The radical impact of oxygen on prokaryotic evolution—enzyme inhibition first, uninhibited essential biosyntheses second, aerobic respiration third

Abstract

Molecular oxygen is a stable diradical. All O₂-dependent enzymes employ a radical mechanism. Generated by cyanobacteria, O₂ started accumulating on Earth 2.4 billion years ago. Its evolutionary impact is traditionally sought in respiration and energy yield. We mapped 365 O₂-dependent enzymatic reactions of prokaryotes to phylogenies for the corresponding 792 protein families. The main physiological adaptations imparted by O₂-dependent enzymes were not energy conservation, but novel organic substrate oxidations and O₂-dependent, hence O₂-tolerant, alternative pathways for O₂-inhibited reactions. Oxygen-dependent enzymes evolved in ancestrally anaerobic pathways for essential cofactor biosynthesis including NAD⁺, pyridoxal, thiamine, ubiquinone, cobalamin, heme, and chlorophyll. These innovations allowed prokaryotes to synthesize essential cofactors in O₂-containing environments, a prerequisite for the later emergence of aerobic respiratory chains.

The Great Oxidation Event, GOE [1], divides Earth’s history close to its midpoint. Roughly 2.4 billion years ago (Ga), photosynthetic prokaryotes with two chlorophyll-based photosystems linked in series—cyanobacteria—evolved the molecular tools needed to extract electrons from H₂O, and to use them to fix CO₂ and N₂ for growth [2,3]. Oxygen made by cyanobacteria is ground state triplet O₂, a stable diradical with two unpaired electrons of identical spin. Its structure is better written as ^•O–O^• instead of O=O to underscore its diradical nature [4] (Fig. 1). Radicals are molecules having unpaired valence electrons. They have the property of extracting single electrons from available donors so as to restore a stable octet electron configuration, converting the donor into a new radical in the process. Though most radicals are extremely reactive [5], the O₂ diradical is generally unreactive [6]. It is kinetically stable because each of the unpaired electrons in O₂ is delocalized over a two-center, three-electron π bond, resulting in a very large resonance stabilization energy, and consequently a high activation energy barrier [4]. Despite this, O₂ has a very weak σ bond [4], which renders O₂ an extremely energy-rich molecule [7], so energy-rich that it undergoes exergonic redox reactions with every element except gold [8].

Fig. 1.

A timeline of Earth history with the rise of O₂ and the appearance of relevant groups of prokaryotes. Blue boxes emphasize the GOE, the onset of O₂-dependent enzymes, and questions concerning their evolution and functions. The number 365 represents the number of O₂-dependent reactions in KEGG that we mapped to protein families (see text). SOD: superoxide dismutase; Nox: NADH oxidases (diaphorases), which are oxygen detoxifying enzymes [9]. Data from references [1,2,10–14]. Note that ultra-light carbon in 3.8 Ga rocks can be interpreted as evidence of both archaeal methanogens [15] and bacterial acetogens, which carry the same isotope signal [16], because both fix CO₂ via the acetyl-CoA pathway. A broken line reaching to 10% PAL around the end of the GOE indicates the Lomagundi excursion [10]. The reasons why O₂ levels remained near the Pasteur point for 1.8 billion years are still discussed. Numerous geological causes [10] and one biological cause [12] for the existence of the boring billion have been proposed. It is undebated that cyanobacteria (and their descendants, plastids) generated the current global supply of O₂ via one single enzyme and one single enzyme activity: the conserved Mn₄CaO₅-containing oxygen-evolving complex (OEC) of photosystem II. O₂ is written as •O–O• instead of O=O to underscore its diradical nature [4]. By 3.4–3.3 Ga, anoxygenic photosynthetic prokaryotes were generating stromatolites in aerial settings [17].

From the origin of the first microbes roughly 4 Ga to the onset of the GOE, the Earth’s oceans and atmosphere were effectively devoid of O₂ [10] (Fig. 1). At the GOE, O₂ became introduced into Earth’s oceans and atmosphere to approximately 1% of its present atmospheric level (PAL), and stayed more or less constant at this level for roughly 1.8 billion years. O₂ levels started to rise again about 580 million years ago, approaching modern values with the advent of land plants ca. 450 million years ago [10–13]. This protracted low oxygen phase of Earth history from 2.4 Ga to 0.58 Ga has been called the ‘boring billion’ [14] to emphasize the lack of geologically interesting events during that period of O₂ stasis, but it has also been called the ‘Pasteurian’ era of life’s history [18] to emphasize the crucial observation that O₂ levels of 1% PAL correspond to the Pasteur point—the level of ambient oxygen (ca. 1% PAL or 0.2% v/v) at which facultative aerobes switch their terminal acceptors from anaerobic to aerobic respiration. The “boring” Pasteurian billion was the era during which prokaryotes learned to cope with the reactivity of O₂, to use O₂ for their own benefit, and to evolve alternatives to enzymes that were inhibited by O₂ (Fig. 1).

Prokaryotes were eye witnesses to the GOE. Since they lived before, during, and after the GOE (Fig. 1), their O₂-dependent enzymes should hold clues about the impact of O₂ on physiological evolution. To characterize the impact of O₂ in biochemical evolution, we identified 365 enzymatic reactions of prokaryotes that utilize O₂ in the Kyoto Encyclopedia of Genes and Genomes (KEGG) database [19] and that map to phylogenies of prokaryotic genes [20]. The results provide novel insights into the impact that O₂ exerted on microbial evolution, the nature of physiological traits that O₂-dependent enzymes imparted, and the dispersal of genes for O₂-dependent enzymes via lateral gene transfer (LGT) following the GOE.

Materials and methods

Collection of oxygen-dependent and oxygen-independent reactions

Data for 11 804 metabolic reactions were downloaded from the KEGG [19] reaction database (version 10th August 2022). Additionally, we manually added the reaction linked to superoxide dismutase (SOD; R00275). In KEGG, SOD was linked to an enzyme commission (EC) number, which was in turn linked to the KEGG orthology identifier (KO) for SOD. However, there was no direct link between the SOD reaction and the KO. As there was no discernable reason for this, we added the enzyme SOD and its reaction for the qualitative descriptions of this paper.

The data of the 11 805 reactions were subsequently filtered for reactions involving O₂ (KEGG compound C00007), yielding a set of 1949 O₂-dependent reactions, most of these being specific to eukaryotes. The reactions were mapped to prokaryotic protein families using KEGG orthology to link reactions to sequences. Protein families were created using MCL [21] as previously described [20,22], and only protein families with 4 or more sequences were used in this analysis. Only reactions that we could link to prokaryotic protein families were retained, yielding 365 O₂-dependent reactions mapped to 792 protein families occurring in prokaryotes, referred to as the O₂-dependent reaction set. Of the remaining reactions, a set of 3018 prokaryotic reactions linked to prokaryotic protein families, making up the O₂-independent reaction set (Table S1). See Data accessibility for access to protein family data. For protein family annotation, all clustered sequences were blasted against the KEGG database using Diamond 2.0.1 [23]. All best hits with at least 25% identity and a maximum e-value of 1 × 10⁻¹⁰ were used for annotation. Based on these hits, one KO (KEGG orthology identifier) annotation was assigned to each protein family based on majority rule (Table S2). Protein families which contained at least 75% of unknown sequences without any hits in the KEGG database were not annotated.

