Imaging ROS signaling in cells and animals

Reactive oxygen species (ROS) act as essential cellular messengers, redox regulators, and, when in excess, oxidative stressors that are widely implicated in pathologies of cancer and cardiovascular and neurodegenerative diseases. Understanding such complexity of the ROS signaling is critically hinged on the ability to visualize and quantify local, compartmental, and global ROS dynamics at high selectivity, sensitivity, and spatiotemporal resolution. The past decade has witnessed significant progress in ROS imaging at levels of intact cells, whole organs or tissues, and even live organisms. In particular, major advances include the development of novel synthetic or genetically encoded fluorescent protein-based ROS indicators, the use of protein indicator-expressing animal models, and the advent of in vivo imaging technology. Innovative ROS imaging has led to important discoveries in ROS signaling—for example, mitochondrial superoxide flashes as elemental ROS signaling events and hydrogen peroxide transients for wound healing. This review aims at providing an update of the current status in ROS imaging, while identifying areas of insufficient knowledge and highlighting emerging research directions.


Introduction
Reactive oxygen species (ROS) are oxygen metabolites that are highly active in terms of oxidative modifications of cellular macromolecules including proteins, lipids, and polynucleotides. Superoxide radical (O 2 ·− ) is usually the primal ROS species produced and is subsequently converted into hydrogen peroxide (H 2 O 2 ) through spontaneous or superoxide dismutase (SOD)-catalyzed dismutation. And reaction of O 2 ·− and nitric oxide (·NO) generates peroxynitrite (ONOO − ), a ROS and reactive nitrogen species (RNS) species. Of all cellular ROS sources, electron leakage from the mitochondrial electron transfer chain (ETC) to molecular oxygen generates a steady flux of O 2 ·− and thus constitutes the major site of cellular ROS production [1,2]. Other enzymes, including NADPH oxidases, lipoxygenase and cyclooxygenase, cytochrome p450s, and xanthine oxidase, also participate in ROS generation [3].
The cellular redox homeostasis is set by a delicate balance between ROS production and the antioxidant system. The ROS-scavenging enzymes include SODs, which convert O 2 ·− to H 2 O 2 , and catalases, which convert H 2 O 2 to water. The antioxidant system consists of glutathione, peroxiredoxin, thioredoxin, and NADPH. Collectively, they form an antioxidant pool, while a third category of enzymes such as glutathione peroxidase and thioredoxin reductase catalyze the interconversion and equilibrium among the reduced/oxidized species of different reductants [4,5].
When ROS are produced excessively or endogenous antioxidant capacity is diminished, indiscriminate oxidation elicits harmful effects, resulting in "oxidative stress". Mounting evidence has established strong links between oxidative stress and a wide variety of pathologies including malignant diseases, diabetes mellitus, atherosclerosis, ischemia-reperfusion injury, and chronic inflammatory processes as well as many neurodegenerative diseases [3,[6][7][8][9][10][11][12][13][14]. Moreover, the oxidative stress theory of aging states that systematic accumulation of oxidative damage from multiple ROS sources constitutes the core process that drives the biological clock of aging [15,16]. Nevertheless, homeostatic ROS are required to maintain a redox environment optimal to biochemical activities of the cell. It has been shown that acute application of SOD mimetics to cardiomyocytes halves the rate of occurrence of spontaneous Ca 2+ sparks [17] and decreases action potential-elicited Ca 2+ transients and contraction [18]. "Reductive stress", a state with too little ROS production and/or too strong antioxidant reactivity, can also lead to pathology. For instance, cardiac-specific expression of human alphaB-crystallin autosomal-dominant mutant hR120GCryAB led to reductive stress causing cardiomyopathy through inducing protein aggregation [19,20].
