1. Thiol peroxidases as sensors

The rate constants for the oxidation of freely accessible SH groups in low–molecular-weight compounds by H₂O₂, even if extrapolated to full SH dissociation, do hardly exceed 50 M⁻¹ s⁻¹ (512) and, thus, fall short by orders of magnitude when compared to those of GPx- or Prx-type peroxidases (see section II.A). Protein-bound cysteines are by no means more reactive (31, 123, 134), unless they are embedded in a micro-architecture that facilitates cleavage of the hydroperoxy bond by polarization and proton shuttling as in the thiol peroxidases (378, 487). Evolution has designed these proteins for highly efficient hydroperoxide reduction. Accordingly, they do not only deserve interest as hydroperoxide-detoxifying enzymes, but also as ideal sensors for ROOH.

In recent years, a sensor/transducer function of peroxidases has indeed been elucidated in transcriptional regulation of yeasts (135). In Saccharomyces cerevisiae a GPx-type peroxidase Orp1 senses H₂O₂ in being oxidized to its sulfenic acid form, as in Eq. 1. The cysteine sulfenic acid residue of Orp1 then forms a disulfide bridge with a particular thiol of the transcription factor activating protein-1 (AP-1)-like transcription factor from yeast (Yap1), thereby directly transducing the oxidant signal (86) (Fig. 2). The physiological meaning of the H₂O₂ sensing by the peroxidase is seen in transducing the oxidant signal to a defined target protein with the specificity typical for protein/protein interactions (86). In another S. cerevisiae strain, the 2-Cys Prx Tsa1 appears to activate Yap1 in an analogous way (357, 407), and in Schizosaccharomyces pombe Tsa1 is the major peroxidase that reacts with the transcription factor Pap1, which is an homolog of Yap1 (333). Also, in S. pombe Tsa1, upon having sensed H₂O₂, forms a disulfide bridge with the stress kinase Sty1, thereby transducing the oxidant signal to the transcription factor Atf1 (333, 497).

Open in a separate window

FIG. 2.

Sensing and transducing the H₂O₂ signal by the Orp/AP1-like transcription factor from yeast (Yap) system in yeast. In this system the glutathione peroxidase (GPx)-type peroxidase Orp1 act as sensor for H₂O₂ and becomes oxidized at Cys³⁶ to a sulfenic acid. The sulfenic acid then forms a heterodimeric disulfide bridge with Cys⁵⁹⁸ in Yap1, which in turn is resolved by two more thiol/disulfide exchange reactions (not shown), ultimately resulting in oxidized Yap1 (Yap1S₂) and reduced Orp1. Oxidized Yap1 can bind to the respective enhancer elements in the promoter of target genes, which protect against oxidative challenge. This way Yap1 acts as the transducer of the H₂O₂ signal. An alternative reaction of Orp1 sulfenic acid leads to an intramolecular disulfide bridge (Orp1S₂) that is reduced by thioredoxin (Trx). In this role of a Trx peroxidase Orp1 acts as a modulator of this system. Trx also terminates signaling by reducing Yap1S₂.

A case of H₂O₂ sensing by a Prx has also been demonstrated in Kinetoplastida. In these unicellular parasites a universal minicircle sequence binding protein (UMCSBP), which is a Zn-finger protein, binds to the peculiar kinetoplast DNA in a Zn- and redox-dependent manner, thereby regulating replication [reviewed in ref. (452)]. DNA binding of UMCSBP is abrogated by H₂O₂, which is sensed by the Prx-type tryparedoxin peroxidase with rate constants that, depending on species, range from 10⁵ to 10⁷ M⁻¹ s⁻¹ (489). The peroxidase directly transduces the oxidation equivalents to CxxC motifs of the Zn-finger domain. The reductive reactivation of UMCSBP depends on the trypanothione redox cascade comprising trypanothione reductase, trypanothione, and tryparedoxin, the latter being a thioredoxin-related protein (TRP) specialized for protein disulfide reduction (349).

Analogous ROOH sensing by peroxidases in mammals have not yet been reported. However, the idea that at least some of the GPx- or Prx-type peroxidases could similarly act as sensors for H₂O₂, alkyl hydroperoxides, and peroxynitrite in mammals is intriguing in several respects. It would comply with the observation that redox signaling also takes place at physiological H₂O₂ fluxes; it could explain why in some cases Prxs improve rather than inhibit signaling (75, 238), and it finally might help to explain oxidative modification of proteins with SH group reactivities that are simply not competitive enough to occur without enzymatic support (134). Mechanistically, transducing mechanisms analogous to those worked out for peroxidases of yeasts and Kinetoplastida would be straight forward for all 2-Cys Prxs. Upon oxidation of their peroxidatic cysteine C_P to a sulfenic acid, they can either form an internal disulfide bridge with their resolving cysteine C_R, which in the ROOH removing function is then reduced by a redoxin, or an intermolecular disulfide bridge with another protein, thus acting as thiol-modifying agent with regulatory consequences. Examples of Prxs found disulfide-linked to other proteins are abundant (93, 187, 333), but also GPx-type peroxidases devoid of a C_R can modify protein thiols, as has been demonstrated for mammalian GPx4. In shortage of GSH, the oxidized active site selenium (U_P) can selenylate proteins, a process that so far has not been implicated in transcriptional regulation, but is pivotal to the differentiation of spermatids into spermatozoa (121, 318, 428, 493) (Fig. 3).

Open in a separate window

FIG. 3.

Possible routes of hydroperoxide sensing by thiols or selenols. (A) General reaction scheme of GPx-catalyzed reduction of hydroperoxides. The selenol is oxidized by ROOH to a selenenic acid, which is step-wise reduced back to selenol by two molecules GSH. A role of selenols in ROOH sensing has so far not been reported. (B) The selenenic acid in GPx4 has been shown to react not only with GSH but also with thiol groups of other proteins. The enzyme, thus, transduces a signal to another protein target and acts as a thiol modifying agent. This mechanism is the molecular basis for the unique transformation of GPx4 into a structural protein during sperm maturation (493). Theoretically, the reaction sequence could be used in redox signaling with GPx4 as sensor and Se-S linked heterodimers as transducers. The sensor-reduced GPx4 could be regenerated by GSH (as in A) or by SH groups of the same or another protein (lower line). (C) 2-Cys peroxiredoxins (Prxs) react with a hydroperoxide at the peroxidatic cysteine (S_P) to the sulfenic acid. Oxidized S_P can then form an intra- or intermolecular S-S bridge with the resolving cysteine (S_R). This bridge is typically reduced by a redoxin (Trx, TXN, or Grx). The oxidized S_P can, however, also react with another protein, thus acting as a thiol modifying agent, as demonstrated in the redox control of yeasts (333). (D) Certain cysteines in protein phosphatases (PPs) react with a hydroperoxide to form a sulfenic acid. This either reacts with a second cysteine to form an intramolecular disulfide bridge in analogy to (C). Alternatively (E), the sulfenic acid reacts with a nitrogen in the peptide bond to a sulfenyl amide. Both forms can be reduced by GSH and Grx. The oxidized enzymes are typically inhibited. As discussed in section II.D.3, the known reaction constants for the reaction of ROOH with thiols in phosphatases are low. Therefore, the possibility that the oxidative signal is first sensed by a GPx- or Prx-type peroxidase and then transduced to a PP, as shown in (B) or (C), respectively, merits consideration.

Prxs can, however, also sense ROOH and transduce the signal in a more indirect way, just by means of their conventional role in removing hydroperoxides. Most of the mammalian Prxs, as well as the nonmammalian 2-Cys Prxs and many nonmammalian GPxs, use Trx or related redoxins as reductants. Under high peroxide flux, they thereby reduce the steady-state of reduced redoxins that are often used as terminators of transcriptional gene activation or interrupters of signaling cascades. Inactivating reduction of OxyR (100) or Yap1 (86, 333) by glutaredoxin (Grx) or Trx, respectively, and reduction of GSH mixed disulfides by Grx (250, 292) may suffice as examples. Further, oxidation of a redoxin may promote signaling due to reversal of inhibition, as has first been shown for the apoptosis signal-regulating kinase-1 (ASK-1), which is blocked by noncovalently bound reduced Trx, but not by the oxidized one (415). Similar functions are ascribed to nucleoredoxin (Nrx) in the wingless and Int 1 (Wnt) pathway (139), and to the dynein light chain 8 (LC8)/Trx-like 14 protein couple in the NF-κB system (227) (see section IV and Fig. 4). Although metabolic regulation by Trx and redoxins, in general, has for long been a widely accepted concept (131), it appears to be still persistently overlooked that oxidation of redoxins in vivo does not likely occur spontaneously, but is catalyzed by Prxs, which came to stage much later (399). Although the pK_a of the exposed cysteine in their CxxC motif is quite low, their rate constants for oxidation by ROOH is usually two orders of magnitude smaller than those of the most abundant Prxs that efficiently oxidize redoxins (489, 490). It therefore appears not too risky to consider most, if not all, regulatory events that are attributed to Trx oxidation to be downstream events of hydroperoxide sensing by a thiol peroxidase, a Prx being the more likely candidate in mammals (123, 134).