Verticality distribution of O₂-dependent and O₂-independent reactions

Verticality values (V) from Nagies et al. [20] were assigned to protein families (Table S1). Verticality describes the relative amount of LGT that a gene family has undergone. High verticality indicates little LGT (vertical evolution, typical for ribosomal proteins) whereas low verticality (typical for O₂-dependent enzymes) indicates abundant LGT. Only prokaryotic protein families spanning 2 or more prokaryotic phyla had an associated verticality value [20]. All the phylogenetic trees underlying this analysis are published in supplementary table 9 in Nagies et al. [20] and available under https://doi.org/10.25838/d5p-15. The distribution of verticality was generated for both the O₂-dependent and the O₂-independent reaction sets. The O₂-independent reaction set contained 3018 reactions and 8322 protein families with associated verticality values, while the O₂-dependent reaction set contained 365 reactions and 547 protein families with associated verticality values. Verticality values of all protein families associated with O₂-dependent and O₂-independent reactions, as well as the average verticality of all protein families associated with each reaction (for O₂-dependent and O₂-independent reactions) were plotted against the number of genomes in which the protein family/reaction was detected (Fig. S1). Finally, we counted the number of O₂-dependent and O₂-independent reactions in the largest genome of each species in our dataset and applied regression models (see Fig. S2 and Table S3A).

Calculation of Gibbs energy and reactant count

The change in Gibbs energy ΔG was calculated for each reaction of both O₂-dependent and O₂-independent sets using eQuilibrator API [24] version 0.4.1. Calculations were performed for physiological conditions (pH 7, 1 mM concentrations of reactants and products, 25 °C, ionic strength 250 mm). In each reaction, O₂ was always written as a reactant (C00007 on the left side of the reaction), such that reaction R00009, R02550, R05229, and R00275 were reversed prior to calculation. In total, eQuilibrator yielded ΔG for 288 O₂-dependent reactions and for 2139 O₂-independent reactions (Table S4).

Reactants and products of O₂-dependent reactions were counted, with reactions written in the direction of O₂ consumption. For this count, only two ROS scavenging reactions were written in the O₂-evolving direction based on their physiological function: R00009 and R02670 (annotated as catalase). See Table S5 for the most common substrates and products of O₂-dependent reactions, and Table S6 for an overview of H₂O₂-consuming and H₂O₂-evolving reactions.

Identification of cofactors

Cofactors for each reaction were identified by integrating data from the IUBMB Comments section of the BRENDA database [25], the EC subclass descriptions if applicable, and literature data associated with each KO entry. In some cases, original literature not listed in KEGG was consulted, the references for which were included in Table S7. The BRENDA database was queried via EC number, while the KEGG literature data for each enzyme was accessed via the KOs associated with the corresponding reaction. In case of discrepancies, literature data was prioritized.

Only cofactors bound by the protein subunits corresponding to the KO annotation were listed. When there were multiple possibilities for the cofactors of an enzyme/subunit/chain, all were listed. Cytochromes and ferredoxins were listed as cofactors only if they were bound by the enzyme as soluble electron carriers. For cytochrome or ferredoxin enzyme domains, the cofactors listed were heme and iron–sulfur clusters, respectively. The reactions sometimes explicitly include an electron donor that provides electrons to the main enzyme indirectly, e.g., through an additional reductase component. Such electron donors were not listed, since they are not immediate ligands of the enzyme in question. In some cases, the natural electron donor was unknown and was therefore not listed. In addition, in cases where the cofactors were substrates or products in the reaction, they were not listed. Cosubstrates, such as 2-oxoglutarate in 2-oxoglutarate non-heme iron-dependent oxygenases, were also not listed.

Occurrence of functional categories per prokaryotic phylum

Using the KO identifiers, each protein family was assigned to at least one functional category (B-level) according to the KEGG BRITE classification. The distribution and occurrence of functional categories was examined for each phylum within all analyzed protein families linked to oxygen-dependent reactions (plotted in Fig. S3).

Statistical analysis

The calculated ΔG of the O₂-dependent and O₂-independent sets were compared with a t-test [26]. Comparisons of the number of O₂-dependent reactions per genome in aerobes and anaerobes and of genome size between aerobes and anaerobes (classified based on [27]) were done with Welch’s t-test [26] (see Table S3B). In Fig. S2 the size of the largest genome of each species in the dataset was plotted against the number of O₂-dependent and O₂-independent reactions associated with the species. Regression models were chosen based on Pearson’s product–moment correlation resulting in a linear model for O₂-dependent values and a logarithmic model for O₂-independent values (see Table S3A).

To compare the distributions of verticality values of O₂-dependent protein families and O₂-independent protein families, a Welch’s t-test was performed on an unfiltered (Table S3C) and a filtered (Table S3D) dataset, in which high verticality values (> 1) were removed. This comparison was corroborated by subsampling the protein families associated with O₂-independent reactions. For this subsampling procedure, 100 000 samples of 547 O₂-independent protein families were generated to compare to the 547 protein families of O₂-dependent reactions. The average verticality (V_avg) was calculated for each sample. With this, an estimate was possible of how likely it would be to create a sample with a mean verticality as low as that of the O₂-dependent distribution from values in the O₂-independent distribution. This subsampling procedure was repeated, but changed each time to mitigate potential effects of unequal protein family size distributions. For this, four bins of equal size for protein family sizes (number of genomes in a protein family) of the O₂-dependent reactions were defined in the range of 4 genomes (minimum protein family size to calculate phylogenetic trees) to 3380 genomes. For each bin, the number of protein families in the O₂-dependent distribution was counted, and during sampling, bins were filled up with an equal number of protein families from the O₂-independent distribution (see Table S3E).

Results

O₂ arose in a world without respiratory oxygen reductases

O₂-dependent enzymes are most frequent in facultative anaerobes with large genomes [28,29] (Tables S3, S8, and S9, Fig. S4). Plots of the frequency of O₂-dependent and O₂-independent reactions in relation to genome size are presented in Figs S2 and S5. Although respiratory oxygen reductases are not the focus of our study, it is important to keep their evolutionary significance in perspective. In today’s environment, the over-whelming majority of O₂ produced by cyanobacteria and plants is used for respiration and energy conservation in aquatic and terrestrial environments, keeping the global carbon cycle in balance and our atmospheric O₂ levels largely constant [30], forest fires and human intervention notwithstanding. But that was clearly not the case before the GOE. The goal of our study is to gain insights into the role of O₂ for prokaryotes at the time of the GOE.

For the purpose of this study, we assume that at the time of the first appearance of environmental O₂ during the GOE, functioning respiratory oxygen reductases (or other O₂-dependent enzymes) had not yet evolved. Contrary to that view, it has been proposed that O₂ reductases evolved from a more ancient family of NO reductases, such that oxygen-utilizing terminal oxidases were already present when O₂ first appeared [31]. However, that proposal is not supported by current evidence [32–38], nor have similar proposals been put forward for other O₂-utilizing enzymes. There are three large and evolutionarily unrelated (independently arisen) superfamilies of oxygen reductases in respiratory chains: (a) the heme copper oxidases (HCO) [32,33] that include cytochrome c oxidases, (b) the heme-containing cytochrome bd oxidases that oxidize membrane quinols [34–36], and (c) the alternative oxidase superfamily (AOX) of non-heme diiron proteins that oxidize quinols [37,38]. Because members of the bd oxidase and AOX superfamilies only have one known electron acceptor substrate, O₂ can be directly inferred as the original substrate for the founding members of those enzyme families following the GOE. Only the HCO superfamily contains members that react with another electron acceptor substrate, the nitrous oxide (NO) reductases [27,33]. Recent surveys with broad sampling have clearly shown that NO reductases evolved multiple times and independently to produce the current spectrum of HCO terminal oxidases [33]. This indicates that O₂ was also the ancestral substrate for the HCO family, as in the case of the bd oxidases and AOX, in line with the low bioavailability of copper (an essential cofactor of HCO enzymes) prior to the origin of environmental O₂ [32].

