Reactive oxygen species or reactive sulfur species: why we should consider the latter

Life is nothing but an electron looking for a place to rest.

     Albert Szent-Györgi

Nothing in biology makes sense except in the light of evolution.

     Theodosius Dobzhansky

Life is created, regulated and sustained by reduction–oxidation (redox) reactions (see Glossary) that drive photosynthesis, respiration and most reactions in between. This depends to a large extent on initially having electrons out of balance, which requires energy. For many simple chemoautotrophs (see Glossary), the immediate environment provides this energy. The situation is more complex for larger organisms, for which this energy is provided almost completely by sunlight. Organisms that use oxygenic photosynthesis (see Glossary; mainly plants and cyanobacteria) use solar energy to pry electrons from water and force them into carbon dioxide, thereby producing reduced carbon and the building blocks of life, with oxygen essentially as a byproduct. Aerobic animals simply reverse this process in obligate symbiosis.

The importance of redox balance, i.e. the ‘redox code’ (Jones and Sies, 2015) and oxygen's abundance and proclivity for electrons is the foundation of the concept of oxidative stress (Sies et al., 2017). A slight excess of oxidants (see Glossary) is important in cellular signaling (eustress); however, additional oxidants (oxidative stress) initiate complex protective antioxidant reactions, and the pathophysiological processes that are initiated by excess reactive oxygen species (ROS; oxidative distress) so adversely affect cellular processes that they are catastrophic for cellular function (Sies et al., 2017). These properties, largely attributed to reactive oxygen species (ROS), have been ingrained in virtually all aspects of biology; an average of nearly 125 reviews with ‘reactive oxygen species’ in the title were published in 2017 and 2018. By late November of 2019 there were another 90 reviews (for a representative sample, see: Beltrán González et al., 2019; Cuevas et al., 2019; Damiano et al., 2019; Fan et al., 2019; Fisher et al., 2019; Huang and Li, 2019; Karki and Birukov, 2019; Kaushal et al., 2019; Momtahan et al., 2019; Munro and Pamenter, 2019; Onukwufor et al., 2019; Saikolappan et al., 2019; Tejero et al., 2019; Villamor et al., 2019; Weinberg et al., 2019; Yang, 2019; Zhang et al., 2019). Not surprisingly, there is also an increasing interest in ROS in signaling and physiology in plants (Foyer, 2018; Janků et al., 2019; Kurek et al., 2019; Leister, 2019; Mhamdi and van Breusegem, 2018; Waszczak et al., 2018).

As a natural consequence of the broad implications of ROS, there has been considerable interest in manipulating ROS to better understand homeostatic processes as well as to mitigate the adverse effects of excess ROS. As will be demonstrated in this Commentary, manipulating ROS is not as straightforward as it might seem (see Olson and Gao, 2019), and attempting to mitigate the adverse effects of ROS with antioxidants has had limited success (Alcala et al., 2018; Casas et al., 2015; Chiurchiù et al., 2016; de Vries et al., 2015; Edrey and Salmon, 2014; Gomez-Cabrera et al., 2015).

Is it possible that ROS are not the central players here? Might we even be looking at the wrong molecules? In this Commentary, I offer up another possibility. I propose that some of the attributes of ROS may be due to other, similar yet distinctly different molecules: reactive sulfur species (RSS). I do not attempt to claim that all activity attributed to ROS is a property of RSS. However, there are striking similarities between ROS and RSS in their chemistry, metabolism and signaling mechanisms that warrant consideration. Here, I also discuss technical and methodological problems that have led to confusion between ROS and RSS, and I offer several examples where homeostatic processes attributed to ROS can just as easily be explained by RSS. Finally, because ‘evolution explains biology’, I make the case that RSS have been inexorably intertwined with life from its origin to the present day. Aspects of severe ROS exposure (i.e. ROS distress), which is often artificially generated, are not considered.

One reason why I believe that we should pay more attention to RSS is that oxygen and sulfur (and their associated reactive species) are chemically very similar; both oxygen and sulfur are chalcogens (see Glossary) with six valence electrons (see Glossary). Of course, there are differences between the two elements – oxygen is more electronegative than sulfur and its most common oxidation states (see Glossary) are −2 and 0 (molecular oxygen), but it also exists in less stable forms of −1, −1/2, +1 and +2. The oxidation states of sulfur range from −2 to +6, and sulfur has over 30 allotropes (see Glossary), whereas oxygen has fewer than 10. Sulfur is larger than oxygen (32 amu versus 16 amu). This means that the electrons of sulfur are farther from the positive nucleus and more ‘promiscuous’, i.e. electron transfer reactions are more favorable for sulfur than for oxygen.