Open in a separate window

FIG. 4.

Redoxins in sensing mechanisms. (A) Reduced Trx has been demonstrated to inhibit apoptosis signaling by binding (and inhibiting) apoptosis signal-regulating kinase-1 (ASK-1). Oxidation of Trx relieves the block (415). Since redoxins are excellent substrates of Prxs, a hypothetical Prx is inserted as real sensor of the system as in (B–D). (B) Nucleoredoxin (Nrx) in the reduced state binds to dishevelled (Dvl), a phosphoprotein that transduces wingless and Int 1 (Wnt) signals from the Frizzled receptor. Upon oxidation, Nrx is released and Dvl can associate with the β-catenin degradadion complex, where it inhibits glycogen synthase kinase-3β (GSK3β) and phosphorylation, ubiquitination, and degradation of β-catenin. β-Catenin moves into the nucleus and activates genes required for proliferation (138). (C) Reduced Nrx associates with flightless-I (Fli-I) and myeloid differentiation factor 88 (MyD88), thereby preventing recruitment of MyD88 to the TLR4 after lipopolysaccharide stimulation (171). This way unnecessary stimulation is avoided. After removal of Nrx, MyD88 can be recruited, and the TLR mediate its signals. Although in this system oxidative processes have not yet been studied, the requirement for oxidants in TLR stimulation may be based in the need to oxidatively remove Nrx from the Fli-I/MyD88 complex, as in (B). (D) Reduced dynein light chain (LC8) binds to IκB and prevents its degradation. Upon oxidation to an intermolecular disulfide LC8 dissociates and gives the way free for IκB degradation and activation of nuclear transcription factor of bone marrow-derived lymphocytes (NF-κB). The oxidation is reversed by the Trx homolog thioredoxin-related protein-14 (TRP14), which is reduced by cytosolic thioredoxin reductase (TrxR1). In this system either LC9 itself is the sensor or is oxidized by an upstream sensing thiol peroxidase (224).

Finally, a role of Prxs in sensing ROOH may be seen in their ability to switch from a peroxidase to a chaperone, as was first demonstrated in yeast (217, 333) but has also been implicated for mammalian Prx1 (403). The chaperone function of Prxs itself appears to be independent from their peroxidase activity (403), as it prevails in the peroxidatically inactive high-molecular-weight forms having their C_P oxidized to a sulfinic acid. The functional switch to the chaperone, however, is initiated via fast ROOH sensing by C_P. Once transformed into a chaperone, the Prx may affect many regulatory pathways, enhanced cytokine expression through Toll-like receptor-4 (TLR4)-mediated NF-κB activation being a revealing example (403).

2. Transcription factors as sensors for hydroperoxides

Certainly, also proteins that are not annotated as peroxidases may sense ROOH through thiol oxidation, if their cysteine residues are activated by neighboring basic residues or other mechanisms. A well-documented example is the bacterial transcription factor OxyR (100). Its Cys199 may be considered kind of C_P, as it is directly attacked by H₂O₂ with a bimolecular rate constant of ∼10⁵ M⁻¹ s⁻¹ (11), which comes close to those of peroxidases. The initial oxidation of Cys199, like in the thiol peroxidases, is followed by a disulfide formation that facilitates oxidation of the remaining cysteines to the fully oxidized active transcription factor, whereas the inactivation is achieved by Grx-catalyzed reduction by GSH (Fig. 1).

Similarly, direct sensing of organic hydroperoxides has been implicated for organic hydroperoxide resistance repressor (OhrR), a repressor of the Ohr gene found in many bacteria (100). Although OhrR is phylogenetically unrelated to any of the peroxidase families, it shares the mechanism with atypical 2-cysteine Prxs (281): it is oxidized by ROOH (not by H₂O₂) at its C_P, which is the arginine-coordinated Cys60 (Pseudomonas aeroginosa sequence). The oxidation of Cys60 to a sulfenic acid, which suffices for de-repression, leads to formation of an intramolecular disulfide bond with Cys124, which can be reduced by dithiothreitol, the natural reductant being unknown.

Any direct interaction of a hydroperoxide or any other oxidant signal with a mammalian transcription factor has so far not been convincingly demonstrated.

3. PPs and PKs as potential sensors

Phosphory-lation and de-phosphorylation are key events in the regulation of enzyme activity in cellular signaling and certainly also dominate mammalian transcription factor activation. In most of the cases, the phosphorylated kinases and phosphatases represent the active modifications, whereas de-phosphorylation leads to inactivation of these enzymes. Exceptions to the rule are, for instance, glycogen synthase (GS) and glycogen synthase kinase-3β (GSK3β), which are active in the de-phosphorylated state. Protein kinases and, more importantly, PPs are widely assumed to sense H₂O₂, other oxidants, or S-alkylants, and it may be guessed that most, if not all, phosphorylation/de-phosphorylation balances are redox-controlled via cysteine modification (Fig. 3D, E).

In receptor protein kinases, a highly conserved MxxCW motif is considered as the general switch that, upon cysteine oxidation, starts the tyrosine phosphorylation required for catalytic activity (339). Oxidative activation is also observed in the various forms of protein kinase C (PKC) that are characterized by Zn-fingers in their regulatory domain, in which two Zn⁺⁺ ions are tetrahedrically coordinated with six conserved cysteine and two histidine residues (77, 194, 258). The Zn-finger is assumed to work as kind of hinge wherein the Zn⁺⁺ functions as a linchpin. Oxidation of the Zn-coordinated cysteines was shown to cause Zn⁺⁺ release (259) and is assumed to favor the active kinase conformation, phosphorylation, binding to Ca⁺⁺, and phosphatidylserine and, thus, membrane recruitment of these enzymes (258, 309). Similarly, Zn⁺⁺ release associated with activation of kinase activity due to oxidation of Zn-coordinated cysteines was reported for c-RAF (190). Since PKCs inter alia mediate oxidative responses such as the oxidative burst of phagocytes (see section IV.D.1), the inhibitory function of Zn⁺⁺ in these enzymes may well explain the still mysterious antioxidant function of the ion, which by itself is redox-inert under physiological condition (49, 309, 382).

In contrast, PPs are more or less readily inhibited by cysteine oxidation/modification. PPs are specific for protein tyrosine phosphates, serine (Ser)/threonine (Thr) phosphates, or for both, and, accordingly, are classified into tyrosine PPs (protein tyrosine phosphatase [PTPs]), Ser/Thr PPs (PSPs), and dual specificity PPs (DSPs). Their mechanism of action differs and so does their sensitivity to oxidants.

PTPs are generally susceptible to oxidative inactivation. They are characterized by an HCx₅R motif that comprises the PTP loop that binds the phosphate groups of phosphotyrosines [(472); reviewed in ref. (67)]. The cysteine in this motif, due to its low pK_a value, is nucleophilic, which is a prerequisite for both substrate de-phosphorylation and oxidative inactivation. More than 100 PTPs of the human genome (486) also include the DSPs that share with the typical PTPs the characteristic HCx₅R motif and may, therefore, be suspected to be equally prone to oxidative inactivation. The DSP family not only de-phosphorylates tyrosine, Ser, and Thr residues, but also further comprises phosphatases acting on nonprotein substrates, such as phosphoinositol phospholipids (PIPs). Examples for the latter specificity are the Src homology-2 (SH2)-domain-containing inositide phosphatases (SHIPs) (149), which de-phosphorylate the membrane-bound PI3K-generated key signaling lipid PI(3,4,5)P₃ at position 5 to PI(3,4)P₂, and the phosphatase and tensin homologue (PTEN), which dephosphorylates the 3 position of both PI(3,4,5)P₃ and PI(3,4)P₂ (304). Sharing the active site motif with the PTP/DSP family, PTEN is a typical example of a redox-sensitive phosphatase (280).