This indicates, in turn, that terminal oxidases, like all other O₂-dependent enzymes, arose in environments that were experiencing a gradual encroachment of O₂ into an anoxic world. As O₂ first diffused into the environment, it led to oxidation of one-electron donors and enzyme inhibition, but it also introduced a novel, energy-rich oxidant into the trajectory of biochemical evolution. Once the HCO, bd, and AOX families of terminal oxidases arose, they rapidly diversified into new subfamilies [33,35,38]. Moreover, prokaryotes that possessed terminal oxidases came to flourish in heterotrophic settings, amplifying both the gene copy number and protein abundance of terminal oxidases in the environment over evolutionary time. Today terminal oxidases are, in terms of substrate turnover, unchallenged as the main consumers of O₂ in contemporary environments. In modern oceans, for example, photosynthetic oxygen production and respiratory O₂ consumption occur at roughly equal rates [39]. In mammals, 90% of oxygen consumption is mitochondrial respiration [40]. Respiration requires quantitatively large amounts of terminal oxidases, which are often complex membrane proteins.

In the present study our concern is not the increase of protein abundance from the GOE to today, rather, the situation at the onset of O₂ accumulation in Earth history and how O₂ could have impacted enzyme origin in the microbes that first encountered O₂. A situation very similar to that found in terminal oxidases is encountered with another O₂-utilizing (but not strictly O₂-dependent) enzyme, one involved in CO₂ fixation, RuBisCO. Today RuBisCO is not only the most common CO₂ fixing enzyme, it is the most abundant protein in nature [41]. The Calvin cycle in which it operates is the most recently evolved [42], the least energy efficient [43] and by far the most widespread of CO₂-fixing pathways. But it did not start that way. RuBisCO’s original function was not in the Calvin cycle but in fermentative breakdown of RNA [44–46]. Changes in Earth’s environment, in particular O₂, impacted the evolutionary trajectory of enzymes that play the central role in Earth’s modern carbon cycle: terminal oxidases that consume O₂ and RuBisCO that fixes CO₂ with electrons from photosynthetic O₂ production. But what was the impact of O₂ on other enzymes and pathways in the immediate wake of the GOE? We asked O₂-dependent enzymes.

O₂ is an energy-rich compound, but unused in SLP

Oxygen is a strong oxidant, with a midpoint potential of +815 mV for the O₂/H₂O pair at pH 7. The enthalpy change in combustion reactions of O₂ with organic compounds (generating CO₂) is typically on the order of −400 kJ·mol⁻¹ of O₂, regardless of the organic compound undergoing combustion, because the energy released is almost entirely the energy stored in the energy-rich O₂ molecule, not in the organic reactant [7]. We estimated the free energy change, ΔG, for 288 of the 365 reactions that use O₂ as a substrate and that map to protein families (Table S1). The ΔG values are presented in Table S4 (along with the values estimated for 2139 O₂-independent reactions; see also Fig. S6), with an average calculated change in Gibbs free energy under physiological conditions of –234 kJ·mol⁻¹ of O₂. This value is sufficiently exergonic to drive substrate-level phosphorylation (SLP). It is therefore a curious observation that strictly O₂-dependent SLP reactions are virtually unknown. A change in free energy of roughly –70 kJ·mol⁻¹ is needed for the synthesis of one ATP [47], hence there is enough energy to synthesize at least one ATP (or more) per O₂-dependent oxidation of organic substrate on average. Out of 365 prokaryotic O₂-dependent reactions, only one (0.3% of the total) generates a product capable of supporting SLP: that catalyzed by the H₂O₂-producing (phosphorylating) pyruvate oxidase (EC 1.2.3.3) [48], POX. This enzyme converts pyruvate and O₂ to CO₂, H₂O₂ and acetyl phosphate with a calculated ΔG of –158 kJ·mol⁻¹. H₂O₂-producing pyruvate oxidase has a very narrow phylogenetic distribution, occurring almost exclusively among members of the Lactobacilli, and the role of O₂ is to oxidize a flavin in the reaction mechanism, not pyruvate itself. If O₂ were an evolutionary vehicle to improve energy yield—a widely held premise about the role of oxygen in evolution [49,50]—why would cells not conserve the ca. –234 kJ·mol⁻¹ of free energy released in > 99% of non-respiratory O₂-reducing reactions? The absence of O₂-dependent SLP reactions suggests that at the onset of the GOE, when O₂ first became available as a substrate, the main initial role of O₂ in microbial evolution was not immediately energy conservation, otherwise cells would likely have found ways to conserve the energy released in reactions of O₂ with organic compounds. The lack of O₂-dependent SLP reactions is noteworthy. If energy was not the initial functional role for oxygen, what then?

Oxygen-dependent enzymes often act on aromatic substrates

Gene-based studies addressing the physiological role of O₂ in evolution typically focus on its use as a terminal electron acceptor in respiratory chains [51,52], but terminal oxidases represent only about 1% of known prokaryotic O₂-utilizing enzyme families [28], hence they depict only one aspect of O₂ impact on prokaryotic metabolism. The responses of anaerobic microbes to O₂ via O₂ detoxification enzymes such as NADH oxidases and rubredoxin:oxygen oxidoreductase are well studied [6], as is the impact of O₂ on prokaryotic gene expression via DNA-binding proteins [53] such as the FeS cluster-containing O₂ sensor FNR (for fumarate and nitrate respiration) [54] and the ArcAB two-component system (for anoxic redox control or aerobic respiratory control), which responds to the oxidation state of the quinone pool [55]. We found that the most common substrates of O₂-dependent reactions are redox cofactors (ferredoxins, NAD(P)H, and flavins), water and 2-oxoglutarate, while common products include water, H₂O₂, CO₂, and ammonia (Tables S5 and S6).

The main utility of O₂ in biochemical reactions is that of a strong oxidant, affording microbes that possess the corresponding genes metabolic access to chemically stable substrates in the presence of O₂. Reactions of triplet O₂ in biological systems always involve a radical mechanism and almost always involve extraction of one electron from a substrate, generating a radical to initiate the mechanism [56]. The most frequent reaction types of prokaryotic O₂-dependent enzymes in this sample are aromatic degradation and amine oxidation (Table 1). Single electron donor dioxygenases, which incorporate both atoms of O₂ into the reaction product (EC 1.13.11.-), represent the most common enzyme category (n = 57), followed by NAD(P)H-dependent monooxygenases (n = 56), which incorporate one atom from O₂ into the product (EC 1.14.13.-). The enzymes from both these groups act mainly on aromatic substrates (Table 1). Dioxygenases typically disrupt aromatic rings [57]. The next most common enzymes are amine oxidases acting on CH–NH₂ groups (EC 1.4.3.-) (n = 48), copper or flavin containing proteins that catalyze the oxidation of primary amines, polyamines and amino acids [58]. NAD(P)H-dependent dioxygenases (EC 1.14.12.-), all of which act on aromatic substrates [59], are the next most common category (n = 38), followed by FeS-dependent monooxygenases [60] (n = 28) (EC 1.14.15.-) (Table 1).