As an exciting paradigm-shifting development, ROS emerge as powerful, ubiquitous, and indispensable cellular messengers, adding to the repertoire of only a handful of second messengers that we know (Ca 2+ , cAMP, IP 3 , and arachidonic acid). The specific ROS targets range from ionic channels and transporters, to kinases and phosphatase, and to transcription factors, and the list continues to grow and permeate throughout pivotal pathways in differentiation and organogenesis [21], cell fate regulation [22,23], stress response [24], and wound healing [25,26]. However, ROS signaling is notoriously complex. As a rule of thumb, the ROS effects are multiphasic and bidirectional, depending on the species of oxidants, their concentrations, history of exposure, and cellular context. The proven failure of antioxidant therapies despite decades of industrious efforts [27][28][29][30] serves us a humbling lesson on how much we still do not know about ROS signaling. Understanding the cell logic and principles of ROS signaling and developing efficient and specific antioxidant therapies to constrain ROS damages would both hinge on precise and quantitative knowledge of intracellular ROS dynamics, concentrations, compartments, and modes of action.
A few harbingers show us the new horizons in ROS research. First, the trend of ROS investigation moving from cell-free preparations to intact cells and even in living animals; second, the development of a set of novel fluorescent protein-based ROS indicators and protein indicator-expressing transgenic animal models, enhanced with cell type and subcellular compartmenttargeting ability; and third, the visualization of exquisite spatiotemporal architecture of intracellular ROS dynamics by time-lapse imaging in intact cells or in vivo imaging in transgenic animal models. In this short review, we summarize these recent advances in fluorescent ROS imaging in cells and animals with emphasis on novel indicators, genetic animal models, and in vivo imaging technology. Emerging concepts on local ROS signaling will also be discussed. Please see references [31][32][33][34] for recent reviews on related topics.

ROS measurement with small-molecule fluorescent probes
Depending on ROS species and cellular environments, lifetime of a ROS molecule in biological systems varies from nanoseconds to seconds. So what is required for a fluorescent ROS indicator is that it should compete with the antioxidants for ROS and produce fluorescently altered products for visualization and quantification [35]. For an "ideal" ROS indicator, the criteria include selectivity for specific ROS species, fast and reversible kinetics, high signal-tobackground contrast, and superb signal-to-noise properties as well as ease with intracellular loading and proper subcellular compartmentalization. It is also desirable to be excitable at a visible wavelength, be resistant to photobleaching, and display no toxicity in general and phototoxicity in particular. Currently available fluorescent ROS indicators fall in two categories, synthetic small-molecule dyes and genetically encoded fluorescent protein-based probes. As will be discussed in the following, major limitations in ROS measurements are related to selectivity, kinetics, and ability for quantitative calibration.
Of the small-molecule fluorescent ROS probes, 2′-7′dichlorodihydrofluorescein (DCFH), dihydroethidium (DHE), and mitochondrial-targeted DHE (mitoSOX) are the most popular ones. The diacetate form of DCFH (DCFH-DA) is a cell-permeable form that allows ester loading of the dye, resulting in intracellular accumulation of the nonfluorescent DCFH. In the presence of H 2 O 2 and other oxidants, two-electron oxidation of DCFH results in the formation of a fluorescent product, 2′-7′-dichlorofluorescein (DCF), which can be monitored by fluorescence microscopy and flow cytometry [33,36]. However, severe limitations and potential artifacts are confounding the DCF measurement of ROS. Apart from its relative nonselectivity to ROS species and oxidants, DCFH oxidation can also be catalyzed by cytochrome c and heme peroxidases. Worse, the oneelectron oxidization product or DCF radical can react with oxygen to produce O 2 ·− and subsequently H 2 O 2 , thus artificially generating the very ROS that it is attempting to quantify. Cautions should also be taken to minimize light exposure because DCFH is both susceptible to photo-oxidation (increasing DCF fluorescence) and to photobleaching (loss of DCF fluorescence). Kinetically, it is difficult for DCFH to track small and rapid ROS transients because DCFH oxidation is irreversible in the intracellular milieu, and the slope of DCF fluorescence rise (dF/dt), instead of fluorescence intensity (F) per se, is often used to measure the level of ROS. After subtraction of the rising basal fluorescence (F base ), local d(F−F base )/dt has also been used to reflect approximately brief ROS transient [37]. This procedure, however, could be complicated because oxidized DCF becomes membrane-permeable [38]. As to its subcellular compartmentalization, DCFH may be enriched in either the cytosol or the organelle mitochondria, depending on loading and experimental conditions [39].