A particularly reactive cysteine being unidentified in PSPs, this type of phosphatase can not a priori be rated as redox sensitive. However, also Ser/Thr phosphatases such as PP2A have been shown to undergo reversible oxidation. PP2A contains 10 cysteine residues in the catalytic subunit including a vicinal pair at positions 266–269, which reminds of a redoxin motif. This CxxC motif proved to be sensitive to oxidation in vitro, and its oxidation resulted in a decreased activity (130). In addition, PSPs can have metal ions in their active center, which are essential for their enzymatic function (449). Thus, apart from redox modification of cysteine residues, also an oxidation of the metal clusters appears conceivable.

Mechanistically, the oxidative inactivation of phosphatases via cysteine oxidation involves reactions known from GPx or Prx mechanisms (127). A pivotal cysteine appears to be oxidized to a sulfenic acid (Eq. 1). This unstable oxidation form typically forms a disulfide with another cysteine residue (Eq. 3), which reminds of the analogous reaction of C_P and C_R in Prx or GPx catalysis (Fig. 3). In oxidized PTEN a disulfide bridge between the catalytic Cys124 and the neighboring Cys71 was detected (275, 418). Similarly, in the DSP Cdc25 phosphatase B, a second cysteine resides in the active site, which in the oxidized form is disulfide-linked to the nucleophilic cysteine of the signature motif (51). In contrast, in PTP1B, the cysteine in the active site motif [I/V]HCxxGxxR[S/T] forms a cyclic sulfenyl-amide (419, 495) in analogy to the redox cycle of the GPx mimic ebselen (423) (Fig. 3E). As in the catalytic cycles of Prxs (378) and GPxs (487), the physiological meaning of conserving the labile oxidation state of the sulfenic acid in a more stable form is seen in the prevention of sulfur oxidation to the sulfinic or sulfonic forms that would be hard to reverse. As in the enzymatic cycles, the oxidized phosphatases are readily reactivated by physiological thiol compounds such as GSH, Trx, Grx, or other redoxins.

However, the almost generally agreed assumption that protein kinases and phosphatases may sense oxidants via cysteine oxidation is overshadowed by lack of any confirming kinetic data. None has been reported for any kinase and the few available rate constants for a direct oxidation of the critical thiols in PPs by H₂O₂ do not exceed 50 M⁻¹ s⁻¹ (31, 87, 459) and, thus, are by far too small to corroborate any physiological relevance of the process. For three redox-sensitive PTPs (PTP1, leukocyte-common antigen-related (phosphatase) [LAR], and Vaccinia H1-related (phosphatase) [VHR]) reaction rates for oxidative inactivation were insufficient (10–20 M⁻¹ s⁻¹); for three Ser/Thr phosphatases (PP2Cα, calcineurin, and lambda phosphatase) they were evidently too slow to be quantified (87). For the PTP-type Cdc25 phosphatases the rate was a bit more promising, being 15-fold higher than for PTP1B (459). In the latter case, the authors stress that the reaction of H₂O₂ with Cdc25 is 400 times faster than with GSH, which though is a poor argument, since the spontaneous reaction of GSH with H₂O₂ is physiologically irrelevant. For the relevant GPx reaction the rate constant is around 10⁸ M⁻¹ s⁻¹.

According to Forman et al. (134), the available kinetic data predict that even the nonenzymatic reaction of H₂O₂ with GSH would completely compete out the oxidation of the active site cysteine of PPs. However, since oxidation of PTP1B and other PPs, as evidenced by glutathionylation (Eqs. 1 and 2), evidently occurs under in vivo conditions (250, 405), oxidant sensing by PPs must be more complex. It could be envisaged that any of the numerous thiol peroxidases first senses ROOH and then, in analogy to the yeast GPx- and Prx-type peroxidases, forms a disulfide bridge with the target phosphatase that is cleaved by GSH, leaving the glutathionyl residue at the target protein. Grx has been implicated in protein glutathionylation but appears to rather de-glutathionylate (229). Alternatively, the rate constants, which were obtained in vitro with isolated enzymes, could be grossly misleading, as they might in vivo be substantially improved by protein/protein interaction within regulatory complexes. Human Prx6 may be taken as an example for a dramatic change in thiol reactivity due to association with another protein: it displays a significant GPx activity only when associated with a GSH-loaded GSTπ (439). Before the discrepancies between in vitro data and in vivo observations have not been convincingly explained, we should cautiously address redox-sensitive PPs and kinases as putative ROOH sensors.

4. Redox sensing by cytosolic inhibitory complexes of transcription factors

In eukaryotes, transcription factors are commonly sequestered in the cytosol in form of multicomponent complexes from which they have to be released for gene activation in the nucleus. The activation of the transcription factor requires modification of one or more components of the complex to enable release of the transcription factor and its nuclear import. Almost regularly, a redox modification of a component is followed by ubiquitination and proteasomal degradation of the same or another component (183). A typical example of this reaction scheme is the stress-signaling Nrf2/Keap1 system, wherein Nrf2 is the transcription factor and Keap1 the complex partner that prevents Nrf2 activation (see section III for details). In this particular case, the inhibitory protein Keap1 not only prevents transcription factor activity by keeping it in the cytosol, but also by targeting it for the proteasomal degradation pathway (254, 531).

Keap1, which again is a Zn-finger protein, is considered to be the redox sensor of the system. Its reactive cysteines may be oxidized to sulfenic acids, form disulfides, or be alkylated by electrophiles (Michael acceptors) such as HNE (201), other α, β-unsaturated aldehydes, or ketones (282, 410), a realm of phytochemicals (469) or other electrophilic xenobiotics (104). The consequence of cysteine modification in Keap1 is a dramatic conformational change resulting in a destabilization of the complex and prevention of Nrf2 degradation. As generally observed in Zn-finger proteins (see section II.D.3), oxidation of Zn-coordinated cysteines leads to release of Zinc and consecutive unfolding of the finger structure. The structural change of Keap1 then prevents ubiquitination and proteasomal degradation (see section III.C.2).

The ubiquitination process proceeds via the conventional scheme: a coordinated cascade of 3 enzymes, the ubiquitin-activating enzymes E1, the ubiquitin-conjugating enzymes E2, and the ubiquitin ligases E3 (Fig. 5). Of the known E3 ligases, the really interesting new gene (RING) family is the largest. The RING motif in these enzymes interacts with E2 and facilitates the transfer of ubiquitin from E2 to a lysine in the target protein (370). One subtype of the RING family is the Cullin family. Cullin-RING E3 ligases utilize 1–7 Cullin scaffolds to assemble several substrate-specific adaptors that recognize and position the target, here Nrf2, in the cullin-E3 complex for proper ubiquitination. The unmodified Keap1 serves as an adapter for the RING box protein (Rbx1)-bound Cullin-3 (Cul3)-based E3 ligase, which targets the Nrf2 within the Keap1/Nrf2 complex (80, 140, 141, 255, 372, 532). Cysteine modification in Keap1 disrupts the presentation of Nrf2 for ubiquitination.

Open in a separate window

FIG. 5.

Simplified scheme of ubiquitination of substrates in mammalian cells. In a first step ubiquitin is activated at the C-terminal glycine by the ATP-dependent formation of a thioester at a cysteine of the E1 enzyme. E1 transfers ubiquitin to a cysteine of the ubiquitin-conjugating enzyme E2 and is released from the complex. The ubiquitin-loaded E2 then forms a complex with the E3 enzyme to which the respective substrate (S) is bound. The really interesting new gene (RING) box protein-1 (Rbx1) targets the substrate for ubiquitination. Ubiquitin can then be passed either directly to a lysine of the RING-bearing E3-bound substrate (Rbx1) or, in case of the homologous to E6AP C terminus (HECT)-domain bearing E3, to another cysteine on E3 and from there to the substrate lysine (186).

The ubiquitination machinery, here briefly introduced as a downstream event of an oxidative Keap1 modification, has also been rated as redox sensitive. Both the ubiquitin-activating enzyme E1 and the ubiqutin-conjugating E2 were shown to be reversibly inactivated by glutathionylation due to treatment with H₂O₂ or diamide (212, 355). The ubiquitin ligases, of which about 600 different ones were identified in man, are mostly characterized by a RING finger domain, in which two Zn⁺⁺ ions are coordinated with seven cysteine and one histidine residues. Clearly, this structural element, like the Zn-finger of PKCs and Keap1, is suggestive of a site for redox regulation, an option that, however, has so far been left unexplored (88). Similarly, the redox regulation of SUMOylation, the analogous protein modification by small ubiquitin-like modifiers (SUMO), appears not to have been investigated to any significant extend (511).

