Use of Computational Biochemistry for Elucidating Molecular Mechanisms of Nitric Oxide Synthase.

Nitric oxide (NO) is an essential signaling molecule in the regulation of multiple cellular processes. It is endogenously synthesized by NO synthase (NOS) as the product of L-arginine oxidation to L-citrulline, requiring NADPH, molecular oxygen, and a pterin cofactor. Two NOS isoforms are constitutively present in cells, nNOS and eNOS, and a third is inducible (iNOS). Despite their biological relevance, the details of their complex structural features and reactivity mechanisms are still unclear. In this review, we summarized the contribution of computational biochemistry to research on NOS molecular mechanisms. We described in detail its use in studying aspects of structure, dynamics and reactivity. We also focus on the numerous outstanding questions in the field that could benefit from more extensive computational investigations.

Introduction

Nitric oxide (NO) is a gaseous radical identified as a signaling molecule by Furchgott, Ignarro, and Murad, who were awarded the Nobel prize in Physiology and Medicine in 1998 for their pioneering findings on this small, highly diffusible molecule [1,2]. Before NO was identified and characterized, the molecule responsible for vasodilation was known as “endothelium-derived relaxing factor” [3,4] and many years passed until it became clear that this was NO [1,2,5]. The finding of its involvement in the cardiovascular system [6] was followed by the realization of its relevance in the context of immune response [7,8], neurotransmission [9], respiratory function [10], and a number of different other biological processes, establishing NO as a crucial physiologic regulator of cellular signaling.

Because of its highly diffusible nature and reactivity, NO is a prototypic cellular messenger, as it can signal in a dose-, time-, and organ-dependent manner by directly targeting cellular biomolecules such as proteins and nucleic acids. NO is generated by a class of nicotinamide adenine dinucleotide phosphate (NADPH)-dependent NO synthases (NOSs), from L-arginine and molecular oxygen (Fig. 1). Several cofactors are involved in the reaction, which allow the electron flow from NADPH to the heme group and molecular oxygen, through a molecule of flavin adenine dinucleotide (FAD), and then flavin mononucleotide (FMN). During catalysis, the tetrahydrobiopterin (BH4) cofactor provides an additional electron, which is replaced during the catalytic turnover [11].

Fig. 1

Two-step oxidation of L-arginine catalyzed by NOS. The NHA intermediate is depicted in its protonated (left) and neutral (right) states.

Three isoforms of NOS that require binding to calmodulin to be active have been identified so far in mammals [12,13]. Two are constitutively expressed (cNOSs), while the third is inducible (iNOS or NOS2). Constitutive NOSs comprise the neuronal NOS (nNOS or NOS1) and the endothelial NOS (eNOS or NOS3), both of which have calcium-dependent activity, whereas iNOS can produce NO efficiently and massively without calcium [14]. nNOS plays a pivotal role in the nervous system, as it is involved in synaptic plasticity for regulation of nerves tone [15]. It is also likely to be involved in long term potentiation, because of the neurotransmitter properties of the NO produced in the central nervous system [6,16]. nNOS also has many crucial functions also in skeletal muscle cells [17], where it localizes below the sarcolemma by interacting with dystrophin [18,19], and it regulates blood flow in muscle cells during exercise. Moreover, nNOS is associated with the cardiac sarcoplasmic reticulum, where it regulates myocardial contraction by exerting highly specific, localized NO production that acts on ion channels or transporters involved in calcium cycling [[20], [21], [22]]. eNOS is expressed mainly in vascular endothelium to guarantee vasorelaxation [23,24], cellular proliferation, white blood cells adhesion, and platelet aggregation [24,25]. cNOSs exert their functions functions mainly by producing low, controlled fluxes of NO, whereas iNOS acts mainly as a cytotoxic, antimicrobial enzyme, which is induced by stress and inflammatory conditions [26]. It has been reported that the activation of iNOS by pro-inflammatory cytokines (such as interleukin-1, tumor necrosis factor α, and interferon γ) results in massive production of NO and sustains host immunity as part of the oxidative burst of macrophages. Other immune cell types also respond to NO [7], resulting in an even wider role of iNOS in immunity. For example, it is well documented that iNOS-derived NO can activate T-cells [8] and interfere with lymphocyte development [27] and death [8,28]. iNOS has also been linked to cancer progression and development, but its role in cancer biology has not been fully elucidated, as both tumor promoting and inhibiting activities have been described [29,30].

Multiple molecular effects are induced by NO, ranging from signaling to irreversible modification or damage (Fig. 2), strongly affecting several physiological processes, indicating that NOS defects underlie many human disease conditions. Understanding the structure and mechanisms of action of NOSs is, therefore, fundamental to developing useful clinical interventions. In this review, we summarize the state-of-the-art information on the structure and reactivity of NOSs, with an emphasis on the contributions of computational biochemistry, molecular modeling, and simulations. Current knowledge and pending questions are discussed in the following sections, and perspectives for molecular modeling studies are proposed in the final section.