Overall, RSS tend to be more ‘versatile’ than ROS. Once produced, H₂O₂ is quickly removed by cellular antioxidant mechanisms (described below). By contrast, RSS may be ‘stored’ as a variety of inorganic or organic polysulfides. When sulfur is oxidized from the formal oxidation state of −2, as in H₂S or cysteine sulfur (CysSH), to sulfane sulfur with a formal oxidation state of −1 or 0, it can be stored as an inorganic or organic persulfide or polysulfide of the form RS(n)R, where R is H or cysteine (CysS) and n=1–5 (e.g. hydrogen polysulfide, H₂S₃, H₂S₄, etc.), cysteine polysulfide (e.g. CysSSH, CysSSSH, CysSSSSH, CysSSSCys, CysSSSSCys) or glutathione (GSH) polysulfides [e.g. GSH(S_n)H, GSH(S_n)GSH] (Ida et al., 2014; Ono et al., 2014). H₂S can be regenerated from these persulfides/polysulfides by cellular reductants (see Glossary; e.g. ascorbate or dihydrolipoic acid) or the sulfur may be transferred from one persulfide/polysulfide to another and can be reused as a signaling moiety. Thus, there is a large source and sink for RSS in cells. Furthermore, although H₂S is also reactive, clearly H₂O is not (see below). It should be noted that ROS and RSS could also be compared with reactive nitrogen species (RNS); however, such comparisons are beyond the scope of this discussion. For the interested reader, they are covered in the comprehensive review by Cortese-Krott et al. (2017).

The mechanisms of ROS and RSS signaling are identical; however, RSS signaling appears to be more extensive than ROS signaling, and may account for some effects that are commonly attributed to ROS. For example, both ROS and RSS signal through their reaction with protein cysteines. Cysteines constitute only ∼2% of the amino acids in vertebrate proteins, yet over 90% of proteins have at least one cysteine (Miseta and Csutora, 2000). Cysteine is one of the most highly conserved amino acids, to a large extent owing to the diverse and pivotal roles that it plays in various aspects of biology, including redox sensing, signaling, catalysis, the folding, processing and trafficking of proteins, protein–protein interactions, DNA and RNA binding, cell differentiation, cell cycling and circadian rhythms (Go et al., 2015). Key to the regulatory function of cysteine is its reversible redox reactivity; cysteine residues are often referred to as redox or sulfur switches (Jones and Sies, 2015).

Twenty-two proteins that can be either peroxidated or persulfidated have thus far been identified (Zhang et al., 2017). Although, experimentally, peroxidation and persulfidation are often attained with H₂O₂ or H₂S₂ concentrations that are probably supraphysiological, in the one instance where peroxidation and persulfidation were directly compared, H₂S₂ was far more effective at modifying the protein (Greiner et al., 2013). Fig. 1 (Meng et al., 2018) illustrates the number and variety of cysteine switches affected by RSS in the cardiovascular system. It is also believed that many proteins are persulfidated in vivo (see below), and there are likely to be considerably fewer endogenously peroxidated proteins.

One of the most exciting developments in this field is the study by Akaike et al. (2017), which was designed to explain numerous previous and puzzling observations that many protein cysteines in cells are persulfidated and polysulfidated. This paper showed that the enzyme cysteinyl-tRNA synthetase (CARS) acts as a cysteine persulfide synthase (CPERS) and catalyzes the transfer of sulfur from one l-cysteine to another, thereby forming a cysteine hydropersulfide (CysSSH; Fig. 2). This process can be repeated, ultimately generating CysSSSH and CysSSSSH, i.e. CysS-(S)_n-H, where n=1–3. CARS also catalyzes the incorporation of CysS-(S)_n-H into tRNA so that it can subsequently be incorporated into proteins. Thus, cysteine persulfides and polysulfides can be introduced into proteins co-translationally, and these modifications will be retained in the mature protein.