Ubiquitination followed by degradation is also used for de-inhibition of transcription factors in other redox-sensitive cytosolic complexes, but the sensing step and the ubiquitinated entities as well as the regulatory principles of the systems differ substantially (Fig. 6). In the hypoxia response system it is also the transcription factor itself that is permanently ubiquitinated and degraded under normal conditions, that is, under normoxia. The redox sensor, however, is not an inhibitory protein such as Keap1 (Fig. 6A), but an enzyme: the proline hydroxylase that hydroxylates the transcription factor HIF-1α and targets it for ubiquitination (Fig. 6B). Similarly, the transcription factor β-catenin in the Wnt pathway is steadily ubiquitinated and degraded under resting conditions. In this context Nrx is assumed to be the redox sensor; in the reduced state it blocks the Wnt pathway by binding to the upstream adapter protein disheveled (Dvl), while allowing nuclear import of β-catenin upon oxidation of its CxxC motif (138, 139) (Fig. 6C). In the canonical NF-κB system, the inhibitory NF-κB-binding protein IκB has to be ubiquitinated and degraded to allow nuclear import of p50/p65. IκB is targeted for ubiquitinating by phosphorylation through IκB kinase (IKK) (Fig. 6D). The redox sensor of the system might be the LC8, which in its reduced states, is bound to IκB and protects IκB from attack by IKK, but allows IκB phosphorylation and degradation upon oxidative dimerization via intersubunit disulfide formation (Fig. 4D and section IV.D.3) (227). For lipopolysaccharide (LPS)-induced NF-κB activation via TLR 4, again Nrx has been implicated as negatively regulating sensor by binding to Fli-1 (flightless), thereby preventing recruitment of the essential adaptor myeloid differentiation factor 88 (MyD88) to the receptor (171).

Open in a separate window

FIG. 6.

Distinct roles of redox-dependent ubiquitination in the regulation of transcription factor activity. (A) In the Nrf2/kelch-like ECH-associated protein-1 (Keap1) system, oxidant sensing by Keap1 terminates the ubiquintination and degradation of Nrf2 and allows its nuclear translocation and transcriptional activation of target genes. (B) In the hypoxia-inducible factor 1 (HIF) system, under normoxic conditions HIF-1α is hydroxylated by HIF prolyl hydroxylase (HPH), ubiquitinated, and degraded. In hypoxia HPH is inactive, ubiquitination is prevented, and HIF-1α can translocate into the nucleus and activate gene expression. (C) In the Wnt/β-catenin pathway, GSK3β phosphorylates β-catenin, thereby facilitating its ubiquitination and degradation and preventing β-catenin-mediated gene expression. Dvl is captured by Nrx. Upon oxidation of Nrx, Dvl is released to inhibit GSK3β activity. β-Catenin is no longer phosphorylated and is translocated into the nucleus. (D) In the NF-κB system, LC8 is associated to IκB, thereby preventing its degradation and release of NF-κB. Nuclear translocation of NF-κB can take place after oxidative modification and release of LC8 and degradation of the inhibitor IκB.

It is needless to state that kinetic data for component interactions in these complex systems do not exist at all and will hard to be obtained. The mentioned redox-sensitive proteins, which, in principle, can directly be oxidized by any ROOH, might therefore not be the real sensors, but, instead, be oxidized by an upstream thiol peroxidase sensing the oxidant.

5. Sensing by selenocysteine-containing proteins?

A sensor's competence to sense ROOH or an alkylant selectively would certainly be enhanced if its sensing sulfur were replaced by the super-sulfur selenium (120). Many selenoproteins display signatures of redox proteins (146) and most of those with an established enzymatic function are indeed oxidoreductases with selenocysteine (Sec) as pivotal catalytic entity (122). As a rule, these seleno-enzymes are substantially faster than their homologs working with sulfur catalysis (33). Simply by chemical reasoning, therefore, selenoproteins would have an optimum chance to beat thiol-based redox sensors in terms of speed and specificity.

Circumstantial evidence has for long suggested that selenoproteins are relevant to transcriptional activation. In selenium-deficient animals a huge number of nonselenoproteins, now overwhelmingly known as Nrf2 targets, were found elevated since the late seventies (78, 79, 270, 395–397). The unexpected phenomenon could not be convincingly explained by GPx1 deficiency alone (396). However, at least one of the selenoproteins had to be involved in Nrf2 activation, since a complete loss of all selenoproteins by an organ-specific removal of the gene encoding selenocysteine tRNA (Trsp) (56) resulted in the induction of GST isoforms in liver, of GSTP1 and NADPH quinone oxidoreductase 1 (NQO1) in liver and macrophages (471) and of heme oxygenase-1 (HO-1) in liver (446). Also a moderate selenium deficiency in which only selenoprotein W, GPx1, and selenoprotein H and M were markedly decreased (249) led to an increase in NQO1, GSTs, sulfotransferases, and UDP-glucuronyl transferases (UGT) as well as HO-1, Prx1, sulfiredoxin-1 (Srx1), and γ-glutamyl-cysteine synthase (335). A firm link between selenium deficiency and Nrf2 activation was finally established by Burk et al. by demonstrating a strong increase in electrophile-responsive element (EpRE)-driven reporter gene activity in livers of selenium-deficient wild-type but not in Nrf2^−/− mice (52).

Certainly, these findings do not strongly corroborate a role of selenoproteins as sensors, as they may be equally explained as adaptive response to impaired hydroperoxide metabolism. The hypothesis of a direct sensing function has not yet been established for any of the selenoproteins, but remains attractive from a chemical point of view. H₂O₂-dependent selenylation of protein thiols by GPx4 (see section II.D.1) is mechanistically equivalent to H₂O₂ sensing by thiol peroxidases and reveals that selenoproteins can in principle act as combined redox sensors and transducers (121, 123). After all, the mammalian Trx reductases, which all are selenoproteins, may be indirectly involved in ROOH sensing by regenerating Trxs as key redox regulators and reducing substrates of Prxs. These aspects certainly deserve more interest when searching for the biological roles for the more than 25 distinct mammalian selenoproteins (167).

2. Keap1 as primary redox sensor of Nrf2 signaling

Keap1, sometimes also called INrf2 for inhibitor of Nrf2 (348), contains three characteristic domains, the broad complex/tramtrac/bric-a-brac (BTB) domain, the intervening region (IVR), and the kelch domain, also known as double glycine repeat (DGR) domain (3). The BTB domain binds Cul3 (see section II.D.4) (141) and is required for homodimerization (538), whereas the IVR contains cysteine residues that are in part Zn-coordinated (98) and are considered relevant to regulation (see below) and a nuclear export signal (NES) (498). The kelch/DGR domain is required for the interaction of Keap1 with the actin cytoskeleton (234, 235) and/or myosin VIIa (499), which immobilizes Keap1 in the cytoplasm. The kelch/DGR domain is also essential for binding of Nrf2, which interacts with Keap1 with its amino-terminal Nrf2-ECH homology-2 (Neh2) domain (208).

The structure of the kelch/DGR domain of Keap1 has recently been resolved (297, 362), which sheds new light on the Keap1/Nrf2 interaction (484) (Fig. 7). According to the hinge and latch model there are two distinct binding motifs in the Neh2 domain of Nrf2, each binding one molecule of a Keap1 homodimer (two-sites binding) (325, 484). The high affinity binding ETGE motif functions as a hinge to pin down Nrf2 to Keap1. The low affinity DLG motif functions as a latch. The link between the hinge and the latch motif is the lysine-rich central α-helix of Nrf2. In this helix six of seven lysine residues point to the same site (484). The Neh2 domain is locked in a position suitable for ubiquitination of the Lys residues (325, 485) which, hence, act as ubiquitin acceptors (532). The hinge and latch model (484), thus, complies with a substrate-presentation mechanism: presentation of Nrf2 for ubiquitination and subsequent proteasomal degradation.

Open in a separate window

FIG. 7.