Fig. 2

Schematic representation of NO-mediated signaling pathways. (A) NO induces the soluble guanylate cyclase (sGC) by binding its heme-group and stimulates the production of cyclic GMP (cGMP) [182]. cGMP production modulates calcium channels and activates the protein kinase G (PKG), leading to a downstream phosphorylation cascade that is important in muscle tone control. (B) Other central sensors of NO fluxes are mitochondria, which adjust the oxygen consumption rate and energy production according to NO levels. NO can affect the mitochondrial respiration rate by direct attachment to Fe-S centers or by the covalent binding to specific tyrosines (C) and cysteines (D) [[183], [184], [185]]. (C) Large amounts of NO produced by immune cells (e.g. macrophages) react with superoxide (O2·), generating the highly reactive peroxynitrite (ONOO-), which leads to protein tyrosine nitration, DNA nitro-oxidation, cell damage, and death. (D) The reaction between NO and nitrogen dioxide (NO2) or redox metals (e.g., Fe3+, Cu2+) generates dinitrogen trioxide (N2O3) or nitrosonium ion (NO+), respectively. Both these species can bind directly to cysteine residues (-SH) of proteins, forming S-nitrosothiols (-SNO). The reaction, termed S-nitrosylation, acts as a posttranslational modification that affects protein function, stability, localization and signaling. The processes in which protein S-nitrosylation plays a role include apoptosis, cell cycle, cell proliferation, gene transcription, mitochondrial homeostasis, and development [82,186,187].

NOS Structure

The NOS structure has two main domains. The N-terminal oxygenase domain (NOSox) harbors the heme porphyrin center (Figs. 3 and 4), that catalyzes oxidation of L-arginine to L-citrulline, requiring BH4 and resulting in the release of NO (Fig. 1), while the C-terminal reductase domain consists of three binding domains for FMN, FAD, and NADPH cofactors (Figs. 5 and 6). The latter domain provides electrons for the reaction, which takes place in the NOSox domain. These two domains are connected through a linker, which includes a calmodulin binding site that is essential for NOS activity. cNOSs binding to calmodulin is responsive to calcium levels, whereas this is not the case for iNOS, which binds calmodulin independently of calcium concentrations.

Fig. 3

Structure of the human nNOSox dimer (PDB: 4D1N). Monomers are colored differently for sake of clarity. The left magnified section (red) shows the ZnS4 motif, with arrows pointing to the cysteines of the first monomer and the central zinc cation depicted in Van der Waals representation. The right magnified section (blue) provides a view of the active site, with the heme moiety in the center, coordinated to Cys420 on one side and dioxygen on the other. Interacting with the heme carboxylates, the pterin BH4 cofactor forms π-stacking interactions with Trp683. Glu597 stabilizes the substrate above the heme moiety, and the aromatic residues Phe589 and Trp414 sandwich the latter to maintain the proper organization of the active site. Trp592, presumably involved in the electron transfer from NOSred to NOSox, is located at the bottom of the heme-binding pocket. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 4

Sequence alignment of the three human NOS isoforms generated with ClustalW [188] for the oxygenase domain, harboring heme-interacting residues (red squares), Zn-coordinated thiolates (green circles), BH4-interacting tryptophan and L-arginine-binding glutamate (blue circles), post-translationally modified cysteines (pink pentagons), and the tryptophan hypothesized to be involved in the electron transfer from FMN to the heme (black star). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 5

Structure of the rat nNOS reductase domain (PDB: 1TLL, chain A), with the NADPH domain in blue, the FAD domain in violet, the FMN domain in red, the α-helix hinge section in gray, the C-terminal tail in green, and the beta-finger in the pink circle on the left. The magnified section A (bottom left) is the active site of the reductase, revealing the placement of the cofactors within the structure, with Phe1395 stacking with the FAD flavin, and Arg1400 interacting with NADPH phosphates. Magnified section B (right) shows two regulatory elements: CT in green and the autoinhibitory segment of the FMN domain (AH) in red, front view (top) and top view (bottom). The two phosphorylated serine residues are also displayed. Magnified section C (bottom right) shows the conserved salt bridge linking the FMN and FAD domains. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 6

Sequence alignment of the three human NOS isoforms generated with ClustalW [188] for the reductase domain, with calmodulin-binding in orange, FMN-binding in red, connecting regions in gray, FAD-binding in purple, NADPH-binding in blue, and the C-terminal tail in green. Phosphorylation sites are indicated by green squares, the phenylalanine shield and the NADPH-interacting arginine by red circles, aromatic residues that may be involved in electron transfer from FMN to heme by black stars, and the salt bridge between the FMN and FAD domains by light blue circles. The auto-inhibitory and beta-finger regions appear as dashed lines within the FMN and connecting domains, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

The electrons from the reductase domain are shuttled toward the oxygenase reactive site through long-range displacement of the FMN domain [31]. Although this ‘swing’ has been the subject of many studies, aspects of its mechanisms of regulation are not yet understood [[32], [33], [34], [35], [36], [37], [38], [39]]. The NOS isoforms are functional upon dimerization, and electron transfer is likely to occur in trans from one monomer to the other [40,41]; however, the details of this mechanism, such as the contact surface and the elements that stabilize the dimer interface, are still being investigated (Section 2.3). The three-dimensional (3D) structures of the full-length NOS variants are unknown, although numerous X-ray structures of the individual domains (Table 1) and models built from cryogenic electron microscopy (cryo-EM) densities and other experimental data are available [[32], [33], [34], [35], [36], [37],[42], [43], [44], [45]], so that the architecture of the full-length NOS can be hypothesized.

Table 1

Available X-ray and NMR structures of the domains of each NOS isoform, along with the PMID associated to the corresponding publication in Pubmed and the corresponding entry in the Protein Data Bank (PDB). We did not report the structures of the heme domain in complex with inhibitors.