DHE is widely used as a small-molecule fluorescent ROS probe specific for O 2 ·− [40]. The reaction between O 2 ·− and DHE generates a highly specific red fluorescent product, 2hydroxyethidium (2-OH-E(+)), shifting its excitation and emission peaks from 350 and 400 to 518 and 605 nm, respectively [41][42][43]. mitoSOX, a DHE derivative with addition of a positively charged triphenylphosphonium group (TPP+), is highly enriched in the mitochondria [44,45], and the binding of oxidized mitoSOX to mtDNA greatly enhances its fluorescence [43]. The chemical reactivity of mitoSOX with O 2 ·− is similar to the reactivity of DHE with O 2 ·− , and the particular product 2-OH-E(+) is unique to O 2 ·− since several studies have confirmed that 2-OH-E(+) is the only product in the presence of O 2 ·− generated by the xanthine-xanthine oxidase-O 2 system [46][47][48].
In intact cells, however, the DHE detection of O 2 ·− is still interfered by a prominent reaction: two-electron oxidation of DHE by oxidants other than O 2 ·− produces ethidium cation (E+), another red fluorescent product that is bound to nuclear DNA and often present at a much higher concentration [43]. It has recently been suggested that selective detection of 2-OH-E+ is possible by excitation at 396 nm because an excitation band between 350 and 400 nm is present for 2-OH-E+ but not E+ [49]. However, other studies have reported that E+ can still significantly contribute to the fluorescence intensity even at 396 nm excitation because of high levels of E+ involved [50]. These indicators are also lightsensitive and prone to auto-oxidation [35,43], further constraining and complicating design and data interpretation in time-lapse experiments. Additionally, it has been shown that mitoSOX at high concentration significantly impairs mitochondrial function [49].
Several other useful small-molecule fluorescent ROS probes have been developed. Particularly, dihydrorhodamine 123 (DHR123) is a nonfluorescent agent that scavenges the OH· generated from H 2 O 2 in an iron-dependent Fenton reaction and is thereby converted into the fluorescent rhodamine 123 [51]. DHR123 reacts also with NO 2 · and hypochlorous acid but is unreactive to O 2 ·− or H 2 O 2 in the absence of catalyst [34,[52][53][54]. A family of boronate-based indicators (e.g., peroxysensor family) has also been introduced for targeting to the cytosol or the mitochondria [55][56][57]. Boronate masks the fluorophore; but, upon exposure to H 2 O 2 , it undergoes a nucleophilic attack and its removal unmasks the fluorescence emission. However, the boronate-based indicators are promiscuous as they also react stoichiometrically with ONOO − , yielding phenols and permitting light emission [58,59].
HKGreen-3, a rhodol-based fluorescent probe, is recently developed by Peng et al. and shows high sensitivity and selectivity for peroxynitrite in both chemical and biological systems [60]. HKOCl-1 is a BODIPY-based fluorescent probe for detecting hypochlorous acid with high specificity [61]. 4-Amino-5-methylamino-2′,7′-difluorofluorescein (DAF-FM) is of popularity for measuring NO due to its high sensitivity, pH stability, and relative resistance to photobleaching. Since the detection relies on conversion of the parent compound into a fluorescent triazole, the presence of oxidants/antioxidants and reaction with other molecules would affect this fluorescence detection [62]. Amplex UltraRed is a fluorogenic substrate for horseradish peroxidase that reacts with H 2 O 2 in a 1:1 stoichiometric ratio to produce the fluorescent product resorufin with long-wavelength spectra (excitation/emission maxima, ∼563/587 nm) [31]. In a recent study, Amplex UltraRed and DAF-FM have been successfully used for in vivo measurement of extracellularly released H 2 O 2 and NO of superficial lumbar spinal cord of anesthetized mice, respectively [63].
The caveat from above considerations is that ROS measurement with small-molecule fluorescent probes is not as straightforward as it seems to be. Proper experimental design, careful choice of loading and light illumination parameters, stringent control of experimental conditions, and judicious interpretation of experimental data should all be exercised. Whenever possible, cross-confirmation with multiple independent approaches is highly recommended. Evidently, developing small-molecule fluorescent ROS probes suitable for faithful measurement of ROS dynamics remains a huge challenge to the ROS research field.