Table 1. Most common prokaryotic O₂-dependent reactions.

n: number of reactions; Dearo: number of dearomatizing reactions (the reaction destroys carbon ring aromaticity); Substr. Arom.: number of reactions where the substrate is an aromatic compound (excludes cosubstrates such as NAD(P)H, flavin, etc.); NH₃ Prod.: number of reactions releasing ammonia/ammonium ion as a reaction product; Dioxygenase: both oxygen atoms incorporated into substrate; Monooxygenase: one oxygen atom incorporated into substrate; Diverse: O₂ can be incorporated or reduced; Acceptor: O₂ is reduced, not incorporated.

EC number	n	e− donor (substrate)	Fate of O2 atoms	Dearo.	NH3 prod.	Substr. Arom.
1.13.11.-	57	Single donor	Dioxygenase	40	0	45
1.14.13.-	56	Donor + NAD(P)H	Monooxygenase	4	0	38
1.4.3.-	48	CH-NH2 group	Acceptor	0	27	21
1.14.12.-	38	Donor + NAD(P)H	Dioxygenase	28	2	38
1.14.15.-	28	Donor + FeS cluster	Monooxygenase	0	0	3
1.14.99.-	23	Misc. paired donors	Diverse	0	0	7
1.14.19.-	21	Paired donors	Acceptor	0	0	7
1.14.14.-	19	Flavin and Donor	Monooxygenase	0	0	6
1.1.3.-	14	CH-OH group	Acceptor	0	0	3
1.14.11.-	12	Paired donors	Diverse	0	0	2
1.5.3.-	8	CH-NH group	Acceptor	0	0	4
1.14.18.-	7	Paired donors	Monooxygenase	2	0	6
1.13.12.-	7	Single donor	Monooxygenase	0	0	4
1.14.20.-	6	Donor +2-OG	Diverse	0	0	2
1.10.3.-	6	Diphenols	Acceptor	2	0	5
1.3.3.-	5	CH-CH group	Diverse	2	0	3
7.1.1.-a	3	Quinones (or cyt c)	Acceptor	0	0	3

^aGiven out of order in the ranking, there are EC categories more common than the translocases (7.1.1.-, terminal oxidases) in the data, but the terminal oxidases are important, hence added to the list. EC numbers and links to all KEGG reactions are provided in Table S10.

Of the 365 O₂-dependent KEGG reactions that mapped to protein families, more than 50% act on aromatics, stable substrates that require either a strong oxidant or a strong reductant [61] to disrupt the aromatic ring (Table 1). However, aromatic degradation in prokaryotes does not require O₂ [59,61,62]. Two O₂-independent routes of aromatic degradation via a benzoyl-CoA intermediate are catalyzed by unrelated benzoyl-CoA reductases. One route employs flavin-based electron bifurcation to generate midpoint potentials sufficiently negative to reduce benzoyl-CoA $(E_{\mathrm{o}}^{\prime }=-622\text{mV})$ [61]. Importantly, the growth rates of aerobic and anaerobic aromatic-degrading microbes are very similar, both having doubling times on the order of 4–6 h [59]. This indicates that the advantage conferred by the O₂-dependent pathway of aromatic degradation is not more rapid growth.

The evolutionary rationale behind the origin of O₂-dependent aromatic degradation might reside in the O₂-sensitivity of the evolutionarily older anaerobic enzymes. The O₂ sensitivity of solvent-exposed FeS clusters is a central underlying theme of O₂ in biochemical evolution [9,63–71]. The anaerobic ATP-dependent benzoyl-CoA reductase from Thauera aromatica contains three O₂-sensitive FeS clusters [72], while the electron-bifurcating benzoyl-CoA reductase enzyme complex from Geobacter metallireducens contains over 50 FeS clusters [61]. The more ancient, O₂-sensitive enzymes cannot function in oxic environments, requiring the origin of enzymes with an alternative reaction mechanism that can operate in oxic habitats [73]. The solvent-exposed nature of FeS clusters is important for their inhibition by O₂. Yet many FeS clusters in proteins are not solvent-exposed and not O₂-sensitive, such that the mere presence of an FeS cluster in a protein is not always a proxy for O₂ sensitivity. Human mitochondrial complex I contains eight FeS clusters, and photosystem I contains three, but neither is inhibited by O₂. In order for O₂ to oxidize an FeS cluster, it has to attain physical proximity, which is possible in the case of solvent-exposed clusters.

Oxygen-dependent enzymes confer novel physiological traits

The frequency of each enzyme (protein family) among the genomes in our sample is shown in Table 2 for the 30 most common O₂-dependent enzymes (the full list is given in Table S2). The most widespread O₂-dependent enzymes are either terminal oxidases, or enzymes employed in (cofactor) biosynthesis, detoxification, and substrate mobilization. In primary metabolism, the O₂-dependent biosynthetic reactions present post-GOE alternatives to O₂-independent reactions that existed in cells before the GOE. The oxidation of amino acids, sulfonates, or 2-hydroxy acids also present alternatives to preexisting anaerobic pathways.

Table 2. Most widely distributed prokaryotic O₂-dependent enzymes.

KO: identifier of the KEGG orthology group used for protein family annotation; N_g: number of genomes; V: verticality value (see Materials and methods). Enzyme function abbreviations are in brackets. Numbers in square brackets correspond to protons pumped (data from references [33,34]). The 7,8-dihydroneopterin oxygenase activity (K01633) is an oxygenase side reaction that proceeds through a carbanion intermediate, generating 7,8-dihydroxanthopterin, which is not a central intermediate in the folate synthesis pathway [74].