A single CARS is present in bacteria, whereas mammals have two forms, one cytoplasmic (CARS1) and one mitochondrial (CARS2). CARS2 is responsible for most, if not all, of the cysteine persulfidation in mice, as persulfidation is halved in heterozygous animals (Cars2^+/−) (Akaike et al., 2017). Complete deletion of CARS2 (Cars2^−/−) is fatal, and the effect on protein sulfur is unknown. CysSSH and CysS-(S)_n-H are synthesized by CARS2 in the mitochondria, and CySSH is then preferentially released into the cytosol, where it can be further persulfidated and, ultimately, incorporated into proteins. CARS2-derived persulfide also enhances mitochondrial biogenesis and is linked to the electron transport chain. Although CARS2 synthesizes mostly CysSSH and CysSSSH in buffer, in HEK293 cells it appears to preferentially form HS⁻/H₂S and thiosulfate by accepting an electron from the electron transport chain, which reduces the polysulfide; however, this needs further verification. The function of extensively polysulfidated proteins in the cell remains to be determined, although it is clear that, in conjunction with the Trx/TrxR and GSH/GSHR pathways, cells have potent oxidoreductase mechanisms for regulating protein cysteine RSS.

As described above, both the Trx/TrxR and GSH/GSHR pathways reduce persulfidated proteins analogous to their action on peroxidated proteins. We recently began to investigate RSS metabolism and the evidence suggests that most (if not all) antioxidant mechanisms found in modern organisms also metabolize H₂S or polysulfides. Both superoxide dismutase (SOD) and catalase (Cat) oxidize the conversion of H₂S to polysulfides in buffer in the presence of an oxidant, and in hypoxia Cat reverses this process and generates H₂S from Trx and NADPH (Olson et al., 2017a,b). The oxygen tension at which Cat switches from an oxidase that catabolizes H₂S to a reductase that produces H₂S is nearly identical to the P₅₀ for oxyhemoglobin. This suggests a physiological function for Cat in matching perfusion with metabolism, as Cat removes vasodilatory H₂S in red blood cells when tissues are normoxic and produces H₂S when the tissues become hypoxic.

We also examined RSS metabolism in HEK293 cells exposed to inhibitors of other antioxidant pathways (Olson and Gao, 2019). We used a number of chemicals that are commonly employed to increase intracellular H₂O₂. These included auranofin (Aur), a thioredoxin reductase (TrxR) and glutathione reductase (GSHR) inhibitor (Rodman et al., 2016; Saccoccia et al., 2014), conoidin A (ConA), a peroxiredoxin (Prx) inhibitor (Haraldsen et al., 2009), tiopronin (Tio), an inhibitor of glutathione peroxidase (GPx) and cystine (CSSC) uptake (Chaudiere et al., 1984; Hall et al., 2014), l-buthionine-sulfoximine (BSO) and diethyl maleate (DEM). BSO and DEM also deplete intracellular GSH; however, BSO decreases intracellular Cys whereas DEM increases it (Albano et al., 2015). We found that Aur, BSO, DEM and ConA increased cellular H₂S, whereas Tiop decreased it; Aur initially decreased, then increased polysulfides, ConA increased polysulfides, whereas BSO, DEM and Tiop had no effect. Thus, although the inhibitors may affect other cellular processes as well, and our experiments need to be confirmed with more selective inhibitors, the results appear to suggest that (1) RSS metabolism is distinct from, and independent of, ROS metabolism and (2) antioxidant mechanisms commonly attributed to ROS regulation are involved in sulfur metabolism in extant organisms.

It is also worth noting that the calculated production rates of ROS and RSS are similar, whether determined from estimates of ROS and RSS in an adult human (DeLeon et al., 2016) or ROS and RSS production in individual HEK293 cells (Olson et al., 2019a). However, because of methodological difficulty in distinguishing between ROS and RSS (described below), RSS production may exceed production of ROS.

In addition to the factors discussed above, there are a number of considerations relating to the research process that may lead us to underestimate the biological importance of RSS. These include the fact that most research is oxycentric, and is performed using tools that do not distinguish between RSS and ROS. Furthermore, research tends to focus on murine models (especially in the biomedical field).