ROS measurement with fluorescent protein-based indicators
Over the last decade, fluorescent protein-based ROS indicators have entered the arsenal for ROS measurement. While protein chemistry introduces a higher level of complexity as compared to the small-molecule chemistry, it at the same time offers tremendous opportunities for rational design (e.g., redox oxidation-sensitive green fluorescent proteins (roGFPs) for redox potential measurement and HyPer for H 2 O 2 measurement) [64][65][66] as well as serendipitous discoveries (e.g., mt-circularly permutated yellow fluorescent protein (cpYFP) for mitochondrial superoxide flashes) [67]. While selectivity is generally improved, reversibility could also be achieved by exploiting the endogenous antioxidant system or new protein chemistry. Another distinct advantage is that these genetically encoded indicators can be specifically targeted to different type of cells using cell typespecific promoters or to different cellular compartments or microdomains by N-or C-terminal fusion with a specific targeting sequence [68,69]. ROS indicator-expressing transgenic animals have been generated [67,[70][71][72][73], allowing for imaging ROS ex vivo and in vivo. However, the fluorescent protein indicators have also their own set of limitations.
Particularly, pH sensitivity is common to most of currently used protein-based ROS indicators, due to reversible fluorescencequenching protonation of the chromophore at pK a close to physiological pH. As such, pH changes in the cytosol or other specific compartments, if not judiciously controlled, could lead to erroneous observation and misinterpretation.
Imaging mitochondrial superoxide flashes with cpYFP Among all ROS sources, mitochondrial ETC is the major site of cellular ROS production. The ETC consists of intermolecular and intramolecular pathways of increasing redox potential (E h ), from −320 mV at the entry point of complex I to +390 mV at the terminal point of complex IV. However, with its E h =−160 mV, O 2 also snaps up 0.15-2 % of the respiratory chain electrons, one at a time, at places prior to complex IV; and one-electron reduction of O 2 forms O 2 ·− [1,2], the primal ROS. This constitutive mode of mitochondrial ROS production plays an important role in setting the ROS and redox homeostasis of the cell.
While studying mitochondrial Ca 2+ signaling with a genetically encoded mitochondrial-targeted Ca 2+ indicator, pericam [74], we serendipitously found that the fluorescent moiety of pericam, cpYFP, can reversibly detect superoxide with high selectivity and sensitivity [67]. Extensive in vitro characterization indicates that, when excited at 488 nm, cpYFP emission displays a several fold increase in response to superoxide generated by the xanthine-xanthine oxidase-O 2 system, but is insensitive to many other oxidants and metabolites, including H 2 O 2 , peroxynitrite, Ca 2+ , ATP, ADP, NAD(P) + , and NAD(P)H. It is also insensitive to E h varying between −319 and −7.5 mV (controlled by mix of oxidized and reduced DTT in different proportions), while displays a pH sensitivity with a pK a ∼8.5 [67]. The reversibility of cpYFP has been evidenced by the fact that SOD added after superoxide formation reverses the cpYFP signal in vitro. In addition, mitochondrial-targeted cpYFP (mt-cpYFP) acts as a ratiometric indicator because its signal at 405 nm excitation is essentially ROS-independent, allowing for the use of the F 488 /F 405 ratio as the readout.