Enzyme name	EC number	KO	N g	V
7,8-Dihydroneopterin oxygenase	1.13.11.81	K01633	4424	0.90
Cytochrome bd ubiquinol oxidase subunit II [0] (Oxphos)	7.1.1.7	K00425	4075	0.63
Cytochrome bd ubiquinol oxidase subunit I [0] (Oxphos)	7.1.1.7	K00426	4074	0.57
Catalase (O2 detoxification)	1.11.1.6	K03781	3485	0.30
L-Aspartate oxidase (NAD+ synthesis; N mobilization)	1.4.3.16	K00278	3388	6.22
Pyridoxamine 5′-phosphate oxidase (PLP synthesis)	1.4.3.5	K00275	3026	0.11
Nitronate monooxygenase (N mobilization)	1.13.12.16	K00459	2974	0.34
Bacterioferritin (Iron storage)	1.16.3.1	K03594	2867	0.52
Glycolate dehydrogenase FAD-linked SU (2-OH acids)	1.1.99.14	K00104	2612	0.36
Cytochrome o ubiquinol oxidase subunit I [2] (Oxphos)	7.1.1.3	K02298	2561	0.14
Coproporphyrinogen III oxidase (Heme synthesis)	1.3.3.3	K00228	2473	1.18
Cytochrome c oxidase subunit I [4] (Oxphos)	7.1.1.9	K02274	2448	0.86
Cytochrome c oxidase subunit II [4] (Oxphos)	7.1.1.9	K02275	2435	0.39
2-polyprenylphenol 6-hydroxylase (UQ synthesis)	1.14.13.240	K18800	2431	1.49
Cytochrome o ubiquinol oxidase subunit II [2] (Oxphos)	7.1.1.3	K02297	2406	0.57
4,5-DOPA dioxygenase (Amino acids & aromatics ox.)	1.13.11.-	K15777	2266	1.15
Catalase-peroxidase (substrate oxidation and O2 detox)	1.11.1.21	K03782	2197	0.14
Superoxide dismutase, Cu-Zn family (O2 detoxification)	1.15.1.1	K04565	2180	1.07
2-octaprenyl-6-methoxyphenol hydroxylase (UQ synth.)	1.14.13.-	K03185	2157	0.72
Malate dehydrogenase, quinone (TCA cycle)	1.1.5.4	K00116	2095	0.31
Glycine oxidase (Thiamine biosynthesis)	1.4.3.19	K03153	1886	0.97
(S)-2-hydroxy-acid oxidase (2-OH acids, glycolate DH)	1.1.3.15	K11473	1849	0.90
Alkanesulfonate monooxygenase (S mobilization)	1.14.14.5	K00299	1831	0.09
Gamma-glutamyl putrescine oxidase (Amino acid ox.)	1.4.3.-	K09471	1809	0.09
tRNA 5-MAM-2-thiouridine bifunctional protein (tRNA)	1.5.-.-	K15461	1789	0.18
4-Hydroxyphenylpyruvate dioxygenase (Amino acid ox.)	1.13.11.27	K00457	1699	0.38
Alkanesulfonate monooxygenase (S mobilization)	1.14.14.5	K04091	1660	0.02
3-Phenylpropanoate dioxygenase (Amino acid ox.)	1.14.12.19	K00529	1540	0.28
4-Hydroxyphenylacetate 3-monooxygenase (AA ox.)	1.14.14.9	K00484	1539	0.09
Protoporphyrinogen oxidase (Heme synthesis)	1.3.3.4	K00231	1488	0.62
Cytochrome c oxidase subunit III [4] (Oxphos)	7.1.1.9	K02276	1433	0.49

The same principle holds for the HCO, bd and AOX superfamilies of terminal oxidases [27,51,75]. The family of cytochrome bd oxidases became integrated into previously-existing anaerobic electron transfer chains that used terminal acceptors such as sulfite and metals, producing a branching from membrane quinols to oxygen that is very common in extant facultatively anaerobic prokaryotes [34,51]. The appearance of HCOs introduced a branching in the electron transfer chains that pivot on c-type cytochromes and often use metals as source of electrons [76]. Large-scale phylogenetic studies of the HCO superfamily indicate that the most ancient HCOs are the A-type HCOs [33], which, like bd oxidases and AOX, normally do not reduce NO, as outlined above. The cytochrome bd oxidases are found among many archaeal groups and might be an ancient lineage of oxygen reductases, but their evolutionary history is, like that of HCOs, complicated by lateral gene transfers [32–38,75,76]. Though the relative ages of terminal oxidase families are not clarified because the protein families arose independently, and because the history of all three families has been affected by LGT [32–34,38], sulfide provides hints. HCO terminal oxidases require copper, which has very low bioavailability in sulfidic environments [75], and are strongly inhibited by sulfide [77,78], whereas bd oxidases are sulfide tolerant [79,80] and do not require copper. This would speak in favor of a greater antiquity for the bd family. The AOX family is probably the youngest of the three oxygen reductase superfamilies, based on its very low affinity for O₂ [51].

In the present sample of prokaryotes, cytochrome bd ubiquinol oxidases appear to be the most common O₂-reducing terminal oxidases (Table 2). These transmembrane enzymes usually oxidize quinols but do not translocate protons across the membrane [34] although they do generate proton motive force via scalar protons (the localization of proton-consuming and proton-generating reactions on opposite sides of the membrane).

Another novel class of enzymes that arose in response to O₂ are detoxification enzymes that scavenge ROS (reactive oxygen species) that result from reactions of O₂: catalases [81], catalase-peroxidases [82], superoxide dismutases [83], and other detoxification enzymes [9], some of which are not represented in KEGG. The main cofactors of the ROS scavenging enzymes are metals [9,81–83]. Their main function is detoxification of reactive oxygen species such as the superoxide radical, ${\text{O}}_2^{·-}$, that forms from one-electron transfers during O₂-dependent metabolism, or hydrogen peroxide, H₂O₂. Many novel enzymatic functions that did not exist prior to the existence of O₂ are found in the biosynthesis and degradation of secondary metabolites [28,84,85].

Genes for O₂-dependent enzymes undergo LGT more frequently than others

Evidence for O₂ accumulation in Earth history is geochemical (Fig. 1). Within that temporal framework, genes for O₂-dependent reactions can be inherited vertically, in which case their advantage is realized only within line-ages, or they can be inherited via lateral gene transfer, in which case their advantage can be transmitted to any line-age. Several gene-based studies of O₂ in evolution address the timing of O₂ appearance [81,83,84] by using molecular clocks to date the age of O₂-dependent enzymes. Dating the appearance of O₂ with phylogenies of O₂-metabolizing enzymes requires, however, that the genes have not been subject to lateral gene transfer during evolution. A recent molecular-clock-based study of oxygen in evolution embraced the vertical view and ascribed the origin of oxygen-dependent enzymes to a single hypothetical lineage called the last universal oxygen ancestor [81]. In the real world of microbial genomes, it is known that all genes in prokaryotic genomes can be transferred and that most, if not all, have been subject to LGT at some point in evolution [20,85–87]. O₂-dependent enzymes are no exception.

If an O₂-dependent enzyme confers the ability to survive in oxic habitats, and if oxic habitats became more widespread across Earth history (Fig. 1), what is useful for one microbe can be useful for another. This predicts that O₂-dependent enzymes should readily spread across lineages via LGT during evolution. To quantify the role of LGT in genes for O₂-dependent enzymes, we utilized values of verticality, V, which provide a measure for how often a gene has been transferred between lineages in evolution [20]. Verticality measures only LGT events across major taxonomic boundaries (phyla or domains), evolutionarily recent fine-scale transfers are ignored in the calculation of verticality such that it provides a robust measure of LGT frequency [20]. High values of verticality indicate low levels of LGT for members of a gene family, for example ribosomal proteins, while low values of verticality reflect a high frequency of LGT for a given prokaryotic gene during evolution.

For the 547 protein families of O₂-dependent enzymes in prokaryotes for which we could assign a value of V, the mean verticality or V_avg is 0.273 ± 0.626 (avg ± SD). For comparison, the 3018 O₂-independent reactions in the data map to 11 754 protein families, of which 8322 had an associated verticality value (Table S1). The O₂-independent enzymes have a mean verticality of V_avg = 0.781 ± 1.926. Although the means overlap, the difference in the two distributions is highly significant at P = 7.43·10⁻⁴⁷ (t-value = −14.92, DF = 1406; Fig. 2). The difference remains significant even when higher verticality values (> 1) are filtered out (V ≤ 1, t-value = −6.22, DF = 603, P = 9.02·10⁻¹⁰; see Table S3C,D).

Fig. 2.

Distribution of verticality values for protein families catalyzing oxygen-independent and oxygen-dependent reactions. The distribution of 8322 protein families with verticality values associated with O₂-independent reactions are shown in the upper histogram (orange) and the 547 protein families with verticality values associated with O₂-dependent reactions in the lower histogram (blue) in logarithmic scale, 100 bins. O₂-dependent gene families show a lower verticality, which means they have been subject to more LGT. For details of the statistical procedures see Materials and methods and Table S3.