Historically, most biological and biochemical research was (and still is) conducted in room air, i.e. ∼20.9 kPa oxygen or 18.5 kPa O₂ in standard tissue incubators. This is close to 200 μmol l⁻¹ O₂ – four times higher than ‘physioxia’ in most cells and as much as 200 times greater than that in the mitochondria (Keeley and Mann, 2019; Olson, 2019; Ward, 2007). Measurement of mitochondrial metabolism at 18.5 kPa, which is still done with some of the most expensive equipment, implicitly favors ROS. Even the term ‘oxidation’ reflects this bias, as it is a chemical term, whereas ‘reduction’ is a mathematical connotation. Because of their chemistries, oxygen and sulfide do not co-exist, either in cells or in the environment; it is not surprising that ROS are artificially increased in experimental conditions flooded with oxygen (Maddalena et al., 2017) and that RSS are scarce.

Another issue to consider when discussing research factors that influence our understanding of RSS is the fact that RSS may be mistaken for ROS. ROS fluorescent probes are not only non-specific for other ROS (Kalyanaraman et al., 2012) but also may have difficulty distinguishing between ROS and RSS (DeLeon et al., 2016). For example, redox-sensitive green fluorescent protein (roGFP), arguably the gold standard for ROS measurement in different intracellular compartments (Cannon and Remington, 2008; Schwarzländer et al., 2015; Waypa et al., 2010), is 200 times more sensitive to RSS than it is to ROS. Amperometric H₂O₂ ‘specific’ electrodes are over 24 times more sensitive to H₂S than to H₂O₂, and RSS are also detected by dichlorofluorescein (DCF), MitoSox Red and Amplex Red. DCF is especially problematic, as it is also readily oxidized by Cat. These discrepancies not only lead to confusion in separating ROS from RSS but also can lead to errors in estimating metabolic production.

A final factor that may affect our understanding of RSS biology is the nearly exclusive use of murine models to investigate human health, as this may bias estimates of ROS production in other organisms. Small rodents with extraordinarily high mass-specific oxygen consumption may indeed produce a lot of ROS, but are these typical animals? Kroghian principals need to be applied to examine animals with more typical levels of oxygen consumption, including hypoxia-tolerant animals, fetal mammals and organisms that inhabit unusual, inaccessible and extreme environments. And, of course, these studies need to be conducted under physioxic conditions. We know so little of what really constitutes ‘life’. It has been estimated that we have discovered only 14% of eukaryotic species and only 0.001% of microbial species (Locey and Lennon, 2016; Mora et al., 2011), most likely because many are hiding in inaccessible and anoxic environs. It is from these that we can hope to get a better idea of the relative roles of ROS and RSS in metabolism and signaling.

Signaling by RSS may explain some effector responses that are commonly attributed to ROS. Being able to substitute persulfidation for peroxidation on cysteines that are involved in regulatory functions, as shown in Fig. 1, is strong evidence that RSS signaling can easily mediate effector responses historically attributed to ROS. There are numerous examples of this, and I will illustrate my point with three: oxygen sensing, ROS/RSS metabolism by SOD mimetics, and the Keap1/Nrf2 pathway.

Although there are a number of proposed oxygen-sensing mechanisms, an increase in ROS under hypoxic conditions has historically been one of the most prominent (Sylvester et al., 2012). In this hypothesis (Fig. 3, top panels), hypoxia prevents the reduction of oxygen to water at complex IV, and the resultant increase in H₂O₂ initiates downstream effector responses. In the RSS hypothesis (Fig. 3, bottom panels), hypoxia prevents H₂S catabolism, and the resultant increase in H₂S₂ initiates downstream effector responses (Olson, 2015). Thus, both mechanisms can be explained by an initial effect of hypoxia on complex IV and SOD-mediated production of the signaling moiety.

A number of porphyrin–metal compounds have been synthesized with manganese in the center and modifications of the porphyrins, such that these compounds are extremely efficient SOD mimetics. Three of these, the manganese–porphyrins (MnPs) MnTE-2-PyP⁵⁺ (MnTE, AEOL10113, BMX-010), MnTnHex-2-PyP⁵⁺ (MnHex) and MnTnBuOE-2-PyP⁵⁺ (MnBuOE, BMX-001), are in clinical or pre-clinical trials (Batinic-Haberle and Tome, 2019; Batinic-Haberle et al., 2018). Their efficacy has, paradoxically, been attributed to their ability to redox cycle between oxygen and intracellular reductants such as ascorbate; in doing so, they generate hydrogen peroxide. This hydrogen peroxide then creates oxidative stress that initiates antioxidant responses, which convey cytoprotective effects to healthy cells and are cytotoxic to malignant cells (Batinic-Haberle and Tome, 2019; Batinic-Haberle et al., 2018). Given the SOD mimetic activity of these compounds, we reasoned that they might also metabolize RSS and, indeed, this is the case. We found that all three MnPs oxidize H₂S to polysulfides, first forming H₂S₂ and later H₂S_3–6 (Olson et al., 2019b). Therefore, the effects of the MnPs could also be attributed to the generation of RSS.