Using mt-cpYFP, we have uncovered a new mode of mitochondrial ROS production-"superoxide flash". Superoxide flashes are sudden, brief, and bursting superoxideproducing events in single mitochondria. In intact cells, spontaneous flashes occur at a low rate in a stochastic manner [67], but their frequency can be regulated over a broad dynamic range. To date, superoxide flashes are universally found in all cell types examined, including cardiomyocytes, skeletal muscle cells, neurons, glials, fibroblasts, and several types of cancer cells [23,67,70,[75][76][77][78][79][80][81][82]. They are also highly conserved in species ranging from mammals (mouse, rat, and human) and to Caenorhabditis elegans (unpublished data). Ex vivo and in vivo imaging in transgenic mouse models with cardiac-specific or pan-tissue mt-cpYFP expression have allowed for detection of superoxide flashes in beating hearts under Langendorff perfusion [67] and in gastrocnemius muscle and sciatic nerve of living mice under anesthesia (Fig. 1) [70,71]. In addition, superoxide flashes of similar characteristics are active in freshly isolated, respiratory mitochondria [83], indicating that single mitochondria contain the full machinery for the genesis of superoxide flashes. Notably, the rate of flash occurrence varies depending on species, tissue and cell types, metabolic states, the presence of stressors, and disease conditions; the amplitude and duration of the flashes, however, appear to be stereotypical at multiple levels, from isolated mitochondria, to intact or plasma membrane-permeabilized cells, to whole tissues or organs, and to living animals.
The superoxide origin of the mitochondrial flashes has stirred some hot debates. Sweetlove et al. reported similar phenomenon in Arabidopsis mitochondria, but interpreted its nature as transient alkalization of the mitochondrial matrix [84,85]. Recently, we and others have systematically examined the respective contributions of ROS and pH to a flash and concluded that ROS burst is the dominant signal while pH change, if any, has only a minor contribution [83]. In experiments combining the protein and small-molecule ROS indicators, it has been shown that mitoSOX, which is pH insensitive, reports a concomitant stepwise increase during a mt-cpYFP flash [78,83,86]. Furthermore, owing to its Modified from Fang et al. [71] reversibility, mt-cpYFP is able to track the time course of flashes with the briefest duration of only ∼1 to 2 s [80].
Differing from constitutive mitochondrial ROS production, the genesis of superoxide flashes involve different but overlapping molecular mechanisms. The flash ignition is tightly coupled to transient opening of mitochondrial permeability transition pore (mPTP), evidenced by sudden dissipation of mitochondrial membrane potential and partial and irreversible loss of fluorescent solutes (MW, 752-980 Da) preloaded to the matrix [67,75]. The involvement of mPTP is also supported by the fact that the flash production is partially sensitive to pharmacological inhibition or molecular knockdown of cyclophilin D, consistent with a regulatory, but dispensable, role of this protein in mPTP activity [87][88][89][90]. Another prominent property of the superoxide flashes is their dependence on functionally intact ETC. The flash production is abolished by disruption of the ETC at any possible site, complex I through V, in a manner distinctly different from ETC regulation of constitutive ROS production. For instance, antimycin A, an inhibitor-targeting complex III abolishes the bursting quantal ROS production in the form of the flashes [67] while stimulating continuous ROS production [1,91,92]. That inhibition of complex V also suppresses the flash production is consistent with an intimate relationship recently suggested for the ATP synthase and mPTP [93,94].
By in situ, ex vivo, and in vivo ROS imaging, we and others have shown that superoxide flashes represent not only a digital readout of mitochondrial metabolic status but also a novel biomarker of mitochondrial stress. The rate of skeletal muscle superoxide flashes in live anesthetized mice increases after intraperitoneal injection of glucose or insulin, indicating that superoxide flashes are coupled to whole-body dietary glucose metabolism [70]. In isolated skeletal muscle cells with electroporation-mediated transient mt-cpYFP expression, superoxide flashes occur at a markedly elevated frequency and display a similar though less profound response to glucose plus pyruvate stimulation, in the absence of insulin and other whole-body factors [78]. A further elevated superoxide flash activity is observed in skeletal muscle of RyR1 Y522S/WT malignant hyperthermia mice which exhibit marked temperaturedependent increases in ROS and RNS generation [76]. Thus, imaging superoxide flashes in vivo exemplifies how both the integrative whole-body metabolic response and the mechanistic single-mitochondrion behavior can be investigated in one single experiment.