Model for redox-regulated Nrf2 activation based on conformational change of Keap1. In the resting state (A), Keap1 forms a homodimer via the broad complex/tramtrac/bric-a-brac (BTB) domains and associates to Nrf2 with its kelch domains. The Nrf2-ECH homology-2 (Neh2) domain of Nrf2 contains two Keap1 binding motifs with different affinity to the kelch domains, with ETGE representing the amino acid sequence with the higher affinity and DLG, the sequence with a much lower affinity. This two-site binding locks the central α-helix of the Neh2 domain of Nrf2 with is seven lysine residues (7 Lys) into a position suitable for ubiquitination. The intervening regions (IVR) contain the most reactive cysteine residues (Cys273 and 288) and build the adapter region for the Cullin-3 (Cul3)-based E3 ligase system. Nrf2 is polyubiquitinated and degraded by the proteasome. Under stressed conditions (B), oxidative/electrophilic modification of Cys272 and 288 induces a conformational change of Keap1 resulting in the dissociation from the weaker binding DLG motif in Nrf2, whereas binding to the ETGE motif remains [hinge and latch model as proposed by (484)]. P21 interacts with the DLG motif and supports dissociation of Nrf2 from this site. In contrast, modification of Cys151 in the BTB domain does not change the conformation of Keap1 but rather disrupts BTB Cul3 interaction. By both modifications ubiquitin ligase activity is lost and newly synthesized and released Nrf2 can move into the nucleus.

The hinge and latch model can also explain how thiol-modification disturbs the presentation of Nrf2 for ubiquitination. Murine and rat Keap1 contain 25 cysteine residues, whereas human Keap1 has 27 (97, 98). Nine of these are predicted to be reactive due to their location adjacent to basic amino acids. The cysteines 257, 273, 288, and 297 lie in the IVR and, by MS analysis of tryptic peptides, were demonstrated to be most sensitive to alkylation by dexamethason mesylate in vitro (97). Further, optical monitoring of the reaction of Keap1 with dipyridyl disulfide revealed a sequential reaction of the cysteine residues, likely starting with an S-thiylation of the most reactive one and followed by thiol/disulfide exchange to finally form internal disulfides within Keap1. Competition experiments with a representative set of Nrf2 activators demonstrated that Keap1 modification capacity of these heterogeneous compounds roughly correlated with their in vivo activity (97). By these and many other thiol modification studies (256, 287) the hypothesis of Keap1 being the redox sensor of the Nrf2 system was corroborated. In vitro and in vivo mutation studies finally revealed that Cys273 and 288, which are the Zn-coordinated ones, are crucial for the response of Nrf2 to activators (282, 504, 531), whereas mutation of Cys151 mediates stressor-induced dissociation of Keap1 from Cul3 (389, 521) (Fig. 7).

However, the original hypothesis that the primary sensing event, that is, modification of critical cysteine residues, leads to instant release of Nrf2 has been revised: the hinge and latch model predicts that thiol modification of Keap1 changes its conformation in a way that only binding of Nrf2 through its DLG motif, the latch, is reversed, whereas Nrf2 remains attached to Keap1 by its hinge, the ETGE motif. Thereby, the presentation of Nrf2 for ubiquitination is impaired. Accordingly, the unmodified critical cysteines 273 and 288 are crucial for ubiquitination of Nrf2 (531) and the degradation of Nrf2 in the unstressed situation (254). Upon challenge, the still hinge-bound Nrf2 is stabilized by preventing ubiquitination and may finally be released by downstream events (see section III.C.3). Moreover, continuously newly synthesized Nrf2 can bypass the still Nrf2-loaded Keap1 for transduction (254).

How thiol modification leads to the conformational change is likely explained by the zinc finger nature of Keap1 (95). Zinc is bound to Keap1 with an impressive K_ass of 10¹¹ M⁻¹. Zn binding to Keap1 evidently involves the critical cysteines, as thiylation of the latter by dipyridyl disulfide is accompanied by a stoichiometric release of Zn; by a Cys to Ala mutation, the K_ass is dramatically decreased, and profound conformational changes are associated with Zn release, as evidenced by shifts in tryptophan fluorescence or depression of fluorescence of a hydrophobicity probe. The Zn-stabilized structure of Keap1, thus, is disturbed by derivatization of Zn-binding cysteines and equally by Zn chelation or Zn shortage, as has been similarly discussed for other redox-sensitive zinc proteins (82, 258, 309). However, it can also be envisaged that cysteine-coordinated Zn in Keap1 contributes to the pronounced reactivity of the cysteines, thereby facilitating Nrf2 activation. The pivotal role of zinc in Keap1 function might provide another lead to explain its antioxidant effect (49) (see also section II.D.3).

Finally, the realm of Nrf2 activators raises the question how their chemical heterogeneity complies with the expectation that a sensor should sense specifically. Emerging evidence indeed reveals that groups of activators act on different sets of cysteines in Keap1, which may be translated into a specific biological effect. From these observations the term cysteine code was created and breaking this code for each Nrf2 activator will help to understand the various effects resulting from Nrf2 activation (256, 521).

3. Downstream signaling events

The essence of the events following sensing by Keap1 is that free cytosolic Nrf2 has to reach its target EpRE in the nucleus. The nuclear transport of Nrf2 is mediated by the conventional set of nuclear importin and exportin proteins (156). Typical nuclear localization signals (NLS) or NES of cargo proteins to be recognized by the import and export machinery are found in the sequence of Nrf2: two NES, one located in the leucine zipper dimerization (Zip) domain and a second one in the transactivation domain (TAD) (215, 289), and three NLS (478).

The basic requirements for nuclear traffic being evident, we first have to elaborate on the questions how free Nrf2 is generated and primed for nuclear import (Fig. 8). As mentioned in the previous section, cysteine modification of Keap1 by itself is not considered sufficient to release Nrf2 from the cytosolic complex. The view favored at present therefore is that the Nrf2 that is synthesized de novo bypasses cytosolic Keap1 when the latter has been modified and hence loaded with stabilized Nrf2, whereas Nrf2 meeting Keap1 under resting conditions has a poor chance to escape from degradation (287). However, since permanent wasting of Nrf2 by the unchallenged organism appears uneconomic, mechanisms to release Nrf2 from the cytosolic complex merit consideration and do indeed exist. (i) PKC phosphorylates Nrf2 at Ser40 (192), the isoforms later being identified as PKCδ (283) and PKCι (350). Ser40 is located between the DLG and the ETGE motifs, and phosphorylation inhibits binding to Keap1 and simultaneously facilitates nuclear import (347, 348). It appears, however, to be unsettled if this phosphorylation is also achieved within the Keap1/Nrf2 complex (35, 192, 350). (ii) The CDK inhibitor p21 binds to stabilized Nrf2 at Keap1 and possibly facilitates Nrf2 release (63). (iii) Prothymosin α (239, 361), and (iv) the sequestosome 1 (SQSTM1; p62) (269) interact with Keap1 in a way that displacement of Nrf2 appears unavoidable. SQSTM1 was further shown to lower Keap1 levels by increasing its rate of degradation (76). Accordingly, overexpression of SQSTM1 results in transcriptional activation of Nrf2 target genes (76, 257). Collectively, these findings suggest that also the Keap1-bound Nrf2 that has been saved from ubiquitination can be made available for nuclear import (482). It has further also been proposed that Nrf2-loaded Keap1 enters the nucleus (498).

Open in a separate window

FIG. 8.

Nuclear translocation of Nrf2. Activation of Nrf2 does not necessarily require activation of a specific surface receptor. Stimuli can be exogenous or endogenously produced. In response to oxidative/electrophilic stress, the interaction of Keap1 and Nrf2 is disturbed and Nrf2 is no longer degraded. Nrf2, either released from Keap1 or newly synthesized, enters the nucleus. For translocation, Nrf2 has to be phosphorylated at Ser40. The phosphorylation is achieved by PKCδ or ι and/or other kinases (see text). Phosphorylated Nrf2 enters the nucleus, forms a complex with the basic leucine zipper (bZIP) factors CREB-binding protein (CBP)/p300, and activates gene expression by binding to the electrophile responsive element (EpRE). Phosphorylation at Tyr568 by the tyrosine kinase Fyn leads to nuclear export of Nrf2 and terminates the signal. It is not known whether Nrf2 has first to be de-phosphorylated at Ser40 or whether the bis-phosphate is exported. Exported Nrf2 can re-associate with Keap1.