Isoform	Organism	Domain structure	PDB ID	PMID
nNOS	Human	Heme	4D1N	25286850
	Rat	Heme	4FVW, 3HSN, 2G6H, 1ZVI, 1LZX	23586781, 19791770, 16804678, 16033258, 12437343
		Reductase	1TLL	15208315
		FNR-like (FAD/NADPH)	1F20	11473123
iNOS	Human	Heme	1NSI	10409685
		calmodulin binding + calmodulin	5TP6, 2LL6	28121131, 22486744
		FMN/calmodulin binding + calmodulin	3HR4	19737939
	Mouse	Heme	3NQS, 1DWV, 1DF1	20659888, 10769116, 10562539
		Heme mutant	3DWJ, 1JWJ	18815130, 11669619
eNOS	Human	Heme	3NOS, 4D1O	10074942, 25286850
		calmodulin binding + calmodulin	2MG5, 2LL7, 2N8J, 1NIW	24495081, 22486744, 27696828, 12574113
	Bovin	Heme	1NSE, 2G6O, 1ED6	9875848, 16804678, 11331003

Computational studies would address several outstanding questions about the structural mechanisms of NOSs; however, only a few molecular dynamics simulations of NOSs have been reported, on very short timescales, no extensive structural analysis, and most as part of more comprehensive experimental studies. Other computational approaches, such as homology modeling and docking, have been used together with experiments to determine the quaternary structure of NOS [[32], [33], [34], [35], [36], [37],[43], [44], [45], [46]]. However, they suffer of major limitations in terms of accuracy and the capability to account for highly flexible and conformationally heterogeneous proteins. The interaction of NOSs with selective inhibitors has been reported [47,48] but there has been no extensive computational investigation of the free enzyme structure.

In this section, we summarized the current knowledge on NOS structural features from studies with computational methods and raise questions that would benefit from more extensive, accurate studies with molecular modeling and simulations.

The Oxygenase Domain

The NOSox is composed of a unique heme domain, harboring a heme-porphyrin catalytic center, a structural zinc tetrathiolate (ZnS4) motif, and the BH4 pterin cofactor. Numerous high-resolution structures of the three human isoforms of NOSox are available (Table 1), which provide considerable source of information for unraveling the structural features of this catalytic domain.

The geometry of the active site is highly conserved among NOS isoforms and mammalian species [[49], [50], [51]]. The heme iron is axially coordinated to a cysteine thiolate on one side (Cys420 in nNOS), and the other side of the porphyrin is the binding site for the dioxygen molecule required for L-arginine oxidation (Fig. 3). The heme insertion within the NOSox domain might be favored by the interaction of NOS with heat shock protein 90 (hsp90), by a structural deformation that allows access of heme to its binding cleft in the protein [52,53], concomitantly promoting NOS dimerization [54]. A conserved glutamate residue (Glu597 in nNOS) is likely to play an important role in substrate binding [51], as shown by experimental mutagenesis [55,56]. The pterin redox cofactor binds in the vicinity of the active site through interaction with the heme propionate groups, and this process is thought to promote the binding of L-arginine [57]. Conserved aromatic residues near the active site form stacking interactions with the porphyrin moiety (Trp414 and Phe589 in nNOS) and the pterin cofactor (Trp683 in nNOS), which is involved in an extensive hydrogen-bond network. This network is likely to promote the stabilization of the NOSox dimer interface and the binding of L-arginine to the enzyme [49]. An important tryptophan residue located at the back of the heme pocket (Trp592 in nNOS) has been proposed to shuttle the electron from the FMN cofactor to heme (Section 2.3).

Another element that favors dimerization is the zinc tetrathiolate (ZnS4) motif. Indeed, the zinc ion is tetra-coordinated with two thiolates (Cys331 and Cys336 in nNOS) from each monomer, contributing to the maintainance of the architecture of NOS by bridging the two NOSox domains (Fig. 3). The absence of this cation or the modification of one of the coordinated cysteines drastically destabilizes the dimer and thus reduces the NOS catalytic activity [[58], [59], [60]].

Surprisingly, no computational studies have been carried out so far on the dynamic properties of NOSox and the catalytic site with NOS in the closed state. To the best of our knowledge, the only study with MD simulation of isolated NOSox involved a very short trajectory below the nanosecond timescale. This 300-ps MD simulation was performed within a reactivity study of iNOS active site with fixed heme, dioxygen, coordinated cysteine, and surrounding water molecules [61]. The simulations were performed with Turbomole/Jaguar, coupled to DL-POLY. The QM and MM simulations were treated at the B3LYP/LACV3P*+ level of theory and with CHARMM potentials, respectively. Electronic embedding was applied. Given the very short timescale, this study did not allow a proper assessment of the dynamic behavior of the NOSox domain, which would have required longer, unconstrained MD simulations. In another study, the interaction of NOSox with caveolin-1 was studied by docking and short MD simulations (10ns) with a CHARMM force field, and accompanied by experimental investigations [62]. Caveolin specifically decreases eNOS activity by hindering calmodulin binding and further activation [63,64]. The MD simulations were performed to assess the stability of the docking poses and were too short to elucidate the dynamics of the complex between NOSox and caveolin-1. Furthermore, few details were given about the simulation, so that the data would be difficult to reproduce. Their results suggest that caveolin-1 prevents binding of the BH4 cofactor by interacting with the eNOS Trp447 (equivalent to Trp683 in nNOS). However, given the lack of technical details and the short timescale, this hypothesis must be confirmed in studies of the competitive binding of caveolin-1 to eNOS against BH4.