Investigation of superoxide flash production in mice deficient of SOD2 has revealed that superoxide flashes negatively regulate neural progenitor proliferation and cerebral cortical development through modulating activation of ERK [23]. Remarkably, a reversible 20-fold increase of superoxide flashes occurs in response to hyperosmotic stress, due to the synergistic effect mitochondrial Ca 2+ uniport, and basal ROS elevation. The high activity of superoxide flashes, in turn, contributes to activating JNK and p38, essential signals for adaptive cell survival responses [81]. In cultured cardiomyocytes, a flurry of superoxide flash activity occurs in a 5-10 min window after reoxygenation from hypoxia or anoxia [67,95]. In the pathology of Huntington disease, the elevated flash activity induced by elevated mitochondrial Ca 2+ signaling acts to exacerbate mtDNA damage [82]. Likewise, superoxide flashes act as early mitochondrial signals mediating the apoptotic response during oxidative stress in HeLa cells [79].
Collectively, these recent advances indicate that superoxide flashes offer a rare window through which we can glimpse into the whole-body metabolic response at the single-mitochondrion level, and gauge a wide variety of stresses converging to the mitochondria. To our knowledge, many types of cpYFP transgenic organisms, from mice, zebrafish, C. elegans, Drosophila melanogaster, and yeast, have been generated or are currently being created to address multidisciplinary questions in broad settings. We are eager to see what these new models and approaches can teach us about ROS signaling in biology and diseases.

Imaging H 2 O 2 with HyPer
HyPer is a ratiometric fluorescent indicator of H 2 O 2 in which cpYFP is inserted into the regulatory domain of an Escherichia coli peroxide sensor OxyR [66]. Naturally used by the bacterium to trigger transcriptional response to oxidative stress, OxyR contains an H 2 O 2 -sensitive regulatory domain and a DNA-binding domain, and, upon oxidation by H 2 O 2 , intramolecular disulfide bond forms between two cysteine residues (Cys199 and Cys 208) and the resultant conformational change shifts the excitation maximum of the attached cpYFP from 420 to 500 nm (emission maximum at 516 nm) [66]. HyPer is able to detect nanomolar H 2 O 2 in vitro and, when expressed in cells, responds to micromolar H 2 O 2 added externally [66] or changes of intracellular H 2 O 2 upon growth factor stimulation [66,96]. In an elegant study, Niethammer et al. have exploited HyPer expressed in transgenic zebrafish larvae to visualize a regional, graded, and transient H 2 O 2 signal produced by dual oxidase (Duox) in response to tail fin injury. Functionally, this Duox-elicited H 2 O 2 signal is required for rapid recruitment of leukocytes in the process of wound healing [25].
A series of HyPer mutants have been developed in order to improve its dynamic range and reaction kinetics for H 2 O 2 detection. In particular, HyPer-2, an A406V single-point mutant of HyPer, exhibits twice-expanded dynamic range, but the response to H 2 O 2 is much slower, doubling both halfoxidation and half-reduction from ∼6 and ∼200 s for HyPer to ∼13 and ∼400 s for HyPer-2, respectively [97]. Recently, a H34Y mutant of HyPer, HyPer-3, was developed, which shows expanded dynamic range compared to HyPer and faster oxidation-reduction kinetics compared to HyPer-2 [72]. Notably, in HyPer-3 transgenic zebrafish, similar H 2 O 2 gradients along the fish tail regions were observed upon wounding; however, HyPer-3 showed a higher fluorescence ratio (F500/F420) [72] than what reported by HyPer [25] demonstrating its advantage for H 2 O 2 detection.
Both HyPer and HyPer-3 are applicable for fluorescence lifetime imaging microscopy (FLIM) [72]. Instead of measuring the fluorescence intensity, the physical parameter measured in FLIM is the time constant (τ) of the excited fluorophore returning to its basal states and, in this case, the change in τ is quantified to reflect the redox state of the ROS indicator and hence the ROS level. Because τ is independent of indicator concentration, FLIM measurement is essentially insensitive to indicator expression level, non-uniform distribution, and partial photobleaching. It is also advantageous for quantifying signals at different depths in a biological tissue because τ is less interfered by light scattering and reabsorption (inner filtering). Moreover, FLIM generates absolute quantitative readouts while requiring only a single-wavelength excitation, provided that the indicator is calibrated in situ (e.g., in permeablized cells) or in vitro under conditions closely resembling intracellular environments.