Thus, nuclear import of Nrf2 is initiated by the key oxidative modification of Keap1 and further facilitated, inter alia, by PKC-dependent phosphorylation, which is also considered to be favored by thiol oxidation (Fig. 8; section II.D.3). Moreover, Nrf2 itself and, hence, its nuclear localization is redox sensitive (Fig. 9). Its activity can also be enhanced by inhibiting the nuclear export of Nrf2. The NES of Nrf2 are recognized by chromosome region maintenance (Crm1)/exportin, but in the nucleus are largely masked, probably by heterodimerization of Nrf2 with musculo-aponeurotic fibrosarcoma (Maf) proteins that retain Nrf2 in the nucleus. In consequence, the import/export balance is in favor of import even under basal conditions, which accounts for the constitutive Nrf2 activity that guarantees the basal expression of stress response genes. Whereas the NES in the bZIP domain is redox insensitive (289), oxidation of Cys183 in the NES in the TAD of human Nrf2 is discussed to inhibit the access and binding of Crm1 (288). Under oxidative stress conditions, therefore, the function of the redox-sensitive NES in the TAD is impaired and the overall weight of export signals is further decreased (287). Similarly, an oxidative modification within an NES of nuclear Keap1 can retain Nrf2 in the nucleus (498). Irrespective of the mechanism, the net effect is a nuclear accumulation of Nrf2. Cysteine modification, depending on its chemistry, may be reversible, for example, by GSH or a redoxin, and reversal of nuclear localization of Nrf2 by Trx has indeed been reported (162). Cys506 is critical for binding of Nrf2 to its responsive element and the interaction with the cofactor CREB-binding protein (CBP)/p300 (34). Apart from Cys506 also Cys119 and 235 (numbering in the mouse Nrf2) have been found to have similar functions (176). Hence, re-establishing reductive conditions in the nucleus deserves attention not only as required for DNA binding but also as shut-off mechanism of the system (see section III.C.5).

Open in a separate window

FIG. 9.

Organization of Nrf2. The Nrf2 protein comprises six domains (Neh1–6) with different functionalities. Neh2: Keap1 binding site; Neh4 and 5: transactivation (TA) site harbors the nuclear export signal (NES) with the redox-sensitive C183, and is required for binding of Nrf2 to its responsive element and interaction with the coactivator CBP/p300; Neh6: linker, site responsible for Keap1-independent nuclear degradation under oxidative stress (324); Neh1: contains the Cap ‘n’ Collar region (CNC), the redox-insensitive NES, the nuclear localization signal (NLS) with the redox-sensitive C506, and the bZIP motif; Neh3: the C-terminal domain contains Y568 and is required for transactivation (345). S40 is the phosphorylation site for PKCδ and ι; Y568, the phosphorylation site for Fyn. Arrows mark putative sites for casein kinases. Two novel NLSs spanning amino acids 42–53 and 587–593 identified in the mouse sequences are not included for sake of clarity (478). For numbering of different species see ref. (176).

Further, nuclear export of Nrf2 is regulated by phosphorylates (Fig. 7). The Fyn kinase phosphorylates Nrf2 at Tyr568 in the nucleus which leads to the strengthening of the interaction with Crm1 and to an enhanced nuclear export of Nrf2 (213, 348). A translocation of Fyn into the nucleus is observed several hours after Nrf2 activation and requires the phosphorylation of Fyn at a so far unknown Thr residue by GSK3β. For this activity GSK3β has to be phosphorylated at Tyr216, which in contrast to the phosphorylation at Ser9 by Akt, which inactivates GSK3β (327), keeps GSK3β in the active state. The Tyr216-specific kinase is still unknown, but it is activated by H₂O₂ (214). This way GSK3β may contribute to the termination of the Nrf2 signal.

4. Nrf2-mediated transduction events

In the nucleus Nrf2 dimerizes with small Maf proteins (207) or other bZIP transcriptions factor such as c-Jun or activation transcription factor-4 (ATF4). The heterodimer binds to the Nrf2-responsive element EpRE. The Nrf2/bZIP complex then recruits the CBP and p300 (536) and initiates transcription of target genes (Fig. 8). Formation of a heterodimer with c-Fos or c-Fos-related antigen-1 (Fra-1) inhibits Nrf2 activity due to binding to the AP-1 site within the EpRE. Also, overexpression of small Maf proteins leading to Maf homodimers and possibly unproductive Maf-Nrf2 heterodimers inhibited Nrf2 activity (90). The Nrf2 gene itself carries an EpRE and is transcriptionally stimulated by Nrf2 activators (263).

Binding of Nrf2 to EpRE is competed out by the BTB and CNC homology-1 (Bach1) transcription factor (91). Oxidative conditions inactivate Bach1 and allow Nrf2 binding to EpRE (203). Inactivation of Bach1 is mediated by a redox-sensitive phosphorylation at Tyr486 by an unknown kinase, which leads to rapid export of Bach1 (243).

5. Shut-off mechanisms

Apart from nuclear export of Nrf2 (see above, section III.C.3), termination of Nrf2-mediated transcription may be achieved by various mechanisms. (i) Also the nucleus contains Keap1, whose function is not fully clarified, but it may be envisaged that Keap1 also targets the nuclear Nrf2 for degradation by the nuclear proteasome (348). (ii) Polymerization of the actin skeleton is considered to be essential for retaining Nrf2 in the cytosol, thus preventing Nrf2 activity. PI3K inter alia regulates the response to oxidants by de-polymerization of actin, which is considered to allow Nrf2 to escape from Keap1 and to enter the nucleus together with actin (234). Re-polymerization of actin allows Nrf2 to exit the nucleus. In this context it might be revealing that glutathionylation of actin leads to depolymerization and is reversed by Grx (267, 506). (iii) Nrf2 activates transcription of its own cytosolic inhibitors such as Cul3, Rbx1, and Keap1 (242). (iv) The most efficient shut-off is certainly achieved by means of the myriad of enzymes simply eliminating the system's signals or preventing their formation.

Notably, Nrf2 targets such as Trx and Trx reductase, Srx and γ-glutamyl-cysteine synthetase synergize in improving the capacity of the entire Prx- and GPx-mediated hydroperoxide metabolism and thus eliminate oxidant signals and prevent the formation of damage signals arising from oxidative stress (125, 200). The scenario is complemented by induction of particular peroxidases such as Prx1 (200) and GPx2 (21). Similarly, the induction of many GSTs prevents activation of the system by eliminating the nonoxidant alkylants (175). Apart from eliminating Nrf2-activating signals, however, the expression of antioxidant enzymes may also interfere directly with ongoing Nrf2 transduction: murine liver txnrd1^−/− cells that lack the selenoprotein Trx reductase-1 do not display any gross changes in the cellular redox homoeostasis, likely because GPxs can easily substitute for the Trx-dependent Prxs under most conditions. However, the txnrd1^−/− cells displayed a pronounced persistence of Nrf2 in the nucleus. This, for the first time, points to the requirement of reduced Trx for switching off Nrf2 transduction (470) and provides a mosaic stone possibly explaining the role of selenium in Nrf2 signaling (see section II.D.5).

1. The source of the oxidants

In most of the schemes illustrating redox-regulated NF-κB activation, activating ROS, like a deus ex machina, show up in the center of an exploding star, leaving open the problems of the chemical nature and biological source of the magic activator. Intriguingly, however, a list of compounds known to induce O₂^•− formation in phagocytes already in 1985 (126) almost coincides with the one of NF-κB enhancers published by Schreck et al. in 1991 (438). It comprises a variety of PAMPs, antibodies, calcium ionophores, phorbolesters, substrates/products of LOX, and lipid mediators of inflammation. The two lists become practically congruent just by adding the pro-inflammatory cytokines IL-1 and TNFα to the older one. It incidentally was this pronounced similarity of activators of NOX and NF-κB enhancers that led to the concept of NF-κB activation by oxidants (128). It now corroborates that PAMPs or pro-inflammatory cytokines, when activating their respective receptors, simultaneously activate NOX and/or LOX enzymes.

The ability of calcium and PKC activators such as phorbol esters to trigger an oxidative burst in phagocytes had for long revealed a role of PKC-type enzymes in linking receptor occupation to the activation of NOX. Members of the PKC family are known to regulate many cellular processes (258) (see sections II.D.3 and III.C.3) and have meanwhile been shown to activate NOX- and LOX-type enzymes (see section IV.D.1), and to also interfere with the NF-κB system at other sites (see section IV.D.2). As mentioned in section II.D.3, they are characterized by redox-sensitive zinc finger domains. While they are activated by oxidation of the cysteine residues that are coordinated to zinc (152, 153), they may be inactivated by oxidation or alkylation of their catalytic cysteines (154). It, thus, appears conceivable that they amplify an NOX- or LOX-mediated oxidative response but contribute to its termination under stronger oxidative conditions.