Designing effective isoform-selective inhibitors is important for targeted therapies, and modeling and simulations could be valuable in this context. Docking, Quantitative-Structure-Activity Relations (QSAR), and GRID/Consensus Principal Component Analysis (CPCA) approaches have been used to probe the chemical environment within the heme binding site and the ligand-host interactions [48,65], sometimes coupled with homology modeling, MM-Poisson-Boltzmann (Implicit Solvent Model) Surface Area calculations, thermodynamic integration, and short MD simulations [[66], [67], [68], [69], [70], [71], [72]]. These computational studies shed light on the subtle electrostatic differences among the active sites of the NOS isoforms, opening new avenues for enhanced selective inhibitor design. Many questions remain, however, about NOSox structure and regulation. Numerous experimental data are available on the structure of the NOS heme domain in the closed state, but several aspects remain to be elucidated. For comparison with a homolog, extensive in silico studies (e.g., microsecond all-atom and coarse-grain MD, docking) have been reported on P450 enzymes [[73], [74], [75], [76], [77], [78]], including the investigation of the structure of the active site, substrate tunneling and binding modes, mutational effects, and interaction with membranes. These show the usefulness of molecular modeling in a context similar to NOSox. Use of computational methods with regard to NOSox has essentially focused on the design of selective inhibitors. However, as for P450 enzymes, longer timescale MD investigations with unbiased simulations or enhanced sampling approaches could provide important insight into NOSox domain structural behavior, its interaction with other proteins (including Hsp90 and caveolin [79]), the structural effect of post-translational modifications (PTMs), such as S-nitrosylation/S-sulfhydration of Cys441 [[80], [81], [82]] and S-glutathionylation of Cys382 in eNOS [81], and into the role of BH4 and ZnS4 in the stabilization of the dimer interface.

The Reductase Domain

The NOS reductase domain (NOSred) has three different sub-domains, namely NADPH, FAD, and FMN, each of which binds a specific cofactor for electron transfer (Fig. 5). This organization is similar to that of the cytochrome P450 reductase protein (CPR), which shares 60% sequence homology with NOSred and catalyzes analog electron transfer from NADPH to P450 reactive site [83].

The NADPH and FAD domains assemble to form the ‘FNR-like’ unit, while the individual FMN part is thought to serve as an electron shuttle toward the heme center, as observed in CPR [84,85]. An α-helix hinge section connects the FMN and the FAD sub-domains, ensuring proper alignment of the two flavins in a position to promote electron transfer. The electron goes from NADPH to FAD then to FMN. Upon activation, the FMN domain undergoes large-scale movements to dock on the NOSox domain and terminate electron transfer toward the heme center. This phenomenon occurs in trans, from the reductase domain of one monomer to the NOSox of the other monomer, and is triggered by binding of calmodulin [40,41,86,87].

Several NOSred elements have been suggested to respond to calmodulin binding and regulate FMN domain, unlocking from NADPH/FAD domains in an isoform-dependent manner. First, a conserved auto-regulatory segment is present in cNOSs FMN-binding domain, whose N-terminal α-helix shares sequence similarity with the calmodulin-binding motif [88]. This segment harbors a phosphorylation site at Ser847, suggesting an additional layer of post-translational regulation. Second, the C-terminal tail, the length of which differs among the three NOS isoforms, has been suggested to modulate FAD-FMN interactions in synergy with the auto-regulatory sequence and calmodulin binding [89,90]. The C-terminal tail of cNOS includes the Ser1412 phosphorylation site, which is also likely to play a role in regulating NOS activity. Third, the beta-finger (a small insertion encompassing the so-called CD2A loop) present in the flexible hinge region may play an important role in the closed-to-open state switch, presumably by modulating interactions with the FMN domain, especially in eNOS [[91], [92], [93]]. The subtleties of such regulatory mechanisms at the atomic level are still poorly understood.

Only one structure of the isolated rat nNOS FAD/NADPH-binding domain and another of the human iNOS FMN- and calmodulin-binding domains bound to calmodulin are available [43,94], as shown in Table 1. Nevertheless, the crystal structure of the rat nNOS reductase resolved by Garcin et al. [32] provided unprecedented insight into the structural features and organization of this domain. This was the only successful attempt to crystallize the three subdomains of a NOS reductase domain with the NADPH and flavins cofactors in their respective binding pockets. The structure revealed the extensive network of hydrogen bonds between the three cofactors and the protonation state of the FMN, which appeared as a semiquinone (Fig. 7). In addition, inspection of the FMN/FAD domains interface revealed important hydrophobic contacts and salt bridges, especially between the Glu816 and Arg1229 residues, both conserved in cNOSs. Arg1400, present in both nNOS and eNOS (Arg1165), was suggested to play a role in the selective binding of NADPH, and in locking of FMN in its electron-acceptor state in the absence of calmodulin. The FAD-shielding residue Phe1395 is thought to be involved in the repression of the electron transfer (Fig. 5) in the calmodulin-free state by acting as an aromatic shield between NADPH and FAD [95], as observed in the crystal structure [32].