As is the case for mt-cpYFP, HyPer and its mutants are pH sensitive: a shift of 0.2 pH units is sufficient to change the F500/F420 ratio as much as those corresponding to a full reduction or oxidation [66]. Thus, monitoring pH changes should be included as an essential control in HyPer measurement. A second concern is about its ROS selectivity. Although in vitro data have shown that HyPer is insensitive to other oxidants including O 2 ·− , GSSG, nitric oxide, and ONOO − [66], events similar to mt-cpYFP flashes were detected with mitochondrial targeted HyPer, accompanying a simultaneous stepwise increase of mitoSOX fluorescence [86]. SypHer, a pH indicator and insensitive to H 2 O 2 (by a disruption of the H 2 O 2 -sensing cysteine pair (C199S), of HyPer), detects similar mitochondrial flash events [66,86,98]. Thus, since HyPer and SypHer comprise a cpYFP as the fluorophore, it is possible that the cpYFP part in these two indicators reports mitochondrial superoxide flashes [86], as do mt-cpYFP and pericam [67,78].

Imaging redox potential with roGFPs
As an example of rational design, roGFP1 have been generated by substituting two surface exposed amino acids in GFP with cysteines (S147C and Q204C). The introduced Cys 147 and 204 are situated next to each other on two adjacent βstrands and form disulfide bonds due to significant conformational changes of roGFP1 upon oxidation [64,65]. These cysteines are located near the chromophore of GFP and the formation of the disulfide bond leads to a simultaneous shift of the absorption properties. The roGFPs have two fluorescence excitation maxima at about 400 and 490 nm, corresponding to the neutral fluorophore and anionic form of the flurophore, respectively [64,65]. The disulfide formation promotes protonation of the chromophore and increases the excitation peak near 400 nm at the expense of the peak near 490 nm. Therefore, they serve as dual-excitation ratiometric indicators for E h measurement in vitro and in vivo [64,65]. The first generation of roGFPs with different mutation sites, including roGFP1-6, all have midpoint E h of −272 mVor below, which made them most useful in reducing compartments, such as the cytosol and the mitochondrial matrix [64]. In particular, roGFP1and roGFP2 expressed in cytoplasm report a basal E h of −315 to −325 mVand both respond to a variety of oxidant stimuli [65]. Interestingly, the mitochondrial matrix of HeLa cell is highly reducing with a midpoint E h near −360 mV as reported by roGFP1; membrane-permeable reductants and oxidants reversibly change the E h in the matrix of mitochondria [64]. Recently, transgenic animals of roGFPs have been developed and provided very useful tools for investigating the physiology and pathology of ROS signaling [73,99]. A combined approach using transgenic mice with mitochondrial-targeted roGFP and two-photon laser scanning microscopic imaging in brain slices have shown that normal autonomous pace-making produced oxidative stress specific to dopaminergic neurons in substantia nigra pars compacta that are usually vulnerable [73]. In mitochondrial-targeted roGFP2 transgenic Drosophilae, it has been demonstrated that elevated ROS contribute to pathogenesis in a neurodegenerative mutant ATPalpha DTS1 and in a model of mitochondrial encephalomyopathy [99].
It should be cautioned that it takes minutes or longer for current roGFPs to equilibrate with the environmental redox potential changes and their reversibility is too slow to detect transient ROS events [32]. Indeed, roGFP2 expressed in either cytoplasm or the plasma membrane showed no response to stimulation with epidermal growth factor or lysophosphatidic acid, which induces H 2 O 2 production that can be detected by DCFH [65,100,101]. By measuring the steady-state redox levels in different cellular compartments, future investigations may exploit roGFPs to complement the measurements using selective, fast responding, and reversible ROS indicators. Importantly, to meet the needs of measuring E h in severe oxidative stresses, it would also be desirable to obtain a collection of redox indicators with different midpoint potentials.