It is generally accepted that a moderate increase in oxidative stress increases cellular ‘antioxidant’ defense mechanisms, and that this is mediated to a large extent by nuclear factor erythroid 2-related factor 2 (Nrf2) activation of antioxidant response elements (ARE) in the nucleus (Motohashi and Yamamoto, 2004). Nrf2 is activated after it is released from Keap1 (Kelch-like ECH-associated protein 1) by peroxidation of Keap1 Cys¹⁵¹ in the cytoplasm, allowing Nrf2 to translocate to the nucleus and activate the ARE. Persulfidation of Cyc¹⁵¹ on Keap1 also frees Nrf2, allowing it to translocate to the nucleus and also activate ARE (Guo et al., 2014; Hourihan et al., 2013; Yang et al., 2013). I think one must wonder, given the various effects of antioxidant enzymes in RSS metabolism described above, how much of the Nrf2–Keap1 system actually regulates RSS.

However, the appearance of oxygenic photosynthesis was delayed for over a billion years because the light-gathering antennae of cyanobacteria were not able to capture enough energy to oxidize water. From 2.3 billion years onwards, atmospheric O₂ would episodically rise (to ∼1%) and fall, while the oceans remained essentially anoxic; it was under these conditions that the first eukaryotic organisms appeared 1.5 bya. It took until 0.6 bya for the oceans to finally become oxic. The oxidation of the oceans is suggested to have been the greatest threat to life's existence. This idea, the ox–tox hypothesis, states that organisms existing at this time developed sophisticated antioxidant defenses, retreated to anoxic environs or died (Jones and Sies, 2015; Kurland and Andersson, 2000). The problem with these ideas, as I see it, is that all these antioxidant defense mechanisms – the enzymes SOD and Cat, and the NADPH-dependent catalytic reduction cascades involving Prx, Trx/TrxR and Grx/GrxR – appeared around the time of anoxygenic photosynthesis ∼3.6 bya (Dibrova et al., 2013; Grace, 1990; Hanschmann et al., 2013; Inupakutika et al., 2016; Knoops et al., 2007; Lu and Holmgren, 2014; Miller, 2012; Novoselov and Gladyshev, 2003; Zhang and Weissbach, 2008), when they would have been useful for dealing with RSS, not ROS. Given the chemical similarities between RSS and ROS described above, only minimal adjustments would have been needed for these antioxidant pathways to adapt to the latter; furthermore, these pathways may have retained the capacity to metabolize RSS, and may still do so in extant organisms.

The similarities between ROS and RSS, from the perspective of their chemistry (oxygen and sulfur), their signaling mediators (H₂O₂ and H₂S₂) and their biological targets are striking. This can blur the distinction between ROS and RSS, and it is further confounded by a number of practical methodological considerations involving ROS and RSS detection, the conditions under which the experiments are performed and the organisms involved. Collectively, these factors are so striking that we must take a step back to gain a better perspective of what is actually going on.

Arguably, the most important tenet of comparative physiology is the Krogh principle: ‘For such a large number of problems there will be some animal of choice, or a few such animals, on which it can be most conveniently studied’. The limitation of this philosophy, as I see it, is that it first requires a problem. For those investigators in the ROS field it seems that the problem has been identified; it is ROS, so let's look everywhere for it. I see an inherent danger in this approach; do we see what we are looking at, or what we are looking for? I think this is especially relevant when thinking about ROS and RSS. I hope this Commentary gives comparative physiologists ‘something different to look for’ with respect to the chemical identity of the real reactive species involved, how these species are regulated under physiological conditions, and how they interact with homeostatic effector systems in a variety of organisms. I think this is one of the biggest challenges in physiology and one of the greatest opportunities for comparative physiologists.

K.R.O. wishes to acknowledge the numerous colleagues and students who have contributed to this research.