Fig. 7

Schematic representation of NOS reactivity. a) Electron and proton transfers in the reductase domain leading to the four-electrons reduced state through two electrons reduced to the quasi-equilibrium state. b) Three different oxidation states of the flavin moiety (present in FAD and FMN). c) Mechanism of L-arginine oxidation: first half-reaction on the left (step1, orange path), second half-reaction on the right, with NHA protonated as reviewed by Santolini et al. [118] (step2a, blue path), and NHA neutral as proposed by Shamovsky et al. [117] (step2b, green path) involving formation of the Cpd II-P intermediate with extensive spin delocalization depicted by red dashed lines. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

The functionality of the dimeric form of nNOSred observed in this crystal is, however, a matter of controversy, as the dimer interface is presumably between the NOSox domains rather than the NOSred in the three isoforms, as suggested by recent models guided by cryo-EM data [33,36,37,46]. Likewise, experimental studies show that iNOS dimer is still functional even in the absence of the reductase domain of one monomer [40], indicating that the essential dimeric interface is localized between the two oxygenase domains. Modeling of the nNOS open-state structure from other EM data suggests that interactions between reductase domains might have a stabilizing effect [35]. This remains compatible with the assumptions mentioned above, as the main dimer interface is still found between the oxygenase domains. Dimerization of the reductase domain of the homologous CYP102A1 system has also been suggested [96].

The chronology of the NADPH and calmodulin binding events is still under discussion. Volkmann et al. proposed that binding of NADPH on the eNOS-calmodulin complex triggers electron transfer [37], whereas it is more commonly considered that NADPH binds calmodulin-free nNOS to lock the FMN domain in its electron-donor state until activation upon subsequent calmodulin binding [97,98].

Few studies involving computational biochemistry to investigate NOSred structure have been reported. Homology modeling and protein-protein docking methods have been used to build models from cryo-EM data (Section 2.3), and only one study reported MD simulations on eNOSred, providing a rationale for the effects of phosphorylation on NOS structure on the basis of the analyses of short trajectories [99]. In this study, Devika et al. performed 40-ns MD simulations using the GROMOS96 43a1 force field on the human structure of eNOSred built by sequence homology. The structure was rebuilt with Modeller and structures of homologs as templates (i.e., the rat nNOSred with PDB entry 1TLL, the rat FMN-like unit with PDB entry 1F20, and the human FMN and calmodulin-binding domains in complex with calmodulin with PDB entry 3HR4). Although the results provide insight into mutational effects on the structure of eNOS reductase, MD simulations with homology model as a starting structure should be considered with extreme caution, as they can result in large deviations from the native structure [100].

Molecular modeling has been of particular interest for studying structural features of the homologous CPR system, and a number of computational studies have been reported [[101], [102], [103]]. Thus, extensive unbiased and enhanced sampling MD simulations could provide additional insight, at the atomic level, into the numerous questions about the inner structural organization of NOSred, such as: i) What are the mechanisms of NOSred activity modulation by the regulatory elements? ii) Which structural changes are induced by PTMs in this domain (including phosphorylation of several serines in cNOSs, phosphorylation of Thr495 and S-glutathionylation of Cys689 and Cys908 in eNOS [81,[104], [105], [106]])? iii) Are the regulatory mechanisms of the NOS isoforms different? iv) Which amino acids are involved in the regulatory events, ligand binding, and interaction among the sub-domains? v) How does Arg1400 discriminate between NADPH and NADH? vi) How does NADPH and calmodulin binding affect the structure and dynamics of the reductase domain? Computational studies could provide new insight into the conformational changes of NOSred triggered by its multiple regulatory elements, which would be of utmost importance for understanding the complex NOSred regulation.

Inter-Domain Interactions

The full-length structure of NOS remains elusive in X-ray crystallography because of the flexibility of the NOSred domain. The structural and dynamic processes that drive inter-domain electron transfer and the regulatory effects of calmodulin binding are key to understand the molecular mechanisms underlying NOS activity, and have been investigated with structural biology methods.

As high-resolution structural data are limited to the isolated domains, homology modeling, protein-protein docking, and MD simulations have been used to characterize the sites of interaction between the FMN domain, calmodulin, and NOSox/NOSred. Cryo-EM techniques, combined with homology modeling and/or protein-protein docking, have been used to identify the architecture of holo-NOS and the changes of FMN from the closed to the open state [[35], [36], [37],46]. Other full-length NOS models have been built with modeling approaches, directly from the X-ray structures [32,34,43], and by integration of experimental data from hydrogen-deuterium exchange mass spectrometry [33] or electron paramagnetic resonance [44,45].

Overall, modeling of the multi-domain NOS structure indicates that binding of calmodulin might regulate the swing of the FMN domain from the reductase to the oxygenase domain, thus modulating electron shuttling between the two redox partners. The strong mobility of the NOSred domain with respect to the NOSox dimer has been highlighted with the heme domain acting as an anchor within the full-length architecture. The three NOS isoforms can adopt diverse intermediate conformations, ranging from the closed to the open state. The binding of calmodulin is likely to constrain the movement of the FMN-binding domain toward the output conformation [36].