Figure 2 shows that genes for O₂-dependent enzymes underwent more LGT in their evolution (expressed in lower verticality values) than genes for O₂-independent enzymes. Even the more widely distributed protein families and reactions have comparable low verticality (Fig. S1). The excess of more vertical protein families among O₂-independent enzymes (Fig. 2, upper right) is the signal of vertically evolving enzymes, many of them involved in information processing, such as aminoacyl-tRNA synthetases (for the complete list see Table S1). Genes for ribosomal proteins, though among the most vertically evolving genes in prokaryotes [20], are not plotted in Fig. 2 because their products do not catalyze chemical reactions, hence they are not represented in KEGG.

Figure 3 underscores the small contribution of vertical evolution in prokaryotic genes that code for enzymes involved in O₂-dependent reactions (blue points) relative to genes that are currently in use for molecular systematics studies, including ribosomal and other information processing proteins (black points), and genes that code for O₂-independent reactions (red points). The genes for O₂-dependent enzymes have almost no tendency at all to undergo vertical evolution, they are freely passed around lineages, where they can become fixed or not.

Fig. 3. Genes for enzymes catalyzing O₂-dependent reactions tend to be laterally transferred. Comparison of number of phyla in which a protein family is present, number of genomes where it occurs and verticality values for prokaryotic protein families.

(A) The number of phyla (x-axis) is plotted in relation to the number of genomes (z-axis) and verticality values (y-axis) for 36 highly vertical and universal informational core genes (black) and for the 8322 O₂-independent protein families with associated verticality values (red). (B) The number of phyla (x-axis) is plotted in relation to the number of genomes (z-axis) and verticality values (y-axis) for 36 highly vertical and universal informational core genes (black) and the 547 protein families corresponding to oxygen-dependent reactions with associated verticality values (blue). The following functions correspond to the 36 informational core (IC) genes and are sorted by verticality: rpsJ, rpsK, rplA, alaS, rplB, ffh, rpsE, fusA, ftsY, hisS, truB, metG, rplN, pyrG, rpsH, rpsI, prsA, valS, rplE, rplF, tsaD, argS, truA, ychF, pyrH, uppS, ksgA, rpsS, serS, glyA, tuf, ileS, rpsL, eno, pgk, cyxS. (C) Average verticality values for O₂-dependent (blue points) and O₂-independent (red points) protein families across the 10 functional categories with the highest frequency of O₂-utilizing enzymes. The last column shows the ratio of average verticality between O₂-independent and O₂-dependent protein families per functional category. Values for all functional categories can be found in Table S12, and alternative plots showing the average verticality per functional category, including error bars, can be found in Fig. S7.

Within the 10 functional categories in KEGG with the highest frequency of O₂-dependent enzymes, the most frequently transferred O₂-dependent genes relative to O₂-independent genes within the same functional category encode products involved in the metabolism of ‘other’ amino acids (D-amino acids, glutathione, taurine, selenocompounds, etc.), followed by the functional category ‘Protein families: metabolism’ (which includes various catabolic reactions), genes coding for enzymes involved in xenobiotics degradation (includes cytochrome P₄₅₀ enzymes), amino acid oxidation and breakdown of secondary metabolites (Fig. 3, Figs S3 and S7). In lipid metabolism, fatty acid and carotenoid oxidation, O₂-dependent enzymes are common, as these act on non-fermentable substrates. A summary of transfers across the nodes of a phylogenetic tree is presented in Fig. S8 (see also Table S11).

Evolutionary rationale behind O₂-dependent synthesis of essential cofactors

The most common genes for O₂-dependent enzymes we identified (Table 2) encode steps in the synthesis of essential cofactors including NAD⁺, pyridoxal phosphate (PLP), heme, ubiquinone (UQ), and thiamine (Thi). There are also O₂-dependent and O₂-independent biosynthesis pathways for cobalamin and chlorophyll [73,88,89]. Why should organisms evolve O₂-dependent pathways for the biosynthesis of essential cofactors in the presence of preexisting O₂-independent pathways? One suggestion is that O₂-dependent pathways evolved in the presence of the older anaerobic pathways because they are favored for thermodynamic and kinetic reasons [90]. But the observation that microbial growth rates are similar for O₂-dependent and O₂-independent aromatic degradation [59] offers no hints that the thermodynamics or kinetics of the O₂-dependent pathways confer advantage. Why are there two kinds of pathways for seven essential cofactors?

In the case of parallel O₂-dependent and O₂-independent pathways for chlorophyll synthesis, Chew and Bryant [73] reached a conclusion that is probably applicable in most, if not all, cases of O₂-dependent and O₂-independent alternative pathways: “When a powerful selective pressure, oxygen, apparently inactivated oxygen-independent enzymes in (B)Chl biosynthesis, unrelated proteins with the same catalytic function but with completely different structures and mechanisms, sometimes even using oxygen as a substrate, then evolved.” In other words, the inactivation by O₂ of an anaerobe’s preexisting enzyme generated the selection pressure for the evolution of an alternative enzyme that tolerated the presence of O₂, sometimes even by using O₂ in the radical-dependent mechanism. We suggest that the same evolutionary reason probably applies generally to the origin of O₂-dependent pathways for the synthesis of essential cofactors (or other essential functions) in the presence of preexisting O₂-independent pathways.

The O₂-independent pathways for essential cofactors must be older than the O₂-dependent pathways because prokaryotes that existed prior to the origin of oxygenic photosynthesis undoubtedly required and synthesized NAD⁺ [91], PLP [92], ubiquinone, UQ [93,94], thiamine [95,96], heme [88], chlorophyll [73] and cobalamin [88,89]. The preexisting O₂-independent routes typically involve enzymes with O₂-sensitive low-potential metal centers, such as [4Fe4S] clusters, O₂-sensitive pathway intermediates, or both. The reactions tend to rely on radical mechanisms, which are known to be common to many strict anaerobes [5]. Although the mechanism of O₂ activation is not known for many oxygen-dependent enzymes, the chemistry of O₂ in biological reactions prescribes that one-electron activation is necessary, which is why O₂ can interfere as an inhibitor of other biological radical reactions by extracting the radical. O₂-dependent biosynthetic pathways for essential cofactors offered microbes the tools needed to colonize oxic habitats, from which they would otherwise have been excluded. Enzymes with mechanisms of O₂ tolerance that do not involve O₂ as a substrate fall outside the scope of this paper, but the evolutionary rationale behind O₂ tolerance would also apply. The principle of oxygen’s impact in evolution—inhibition first, then the ability to grow, then respiration—is illustrated in Fig. 4 (and highlighted in the title).

Fig. 4.

Inhibition, growth, respiration. Several essential cofactors have O₂-inhibited and O₂-dependent biosynthetic pathways. The anaerobic pathways are older, obviously, because the cofactors shown are essential to various groups of prokaryotic anaerobes that existed prior to the origin of oxygenic photosynthesis. For references to the pathways see text. O₂-inhibited pathways preclude colonization of oxic niches (“do not enter” signs). Auxotrophy for the cofactor in an oxic niche is only an option if another group in the oxic niche has already evolved an O₂-tolerant biosynthetic route, in which case gene acquisition via LGT becomes an alternative to auxotrophy. The geological time of origin of oxygen respiration is not an issue here, only the relative timing of cofactor biosynthesis, which enables growth in an oxic environment, followed by origin or acquisition of O₂ respiratory terminal oxidases.