Fluorescence resonance energy transfer-based ROS indicators
Fluorescence resonance energy transfer (FRET)-based ROS indicators, which consist of cyan and yellow fluorescent proteins (CFP/YFP) linked by redox sensitive polypeptides, have also been developed. The FRET-based ROS indicators sense the redox state via their internal disulfide bonds, resulting in a conformation change of the protein leading to a FRET response. In this regard, Kolossov et al. developed a series of FRET ROS indicators with different redox sensitive linkers, which consist of α-helical structures in conjunction with redox-sensitive motifs, between CFP and YFP and found that RL5 exhibited a 29 % increase of FRET efficiency from its reduced to oxidized states [102]. Guzy et al. have developed a FRET ROS sensor with the 69 amino acid cysteine-containing regulatory domain from redoxregulated heat-shock protein HSP-33 [103] as the linker of CFP and YFP [104]. With this FRET indicator, they found that hypoxia-induced ROS production requires a functional mitochondrial ETC and that is essential for hypoxia-induced HIF-1α stabilization [104]. A third type of FRET indicators, Redoxfluor, comprises a tandem repeat of partial sequence of carboxy-terminal cysteine-rich domain of Yap1, a yeast transcriptional factor sensing the intracellular redox state [105]. By expressing Redoxfluor to the peroxisome of yeast and Chinese hamster ovary cells, Yano et al. have demonstrated that the redox state within the peroxisomes is more reductive than that in the cytosol in wild-type cells, despite the fact that ROS are generated within the peroxisomes, and the cytosolic redox state of the of cell mutants for peroxisome assembly is more reductive than that of the wild-type cells [105]. Because FRET based-ROS measurement builds on disulfide bond formation of the redox-sensitive linkers, its specificity and kinetics are subjected to the same constraints for other disulfide bond-based ROS indicators discussed above.

Perspectives
The past decade has witnessed significant progress in understanding physiological and pathological functions of ROS, marked by the emergence of a revolutionary concept that ROS acts as cellular messengers that permeate pivotal pathways in the network of intracellular signal transduction. This ongoing revolution has been catalyzed, to a significant extent, by the advent of new small-molecule and fluorescent proteinbased ROS indicators and novel imaging methods, both conferring the ability to visualize and quantify ROS in organelle, intact cells, whole tissues and organs, and even live animals (Fig. 2). We begin to appreciate that, analogous to Ca 2+ signaling, spatiotemporal ROS dynamics exhibit an exquisite hierarchical architecture in intact cells and organisms, from superoxide flashes as elemental mitochondrial ROS signaling events to cell-wide ROS oscillations [17,106,107] supported by the ROS-induced ROS release mechanism [39,108] and to tissue-level ROS gradients for wound healing.
Imaging ROS in situ and in vivo, with high selectivity, quantitative ability, and spatiotemporal resolution will continue to be our most delicate investigative tool for the analysis of the tremendous complexity and subtlety of ROS signaling. To further sharpen the tool, it calls for continued efforts in designing small-molecule fluorescent ROS probes with improved selectivity, reversible kinetics and compartment-targeting property. Meanwhile, biologically inspired novel ROS indicators could be developed as we identify more ROS target proteins and understand better the mechanisms hereby they achieve signaling specificity, sensitivity and reversibility at once. Combined, the promise is that, by searching the enormous chemical space of small molecules and of proteins, we would greatly extend the current repertoire of ROS indicators and ultimately achieve the same level of reliability as we have enjoyed while measuring intracellular Ca 2+ with smallmolecule probes such as fluo-3 [109], fura-2 [110], indo-1 [110], and fluorescent protein probes such as GCamp6 [111] and GECOs, the palette with blue, improved green, and or redshifted indicators [112]. In synergy with the exponentially increasing numbers of indicator-expressing organisms and disease models, the booming technology for super-resolution and single-molecule imaging [113][114][115], and the trend for using miniature, plant-in devices to obtain images in conscious, free-  Fig. 2 Imaging ROS dynamics in vitro and in vivo. By combining transgenic animal models with confocal and multiphoton microscopy, images of high spatiotemporal resolution can now be acquired from isolated mitochondria, cultured cells, intact tissues or organs, and even in living animals moving animals [116][117][118][119][120], imaging ROS in vivo will serve as the most powerful force to transform the landscape and push forward the frontiers in ROS signaling.