The existence of a rotational pivot has been proposed, which would guide the FMN domain toward the appropriate binding area on the heme domain by restricting the conformational space accessible along the swing from NOSred to NOSox [37,107,108]. The work of Ilagan et al. [45], who reported a full-length model of rat nNOS (constructed manually), suggested that the FMN subdomain has higher affinity for the FNR-like unit than for the heme domain. In their study, calmodulin binding induced the conformational changes required to destabilize this balance and favored the open state, without increasing the binding affinity of FMN to the heme domain. Persechini et al. [46] suggested that activation of eNOS by calmodulin binding is a two-step process. First, unhooking of the FMN domain from the FNR-like unit would be triggered by interactions of calmodulin with the numerous NOSred regulatory elements. Second, calmodulin would dock onto NOSox and facilitate the docking of the FMN domain itself. Unfortunately, there is no consensus on the exact activation mechanisms after calmodulin binding, and the hypothesis that it might be different from different isoforms cannot be ruled out.

The nNOSred crystal structure published by Garcin et al. [32,34,43] brought unprecedented insight into the contact area of the FMN sub-domain with the FNR-like unit (Section 2.2). Despite the many investigations on the subject, however, the FMN/NOSox domains interface is not well described. Full-length NOS models designed from cryo-EM densities provide information about the possible docking area of the FMN domain onto NOSox, although detailed mapping of the interface contacts is not available.

Recently, Hollingsworth et al. [109] published a MD study of the iNOSox-FMN-calmodulin complex, and confirmed the role of intermolecular salt bridges. Their model system, encompassing the iNOSox dimer and the FMN and calmodulin-binding domains in the iNOS output state (FMN domain in its electron-donor state), was built by manual docking, guided by HDX-MS data [33]. 100-ns MD simulations with restraints on the iron atom of the heme were conducted in the model system with and without calmodulin, using the CHARMM22 force field with additional parameters generated by the same group. The results suggest that the FMN/heme interface is stabilized by salt bridges. The stability of this complex appeared to be enhanced by electrostaticinteractions of the two domains with the bound calmodulin. Hence, calmodulin binding stabilizes the inter-domain interactions, and its absence leads to disruption of the interface contacts after 20-ns MD simulation.

Sheng et al. [108] investigated the interactions of the iNOSox/FMN domains with 60-ns MD and steered MD simulations and found specific residues that are important for the efficient binding of the FMN domain to the NOSox docking surface. Their model was built with the same X-ray structures as that of Hollingsworth et al., but they docked the two domains (FMN and NOSox) with ZDOCK [110]. The generated structures were then filtered according to criteria obtained experimentally. The linker region was constructed with the Scigress Explorer Ultra platform, and 60-ns MD simulations for different oxidation states of the system (i.e., before and after electron transfer of FMN to heme) were performed using the CHARMM27 force field and in-house parameters for the heme, FMN, and BH4 cofactors. The results suggest that conformation changes leading to the swing of the FMN domain after the electron transfer are redox-dependent. The inter-domain interaction network they report differs from that observed by Hollingsworth et al. [109], with many more interactions between the FMN and the two NOSox monomers. The distance between the FMN cofactor and the heme was found at about 18 Å by Hollingworth et al. [109], bridged by a conserved tryptophan residue that has often been proposed to shuttle the electron between the two moieties [33,34,43]. Sheng et al. reported that the Trp372 (equivalent to Trp592 in nNOS) center of mass remained around 11.7 Å from FMN and 9.4 Å of the iron atom. This result differs from that found for the cytochrome P450 structure, in which the FMN-heme distance is only 6 Å, suggesting direct transfer from the cofactor to the heme center [85]. The simulations by Sheng et al. [108] suggest that, in some conformations induced by the dynamics of the NOS system, electron transfer involves not only the conserved tryptophan, but also shuttling through Tyr631 or Phe593, both located near the FMN cofactor in the FMN sub-domain. This hypothesis is supported by experimental studies that show a drastic decrease in the rate constant of intermolecular electron transfer after mutation of Tyr631 to Phe, which would disrupt important interactions [111]. The differences between the two MD studies above might be due to the use of different model structures of the iNOSox-FMN-calmodulin complex in the simulations. Unfortunately, lack of experimental data obviate a conclusion, highlighting the importance of finding highly accurate models by multiple cross-validation with experimental data and a more extensive sampling of the conformational space in the simulations.

Enhanced sampling MD approaches would make it possible to observe structural rearrangements over a longer time, which is necessary because the NOS system is prone to large conformational changes promoted by different allosteric effects. Such simulations would also be important to confirm the long-term stability of the key interactions identified in the short MD trajectories and other structural studies published so far. It could provide insight into the role of calmodulin binding in the activation of the different NOS isoforms and the effects of PTMs, which might be isoform-dependent.

The complexity of the NOS architecture, a homodimer of up to 330 kDa with several co-factors, and regulatory events (PTMs, allosteric effects) makes it a challenging system to investigate with molecular modeling and simulations. Nevertheless, force field parameters are available for most of the cofactors and ligands, and numerous structures of the isolated domains have been deposited in the Protein Data Bank (Table 1). Several in silico investigations of the structure of the homologous P450/CPR system have been published, providing a framework for investigation of the full-length NOS structure [101,112]. The scientific community would gain much from further use of computational methods, which would allow visualization of the NOS full-length structure and details on its dynamics at the atom level, which is essential for answering numerous outstanding questions.