Flavins and iron are the most common cofactors across O₂-dependent reactions

As a consequence of its kinetically stable triplet diradical state, dioxygen has high activation energy barriers [4]. Because organic substrates usually exist in their singlet ground state, direct reactions of triplet oxygen with organic compounds are spin-forbidden [97] such that oxygen needs to be activated in order to react with typical organic substrates. In enzymatic reactions, activation is usually provided by the enzyme-guided donation of a single electron to generate a superoxide radical O₂^·− or by an electron and a proton to generate a perhydroxyl radical HO₂^·. The one-electron donor is typically either a metal ion such as copper, iron (sometimes in the form of FeS centers), or manganese [98], or an organic cofactor such as a flavin or a pterin [99]. O₂ so activated by transfer of a single electron is extremely reactive, in O₂-dependent enzymes it readily oxidizes a specific substrate at the active site. The most common cofactors for the 365 reactions for which data could readily be identified are given in Table 3 (complete list in Table S7).

Table 3. The most common cofactors across 365 O₂-dependent reactions of prokaryotes. One-electron donors are marked with an (*).

Cofactors	Frequency
Flavins*	156
Iron*	152
NAD(P)+	84
Ferredoxin*	61
Iron–sulfur clusters*	57
Hemes*	39
Copper*	22
Quinones*	17
No cofactor or unknown	21
Coenzyme A	11
Cytochromes*	7
Pterins	5
Nickel	3
Flavodoxin*	3
Rubredoxin*	3
Manganese*	2
Tetrahydrofolate	2
Zinc	2
Selenium	1
Magnesium	1
S-adenosyl methionine*	1
Metal ion (unspecified)	1
Thiamine-pyrophosphate	1

Physiological reactions of O₂ are radical reactions that require single electron donors to initiate the reaction with the O₂ diradical. This is reflected in the frequency of single electron donors (marker with an asterisk in Table 3) among the cofactors for O₂-dependent reactions sampled here. In some reactions, the substrate can provide the radical. In some reactions, the mechanism and the single electron source are unknown. Some of the cofactors are not directly involved in O₂ activation. The frequencies of the cofactors reflect their occurrence across individual reaction types, not their frequency within a protein or their occurrence in the environment. For example, a reaction catalyzed by a heme copper oxidase containing several hemes and several copper ions in all marine cyanobacteria (or in mammalian mitochondria, not sampled here) produces one count of heme and copper each. Similarly, the most abundant enzyme on Earth, RuBisCO, accounts for the lone occurrence of magnesium in the table.

The most common cofactors in O₂-dependent reactions are flavins and iron. Flavins are versatile coenzymes that perform both one- and two-electron transfers. They serve as cofactors and interact with dioxygen in several enzyme families, including flavin monooxygenases such as UbiH and UbiI involved in ubiquinone synthesis, PhzS from the phenazine pathway, or cyclohexanone monooxygenases ChnB. The most common activating mechanism of the flavin monooxygenases involves the reduced cofactor reacting with dioxygen to form the characteristic C(4a)-(hydro) peroxyflavin intermediate through a semiquinone radical pair [99,100].

Iron is the most common transition metal cofactor in oxygenases [101]. A common transition metal-dependent enzyme family in our set was the Rieske non-heme iron oxygenase family, including naphthalene-1,2-dioxygenase, the steroid hydroxylase KshAB involved in cholesterol catabolism, phthalate-4,5-dioxygenase and others. The oxygenase component of these enzymes is characterized by the presence of a catalytic mononuclear iron center and a Rieske iron–sulfur cluster. The latter accepts electrons from the two-electron donor NADH through a reductase component (often flavin-dependent), sometimes via an additional electron carrier. The electrons are transferred to the mononuclear iron center in order to activate dioxygen. The proposed mechanism includes rearrangements encompassing several oxidation states of the mononuclear iron center and a variety of radical and non-radical intermediates [102].

Flavins and iron are present individually, and sometimes together, in over 40% of the enzymes in our sample. NAD(P)H is very common as an electron donor (present in 23% of the reactions), providing electrons to oxygenases through flavins or the reductase component of a multi-component enzyme, as in the Rieske non-heme iron oxygenases. Iron–sulfur clusters are present as prosthetic groups in roughly 16% of the enzymes in our sample and about 17% bind the soluble one-electron carrier ferredoxin. Cytochromes as electron donors are rare for these prokaryotic enzymes. The only O₂-dependent S-adenosyl methionine (SAM) binding enzyme identified in our sample was MnmC, a bifunctional enzyme found primarily in γ-Proteobacteria that catalyzes a specific modification of the tRNA wobble base by a mechanism including methylation preceded by oxidative cleavage [103].

Mononuclear iron centers and FeS clusters are often inhibited by oxygen [9]. This inhibition is likely the reason for the emergence of alternative pathways following the GOE. Why, then, are many alternative oxygen-tolerant enzymes replete with Fe and FeS clusters? The reactivity of oxygen with low-potential metal centers that results in cofactor inactivation in ancient O₂-independent enzymes, whose active sites are not adapted to dealing with oxygen, was not only circumvented, but rather used to good advantage for oxygen activation in the new, alternative pathways: once activated, O₂ can initiate a large number of radical reactions.

The KEGG list of O₂-dependent reactions in prokaryotes includes 21 having no associated cofactors, or where the cofactors were unknown. Several cofactor-independent oxygenases have recently been described [104–108]. Their mechanisms involve a substrate anion, generally generated by base catalysis, that donates a single electron to dioxygen, yielding a stabilized radical pair. The most familiar example of an oxygenase that requires no cofactors to activate dioxygen is the Calvin cycle enzyme RuBisCO, whose oxygenase mechanism involves a substrate-derived radical [107].

No evidence for pre-GOE environmental O₂ in genomes

Evidence for the appearance of O₂ in the atmosphere by the GOE 2.4 billion years ago is universally accepted and uncontroversial [1,2,10,11]. Some studies even date the GOE to a slightly more recent time, 2.3 billion years ago [108]. Several recent reports argue that atmospheric O₂ is older than the GOE. The proposals have it that geochemical interactions between quartz and water generated H₂O₂ early in Earth history and this quartz-derived H₂O₂ could have given rise to O₂ independent of the cyanobacterial OEC [109–111]. From a biological standpoint, an origin of O₂ from H₂O₂ seems extremely unlikely for several reasons. With a standard potential E_o = +1.7 V, H₂O₂ is a stronger oxidant than Cl₂ (E_o = +1.36 V) or O₂ (E_o = +0.81 V), hence a powerful disinfectant and general-purpose biocide, excluding a physical proximity between (so far unobserved) geological H₂O₂ production and life.