Involvement of NOX has been convincingly shown in p47^phox−/− mice. p47^phox is an indispensable component of the phagocytic NOX complex that further contains p22^phox, p40^phox, the GTP-binding protein Rac-1, and the flavo-cytochrome gp91^phox as the O₂^•− producing NOX2 (14). Accordingly, in the lungs of mice deficient in this essential activator the activation of NF-κB elicited by pathogens or LPS was impaired (253, 414). Similarly, in human aortic endothelial cells TLR4-mediated superoxide production and NF-κB activation were shown to depend on the homologous NOX4. A knockdown of NOX4 inhibited IκB degradation and binding of p65 to DNA (365). Mechanistically, an interaction of the C-terminal region of NOX4 with the Toll/IL-1 receptor (TIR) domain of TLR4 was discussed which now is known to be linked by MyD88 (290). Also, the source of IL-1-induced O₂^•−/H₂O₂ production in MCF7 cells was shown to be an NOX. The H₂O₂ thereby produced was reported to activate NIK-mediated phosphorylation on IκB probably by the inhibition of a phosphatase, since ocadaic acid mimicked the IL-1 effect (284).

The precise link between ligand occupation of TLRs and the activation of NOX systems is not yet entirely clear. In the possibly analogous growth factor signaling the activation of NOX1 is achieved by sequential activation of phosphatidyl-inositol-3-kinase (PI3K), formation of GTP-loaded Rac-1, and recruitment of phagocytic oxidase (phox) subunits to the membrane to assemble the active NOX complex (366). The PI3K pathway appears also to be generally activated through TLRs (313), whereby MyD88 links PI3K to the receptor (266) and PI3K produces the phosphatidyl inositol phosphates to form the platform where phox subunits can bind (106, 232) (Fig. 12). A similar sequence may, therefore, be envisaged for Rac-1 activation for NOX2 activation in phagocytes. The pivotal TLR-activated kinase for the phosphorylation of p47^phox (30) and p67^phox (533) in human monocytes is PKCδ, which in turn is activated by SYK and Src kinases associated with the zymosan-recognizing receptor Dectin-1, which in turn cooperates with TLR2 in the induction of inflammatory responses (142).

Open in a separate window

FIG. 12.

Upstream kinases and redox-sensitive phosphatases in NF-κB activation. Upon binding of stimuli to their receptors, which can be RTKs, TLRs or TNFRs, PI3-kinase is recruited to the membrane and binds to RTKs via its Src homolog-2 (SH2) domain to autophosphorylated tyrosines (72), to TLR via direct interaction with the receptor and MyD88 [reviewed in ref. (290)], and to TNFR via riboflavin kinase (525). Once localized at the membrane, the p110 catalytic subunit of PI3K phosphorylates phosphatidylinositol-4,5-bisphosphate (PIP₂) at the 3′ position of its inositol residue forming PIP₃. PIP₃ recruits Akt and phosphoinositide-dependent protein kinase-1 (PDK1), the latter phosphorylating Akt at T308 and S493 in the regulatory domain. Akt thus activated phosphorylates multiple downstream targets one of them being IKKβ. Other IKKβ phosphorylating enzymes are TAK1, activated in the canonical pathway, and mitogen-activated protein kinase (MAPK) kinase kinase-3 (MEKK3), which can be activated by G-protein-coupled receptors (GPCR). PPs counteracting the involved kinases are the dual substrate phosphatases PP2A and PP2C. PP2A has been shown to colocalize with Akt and this way inhibits PDK1 action. It also reverses MEKK3 activation. PP2Cβ reverses Akt-catalyzed IKK phosphorylation and PP2Cη-2 de-phosphorylates TAK1-phosphorylated IKK. The lipid phosphatases PTEN and Src homology-2 (SH2)-domain-containing inositide phosphatase (SHIP-1) reverse the PI3K signal by removing the phosphate groups at position 3′ (phosphatase and tensin homologue [PTEN]) or 5′ (SHIP-1). All phosphatases in the scheme can be inactivated by oxidative modification of cysteine residues. The oxidizing signal comes from NADPH oxidase (NOX) enzymes that are also activated by receptor-bound PI3K there producing the inositol phospholipid products to which phagocytic oxidase (phox) subunits can bind (107, 232).

Novel mechanisms linking NF-κB activating receptors to NOX have recently been proposed. The TNFR proved to be coupled to NOX1 via riboflavin kinase (RFK), formerly known as flavokinase. RFK bridges TNFR1 and NOX1 via binding of the TNFR1-death-domain to the NOX subunit p22^phox. In cells deficient in RFK, TNF-induced O₂^•− production was inhibited (525). Oakley et al. (353) highlighted the involvement of lipid rafts in the synchronized activation of NF-κB and NOX2 by IL-1β. Both IL-1R1 and NOX2 were found colocalized in lipid rafts. Upon IL-1 stimulation MyD88 was recruited to IL-1R1 and endocytosed into endosomes together with Rac-1 and NOX2 in a caveolin-dependent manner. H₂O₂ produced in this complex facilitated TNFR-associated factor (TRAF6) association with the receptor complex, thereby building a redox-active signaling platform that was called redoxosome (354) (Fig. 13). Recruitment of receptors together with NOX enzymes into specific membrane domains could be the missing link between receptor activation and the mysterious coactivation of O₂^•−/H₂O₂-producing enzymes. Compartmentalized NOX activation and redox signaling has recently been reviewed by Ushio-Fukai (494).

Open in a separate window

FIG. 13.

Redoxosome formation in IL-1 signaling. Redoxosome (redox active endosomes which produce O₂^•−/H₂O₂) formation has been demonstrated for the activation of NOX2 by IL-1 in MCF7 cells (354). After docking of IL-1 to IL-1R1 MyD88 is recruited as effector, and initiates endosome formation. Then, Rac-1 in its GTP-bound conformation together with superoxide dismutase 1 (SOD1) is transferred to the membrane, resulting in the recruitment of the second effector, NOX2 with all its subunits. O₂^•− produced by NOX leaves the endosome via anionic channels (AC) and is dismutated to H₂O₂ by SOD1. This creates an oxidative environment, which promotes docking of IRAK and TRAF6 to the receptor complex finally leading to NF-κB activation. An analogous pathway works in TNF signaling where TRADD and RIP1 are recruited instead of MyD88 and TRAF2 instead of TRAF6.

The NOX family is, however, not the only one of interest in the context of redox regulation of NF-κB. A steadily increasing number of observations document that a coactivation of LOX may be equally important. Overexpression of GPx4, which preferentially reduces lipid hydroperoxides, almost abrogated IL-1β-induced NF-κB activation in endothelial cells, whereas changes in GPx1 had a minor effect (45). Similarly, lipid peroxidation and not H₂O₂ was shown to enhance the activation of NF-κB by TNF in the human endothelial cell line ECV304, whereas it appeared not to be involved in IL-1-induced NF-κB activation (43). The 5-LOX inhibitor AA861 inhibited NF-κB activation in parallel with leukotriene B₄ production in A549 cells (70). By similar approaches Bonizzi et al. demonstrated that NF-κB activation by IL-1β depends on 5-LOX in lymphoid cells, but not in epithelial cells, whereas it depended on NOX in monocytes (39). More recently, activation of TLR8 was demonstrated to induce the phosphorylation of the cytosolic phospholipase A2α (cPLA₂α) and to promote 5-LOX translocation, thus stimulating the synthesis of inflammatory leukotrienes (165). The pivotal kinase involved in activating 5-LOX is likely PKCα, which has for long been known to phosphorylate cPLA2 (285).

The precise mechanisms how the activation of TNFRs or TLRs coactivates NOXs and/or LOXs likely vary with the receptor type, the stimulus, and the tissue. In any case, however, the NF-κB activation consistently occurs under conditions in which the production of O₂^•−/H₂O₂ and lipid hydroperoxides is drastically enhanced, which warrants a serious consideration of the interference of these oxidants with the NF-κB-activating cascades.