NOS Reactivity

NOS catalyzes the oxidation of L-arginine to L-citrulline and NO in a two-step mechanism involving dioxygen, the BH4 cofactor, and electron transfers from NOSred (Fig. 1). The production of electrons in the reductase domain is similar to that observed in P450 reductase, and the mechanisms are well described [97,113,114]. First, the binding of NADPH to NOS allows transfer of a first hydride to the FAD cofactor. Then, electron transfer from FAD to FMN and proton addition lead to a quasi-equilibrium state between the two-electron-reduced species [FADH2/FMN ↔ FADH·/FMNH· ↔ FAD/FMNH2] until transfer of a second hydride from NADPH, which results in the four-electron reduced FADH2/FMNH2 state [115], as illustrated in Fig. 7. Complete flavin reduction, which is unlikely to take place physiologically, is gated by the release of NADP+ needed for de integro binding of NADPH and hydride transfer toward the final stable two-electron reduced state of FMN, required for the transfer toward the heme [41,116]. Upon calmodulin binding, the FMN subdomain can swing and dock onto NOSox, where it delivers the electron to the iron center, triggering two-steps L-arginine oxidation. The first mono-oxygenation reaction (first half-reaction) results in the formation of the stable Nω-hydroxy-L-arginine (NHA) intermediate, whose subsequent oxidation (second half-reaction) leads to the production of L-citrulline and NO (Fig. 7). Although NOS has many similarities to the CPR/P450 system [83], the reaction mechanisms in NOS are more complicated, as oxidation implies participation of the BH4 cofactor in electron transfer. Although the mechanisms of NOS reactivity are still mainly hypothetical, recent evidence of P450-like reactivity suggests that it involves distinct heme-oxy species [117]. Considerable research has been conducted on the role of the pterin cofactor, the protonation state of the NHA intermediate, and the heme-oxy species formed during the two steps, as they are essential for understanding the mechanisms of NOS reactivity [11,97,[118], [119], [120], [121]]. Although several computational studies (QM and QM/MM calculations) have been reported, many aspects of NOS reaction mechanisms remain hypothetical, especially with regard to the second half-reaction. Below, we summarized the latest findings on L-arginine oxidation mechanisms, in which computational studies have been of major importance.

First Half-Reaction

O2 activation in NOS is catalyzed by the heme moiety (Fig. 7) in a manner similar to that for cytochromes P450. Electron transfer from FMN switches the iron from the ferric FeIII to the ferrous FeII state, stimulating coordination of the dioxygen molecule to the heme center [118]. After binding of O2, a second electron transfer induces formation of the ferric-peroxide FeIIIO22- complex, which is required for the launching L-arginine oxidation [122]. Several hypotheses have been proposed for the mechanisms that lead to formation of the so-called compound I (Cpd I) cation-radicaloid porphyrin ferryl-oxo FeIVO+· species, which oxidizes the substrate [61,118]. The commonly preferred hypothesis is a mechanism involving two proton transfers (one from the solvent and the other from the L-arginine substrate) and formation of FeIIIOOH species [118,119], as supported by theoretical QM and QM/MM studies [[123], [124], [125]]. The first proton transfer would lead to the formation of the so-called compound 0 (Cpd 0) anionic ferric-hydroperoxo FeIIIOOH, and the second H+ addition would result in the generation of Cpd I and a water molecule by heterolytic cleavage of the peroxide bond. The latter species then oxidizes the deprotonated substrate, leading to formation of the stable NHA intermediate and FeIII. Density functional theory (DFT) calculations performed on a model of bovine eNOSox active site published by de Visser et al. [124] suggest that the latter process involves formation of the unstable ferryl FeIVO- state (Cpd II). The initial geometry, extracted from the PDB file 4NSE, involved the truncated heme and coordinated cysteine, L-arginine, and surrounding amino acids (Glu363, Trp358, Trp359). Geometry optimizations were performed with Jaguar at the B3LYP/LACVP and 6-311+G* levels of theory for the iron and the other atoms, respectively. It is well accepted that the second electron transfer required for O2 activation is provided by the BH4 cofactor. The pterin moiety would allow fast electron transfer toward the FeIIO2 intermediate, stimulating the L-arginine oxidation and preventing uncoupling from the NADPH oxidation cycle [126,127]. The latter process is involved in numerous diseases, including vascular diseases and cancer, and is therefore of major therapeutic interest [25,82,[128], [129], [130]]. It depends on availability of L-arginine and/or BH4, disruption of the NOSox dimer, and PTMs and results in FeIIO2 decay to the release of superoxide anion-radical O2·- rather than NO [[131], [132], [133], [134], [135], [136]]. Consequently, BH4 is a key player in NOS reactivity, as it ensures the bio-availability of NO and prevents oxidative stress [57,127,130]. Many studies, including DFT investigations, suggest that the pterin cofactor stays bound at the active site in its radical cationic state H4B+·, to be re-used during the second half-reaction [120,[137], [138], [139]].

Second Half-Reaction

Continuation of NO· catalysis requires prior reduction of H4B+· back to BH4 to prevent its further time-dependent oxidation to BH2 and NOS uncoupling [140,141]. Without it, the NOSox active site does not have the appropriate chemistry for reduction of the iron center from FeIII to FeII, which is necessary for the O2 binding that initiates oxidation of NHA [142,143]. Experimental evidence suggests that the electron is provided by NOSred in a calmodulin-dependent manner, similarly to the iron center reduction, and is transferred from FMN through the heme moiety [120,144]. Whether the electron is transferred to the metallic center or shuttled towards the pterin radical may depend on the redox state of BH4, and hence on the NOS catalytic cycle stage. Theoretical investigations could provide insight to probe this hypothesis. Once BH4 is reduced back to its neutral form, the second half-reaction can take place.

Unlike the first step, the second half-reaction differs from other known enzymatic mechanisms and is poorly understood. Nevertheless, numerous experimental and theoretical investigations suggest possible reaction intermediates for the formation of NO· [118]. As in the first step, FeIII is reduced by electron transfers from NOSred and BH4 to allow O2 activation. Various subsequent mechanisms have been proposed. In most, NHA is considered to be in its protonated form [119,125,145,146]. Oxidation to L-citrulline would then involve a proton transfer from NHA to the ferric-peroxide FeIIIO22- species after formation of a tetrahedral complex that binds the guanidinium moiety to the peroxide. This would induce the release of HNO, L-citrulline, and FeIIIOH. Other possibilities, reviewed by Santolini et al. [118], would involve the transfer of an electron back to H4B+·, terminating the BH4 cycle concomitantly with the binding of the newly formed NO· to the iron center and release of water (Fig. 7). DFT and QM/MM investigations of this hypothesis did not allow theoretical validation but suggested an alternative mechanism, involving double protonation of the active site to formation of an FeIIIHOOH intermediate [146,147], which might eventually lead to the ferryl-oxo FeIVO+· compound (Cpd I) [145]. Hence, no consensus has yet been found on the NOS second half-reaction mechanism.

Recent computational investigations, however, represent new means for understanding the complex chemisty of NOS. Although, it has been accepted that NHA reacts in its protonated state [57,[148], [149], [150]], Shamovsky et al. [117] proposed a new mechanism featuring neutral NHA (Fig. 7). They performed extensive DFT investigations on the active site of the murine iNOSox structure (PDB 1NOD) at the M06-L/6-31+G* level of theory and extracted the coordinates of the heme, NHA, BH4, and 11 surrounding amino acids from the crystal structure to build the initial geometry. Constraints were applied to the heme center during geometry optimization to maintain its location in the crystal structure. Overall, their calculations suggest the participation of BH4 in a sequence of electron transfers coupled with a proton transfer from the heme propionate and involvement of a protonated Cpd II intermediate (Cpd II-P) instead of Cpd I (Fig. 7). They propose that this mechanism would explain several experimental observations and especially the kinetics of the reaction and the elusive character of some intermediates. More details can be found in their elegant publication [117]. More data are now required to validate or refute this promising hypothesis.

Many aspects of NOS reactivity remain to be elucidated, but the work done so far gives encouraging results for further investigations. QM and QM/MM methods have proved their value in this context, as for their application to the P450/CPR system [[151], [152], [153], [154]]. More extensive use of computational approaches could shed light on NOS chemistry aspects, including: (i) the residues involved in electron transfer from NOSred to NOSox in the different NOS isoforms; (ii) how the electrostatic properties of the active site drive the dual chemistry of NOSox; (iii) rationalization of the different reaction yields found for the different isoforms; and (iv) the chemical features of the NHA intermediate and its stability at the active site.

Summary and Outlook

NOS is a highly complex chemical system with several layers of regulation, involving numerous players that mediate the interactions between the reductase and the oxygenase domains, both of which are involved in the enzymatic reactivity. The two-step oxidation of L-arginine necessitates a finely tuned organization of the active site, without which uncoupling from NADPH and production of ROS rather than NO· can occur.

Many studies have been conducted to unravel NOS molecular mechanisms, including the usage of computational methods, especially docking and QSAR to study ligand binding, and, to a lesser extent, QM calculations to probe oxygenase reactivity. Nevertheless, the full potential of molecular modeling is underexploited. Computational approaches could provide important insight on NOS structure and reactivity at the atom level. Homology modeling and docking have already provided the models of individual domains and of the holo-enzyme. Coarse-grained force fields could be considered for initial exploration of the dynamic behavior of the holo-enzyme and to identify possible allosteric effects to guide more accurate all-atom approaches. Computational tools are now available to treat systems, including intrinsically disordered regions, by coarse graining, which would be necessary to model the flexible linkers between NOS domains [155]. Furthermore, extensive MD simulations could provide information on the dynamic behavior of isolated domains, long range reorganization induced by PTMs, and allosteric effects [[156], [157], [158], [159], [160], [161]]. The integration of MD-derived ensembles with the protein structure network paradigm could also help in predicting structural pathways of communication as the basis of allosteric mechanisms [[162], [163], [164], [165], [166]]. Methods of enhanced sampling MD would be adequate for studying the allosteric effects triggered from distal sites to the inter-domain interface [159,[167], [168], [169], [170]]. In view of the growing amounts of EM data, the application of integrative structural modeling and EM density maps could also provide important information about NOS structural and dynamic features [[171], [172], [173]]. Conceptual DFT calculations and topological analysis of electron density, which have been applied to P450 and other enzymes [174,175], could provide unprecedented insight into the chemical properties of NOSox active site during oxidation [[176], [177], [178]]. Use of QM/MM and DFTB-QM/MM metadynamics might also be considered for studying electron transfer from the reductase to the oxygenase domain and identifying the key players involved in this process [[179], [180], [181]]. Overall, there is a wide range of possibilities for the study of NOS molecular mechanisms with molecular modeling and simulations, opening perspectives that could lead to important breakthroughs on this topic.

Declarations of interest

None