2. Modulation of NF-κB activation via redox-sensitive phosphorylation

As outlined NF-κB activation appears to benefit from oxidative events in the cytosol that result in increased phosphorylation of many of its components, thus facilitating nuclear import and ultimately transactivation. A general PTK/PTP imbalance causing an increased NF-κB activity is indeed observed in aged and LPS-treated rats (226). In principle, such enhanced phosphorylation state may be achieved by oxidative activation of protein kinases or oxidative inactivation of phosphatases (see section II.D.3). In particular, the p65 subunit of NF-κB requires phosphorylation at multiple sites for translocation, transactivation, and other activities (Fig. 14). These phosphorylations involve several kinases, as has been amply reviewed in refs. (147, 172, 173, 363, 501). Phosphorylation of p65 and release from IκB were, in fact, the first regulatory steps of the cascade observed (341, 343) and found to be required for DNA binding of NF-κB and the recruitment of coactivators. The possible regulation of NF-κB via oxidative kinase activation has lately been reviewed under thorough consideration of possible methodological artefacts and tissue specificities (363). A somewhat bold summary of this review would be that oxidative modification of kinases has so far not been convincingly shown to be involved in in vivo NF-κB activation, whereas oxidative inactivation is commonly observed.

Open in a separate window

FIG. 14.

Phosphorylation sites of p65 and kinases and phosphatases involved. Phosphorylation at Ser276 prevents interaction of the Rel homology domain (RHD) with the C-terminus of p65. Thereby, DNA binding and interaction with CBP and p300 is facilitated. Similarly, Ser311 phosphorylation enhances the interaction with CBP/p300. Phosphorylation at Thr435 promoter- and cell-type-specifically decreases histone deacetylase (HDAC) binding after TNF stimulation. Dephosphorylation by PP4 (526) appears to be cell type and promoter specific. Phosphorylation at Ser468 supports nuclear import. Phosphorylation at Thr505 inhibits transactivation activity due to increased association with HDAC1. The consequence of Ser529 phosphorylation is unclear; that of Ser536 affects transcriptional acitivity in a cell-specific way. For more detailed information see text (see section IV.D.2).

The Ser/Thr kinase Akt, which itself is activated by phosphorylation at Thr308 and Ser473 via the PI3K pathway, phosphorylates p65 in its DNA-binding domain, the precise site being unknown (62) (Fig. 14). Akt itself is redox sensitive, being inactivated by H₂O₂ due to disulfide formation between Cys297 and 311 and reactivated by Grx (338). The cAMP-dependent kinase PKAc phosphorylates p65 in the cytosol at Ser276 (535), as do the mitogen- and stress-activated protein kinases (MSK)-1/2 in the nucleus (500). These phosphorylations facilitate release from the cytosolic complex or enhance transcriptional activity, respectively (500, 534). The activity of protein kinase A (PKA), though, is inhibited by oxidants, whereas the redox-sensitivity of the MSKs (and ribosomal S6 kinase 1 [RSK-1], see below) appears not to be investigated (363). PKC-type kinases have been reported to be activated by H₂O₂ and other oxidants (153) due to oxidation of Zn-coordinated cysteines in their N-terminal regulatory domain (152), but inactivated by cysteine oxidation/alkylation of the catalytic domains, the prevailing event at a particular phase of the activation process remaining unclear (154). PKCζ, which interacts with the NF-κB activation cascade at multiple sites, also phosphorylates Ser311 of p65 (279), thereby promoting association with the coactivator CBP (102), but, like other PKCs, also PKCζ is inactivated, for example, by peroxynitrite (252). CK2 phosphorylates p65 at Ser529 with so far unknown consequences (32, 505) and at Thr435, which decreases HDAC binding (352). Apparently, the redox-sensitivity of this kinase also remains to be investigated. IKKβ and IKKɛ phosphorylate p65 at Ser468, which supports nuclear import (316), whereas phosphorylation by GSK3β rather inhibited the activity of p65 (53). More recently, Ser468-phosphorylated p65 has been shown to be preferentially ubiquitinated and degraded in the nucleus, leading to the termination of NF-κB-dependent gene expression (143, 307). Finally, p65 is phosphorylated at Ser536 by RSK-1 and, more typically, by IKKs. Phosphorylation at Ser536 is widely considered to be the decisive event that facilitates the release of p50/p65 from IκB in the cytosol (16, 303, 457), although in T-cells it appears to delay nuclear import (316). However, the phosphorylation at Ser536 also plays an important role in the nucleus, where it prevents recruitment of HDAC C3 to chromatin, allows its own acetylation at Lys310 by p300 (185) and impairs affinity to nuclear IκB and, thus, export from the nucleus (36). IKKα has indeed been reported to be recruited to the nucleus together with p65 and to there phosphorylate chromatin-bound p65 (185). IKKβ, the key player in NF-κB activation (16, 416), is redox sensitive in a sense that it cannot account for an activation of NF-κB by oxidants either. IKKβ can become directly oxidized by H₂O₂ at Cys179. This cysteine residue is located between the Ser residues 177 and 181, which have to be phosphorylated for activation. In consequence, oxidation of the residue prevents phosphorylation and results in inhibition of IKKβ (260, 363). By analogy, oxidative inactivation may also be assumed for IKKα and IKKɛ. A seemingly contradictory report claiming an oxidative activation of IKKα and β associated with enhanced phosphorylation at Ser180 or 181, respectively (231), is likely explained by H₂O₂-mediated inhibition of phosphatases (see below). In this context it may be revealing that Cys179 of IKKβ can also be alkylated by 15d-PGJ₂. The modification of IKKβ with this bulky residue proved to be equally inhibitory, likely due to prevention of IKKβ phosphorylation. Since 15d-PGJ₂ is a product of NF-κB-induced COX2, it is considered to contribute to the resolution of inflammation via IKKβ modification (408).

Thus, the enhanced phosphorylation state observed during, and obligatory for, NF-κB activation cannot likely be attributed to oxidative modification of any of the kinases with the possible exception of PKCs and oxidatively activated RTKs that initiate the atypical pathway (339) (see also section II.D.3). In contrast, oxidative inhibition of PPs at each level of the activation cascade may well account for the often seen enhanced NF-κB activation under oxidizing conditions. A systematic RNAi screen of phosphatases in mouse astrocytes identified a total of 19 phosphatases regulating NF-κB transcriptional activity. In particular, the PP2A-type enzymes were found associated with the IKK, NF-κB, and TRAF2 complexes (286). Depending on the cell type, the NF-κB activation mechanism and cross-talking signaling cascades, even more and other phosphatases may come into play (Fig. 12).

The central event in the typical pathway, IKKβ activation by phosphorylation via TAK1, could be reversed by PP2A (92) or PP2Cη-2 (179) and that of IKKα by PP2Cβ (384). Equally, IKKβ phosphorylation by the lysophosphatidic acid-induced MAP kinase kinase kinase (MEKK3) is counteracted by PP2A (465). TNFα-mediated IKKβ activation is reversed by PP2Cα and β (466). Similarly, the phosphorylation of p65 (Fig. 14) at Ser468 is reversed by PP1/PP2A (53), whereas the nuclear Ser/Thr-specific phosphatase PP4 was shown to act on Thr435 (526). Most recently, the wild-type p53-induced phosphatase-1 (WIP1), also known as PP2Cδ, has been identified as a key phosphatase for the most important p65 phosphorylation site Ser536 and, accordingly, is considered to act as a shut-off device of the NF-κB system (66). Interestingly, NF-κB upregulates WIP1 expression, thus contributing to the termination of its activity (299). Upstream events, in particular the PI3K/Akt pathway that substantially contributes to NF-κB activation by IL-1β, TNFα, and LPS (233), offer further possibilities for oxidative interference via phosphatase inhibition (Fig. 12): PI3K phosphorylates phosphatidyl inositol 3,4 bisphosphate [PIP(3,4)P₂] to phosphatidyl inositol 3,4,5 trisphosphate [PIP(3,4,5)P₃]. Interference with NF-κB activation by phosphatases is possible already at these early steps, since formation of PIP(3,4,5)P₃ is reversed by the action of PTEN and SHIP-1. PIP(3,4,5)P₃ in turn activates the phosphoinositide-dependent protein kinase-1 (PDK1). PDK1 phosphorylates Akt at Thr308 and Ser473, and its inactivating de-phosphorylation is again achieved by PP2A (291). Akt can by itself phosphorylate IKKα at Tyr23 and this phosphorylation is required for the phosphorylation of IκB (16).

Out of the numerous phosphatases that are possibly involved in counteracting NF-κB activation (298) a great deal can be inactivated by reversible or irreversible thiol oxidation, glutathionylation, or other mechanism involving oxidative processes, PTEN, SHIP-1, PP2A-, and PP2C-type enzymes being the most quoted suspects.