At the heart of H₂O₂-dependent abiotic O₂ origin models [109–111] is the chemical reaction catalyzed by catalase,

$$2{\text{H}}_2{\text{O}}_2\to {\text{O}}_2+2{\text{H}}_2\text{O}$$

which is exothermic by ΔH = −101 kJ·mol⁻¹ of H₂O₂ at 100 °C [111] and exergonic with ΔG°′ = −189.9 kJ·mol⁻¹ (1 M H₂O₂, 25 °C, pH 7, 1 atm O₂) and ΔG_R = −154.2 kJ·mol⁻¹·mol⁻¹ (for 1 mM H₂O₂, 100 °C, pH 7 and 0.1 atm O₂), making it an irreversible reaction under physiological conditions. The same reaction can be catalyzed by bifunctional catalase-peroxidases (Table S6), enzymes that degrade H₂O₂ either as a catalase (reaction above) or alternatively as a peroxidase [112] according to the reaction.

$${\text{H}}_2{\text{O}}_2+2\text{H}-\text{Substrate}\to 2{\text{H}}_2\text{O}+2\text{Substrate}$$

using the same active site, with ΔG depending on the substrate. Catalase is the most common H₂O₂ detoxifying enzyme in genomes (Table S6). Proponents of a sand/quartz origin of O₂ might argue that the abundance of catalases in genomes (Table S6) provides evidence in favor of an H₂O₂ origin of pre-GOE O₂, with catalase representing a widespread and ancient relict of a time when O₂ was made from H₂O₂. But catalase activity is essentially ubiquitous among prokaryotes, even among strict anaerobes [113], which generally lack O₂-dependent enzymes and pathways that could utilize O₂ (Figs S4 and S5). The function of catalase in strict anaerobes is to detoxify reactive oxygen species generated by one electron carriers during redox metabolism or by other organisms in the environment [77,114], not to provide anaerobes with O₂ from environmentally-supplied H₂O₂.

Furthermore, the proposed quartz-H₂O₂ mechanism of O₂ origin would, in principle, operate before the origin of cells [111], in which case the genetic code, aminoacyl-tRNA synthetases, and ribosome biogenesis would have evolved in the presence of O₂ and H₂O₂. We found no O₂- or H₂O₂-dependent enzymes involved in core information processing functions. Thus, data from genomes and enzymes are distinctly at odds with suggestions for a pre-GOE (or prebiotic) origin of O₂ from H₂O₂ or quartz–water interactions.

Moreover, a recent incisive study by Koppenol and Sies [115] effectively rules out the possibility that abiotic H₂O₂ could have generated O₂ in amounts that would in any way impact the biosphere. They used simple and uncontroversial kinetics to calculate the expected half-life of H₂O₂ in the presence of 20–200 μM iron(II), a reasonable value for Archaean oceans. The key to their reasoning is the classical Fenton reaction, the rapid reaction of H₂O₂ with Fe²⁺ to yield Fe³⁺ and hydroxyl radicals, which rapidly react with almost any organic substrate. They obtain a conservative estimate for the half-life of H₂O₂ in the presence of Fe²⁺ as 0.7 s. Because Fe²⁺ was ubiquitous in all water-bearing environments on the early Earth, including the oceans, the miniscule 0.7 s half-life of H₂O₂ leaves no alternative to the conclusion that “before oxygenic photosynthesis, organisms were not exposed to H₂O₂, HO^·, ${\text{O}}_2^{·-}$, or ${\text{CO}}_3^{·-}$ ”[115].

Another line of evidence suggesting pre-GOE O₂ comes from the possible presence of small amounts (“whiffs”) of O₂ appearing in rocks 2.5 billion years of age, or ~ 50 MY prior to the GOE, based on the redox state of redox-sensitive minerals in sediments [116]. The evidence for pre-GOE whiffs of O₂ has recently been debated, however, with a new report indicating that the oxidation state of the molybdenum minerals originally supporting the existence of whiffs resulted from oxidation that took place after the GOE, long after the sediments had been deposited [117]. Those interpretations are challenged [118], and the challenge is in turn rebutted [119], suggesting anoxia (< 1 ppm O₂) in the sediment in question 2.5 billion years ago. From a biological perspective, the difference between oxygen appearance 2.4 billion years ago (the standard GOE model), or 50 million years earlier (the whiffs model), or 50 million years later [108] corresponds to a 2% difference in oxygen age, which has no impact on the verticality or functions (Fig. 3) of O₂-dependent enzymes surveyed here. It merely alters the time at which O₂ started to impact enzyme evolution by a factor of 0.02. In this study, we conservatively take the age of O₂ to correspond to the age of the GOE and we do not use molecular clocks.

Discussion

The main evolutionary impact of the origin of O₂ is traditionally viewed from the standpoint of eukaryote origin [49,84] or from the standpoint of mitochondrial respiration and animal evolution [50,120]. Yet eukaryotes appear roughly 1 billion years after the advent of O₂ [13,121], and animals appear in the fossil record near the base of the Cambrian, almost 2 billion years later than the appearance of O₂ [122,123]. Because both eukaryotes and animals emerged in a world that already contained biologically relevant amounts of O₂, their immediate ancestors were already equipped, enzymatically, to deal both with anoxia and with O₂ as a toxin and/or a substrate [124]. The fossilization of early invertebrates required the existence of O₂ for the synthesis of collagen by proline hydroxylases to generate hard body parts [125]. The origin of large animals also required the presence of O₂ [126]. Life on land was accompanied by adaptations to a permanently oxygenated atmosphere [127].

From the moment it was first produced by cyanobacteria, O₂ became an inhibitor of prokaryotic proteins with solvent-exposed FeS clusters, including ferredoxins, by acting as a one-electron acceptor in redox reactions [9,63]. The appearance of environmental O₂ following the GOE confronted prokaryotic life with a few challenges and numerous chemical opportunities. The main challenge was that a very small number of enzymes in anaerobes—sometimes only one enzyme per species, but often in physiologically key positions in metabolism—are inhibited by exposure to O₂ or activated forms thereof [6,9,63–73]. Such enzymes, for example pyruvate:ferredoxin oxidoreductase, a key enzyme in carbon and energy metabolism of anaerobes [65], or nitrogenase, the biosphere’s N₂-fixing enzyme [128], typically have either a radical mechanism [5] or harbor low-potential metal centers, very often FeS centers, that can spontaneously react with the O₂ diradical [71], or both. The O₂-dependent inactivation of essential enzymes can arrest growth until anoxia is restored, which allows the organism to replace the poisoned enzyme by repair or resynthesis [6].

The evolutionary significance of O₂ is traditionally viewed in terms of energetic efficiency, O₂ having enabled improved ATP yield from heterotrophic substrate breakdown. The underlying reasoning is often that “life with oxygen is better than life without: a given amount of glucose processed in the presence of oxygen produces 18 times as much energy as the same amount of glucose processed without oxygen” [50]. Escherichia coli gains 15 ATP per glucose during O₂ respiration under optimal conditions and 4 ATP per glucose from anaerobic fermentation [129], the difference in maximal ATP yield is roughly a factor of 3.8, not 18.

Terrestrial, multicellular eukaryotes are, of course, highly specialized to an O₂-containing atmosphere and possess an O₂-dependent metabolism, yet this specialization took place relatively late in evolution [13,124,127]. Our present investigation deals exclusively with prokaryotes, as they were the only organisms to directly witness the advent of O₂ in evolution. In prokaryotes, O₂ presence does not always mean more energy, and more energy is not always in line with the physiological needs of the cell. The membrane potential, ΔΨ, used by E. coli for substrate import and ATP synthesis under anaerobic growth, −130 mV, is roughly the same as that under aerobic growth, −140 mV [130]. Yeast exhibits a similar response to E. coli’s overflow metabolism, called the Crabtree effect, the preference for fermentation in the presence of glucose and oxygen, which results in an increased rate of ATP production at low efficiency [18,131], not in an increased efficiency of ATP production.

In organisms that respire O₂, the diversity of terminal respiratory oxidases exhibits a variety of bioenergetic capacity. Among the enzymes of the HCO superfamily [51,75,76], the aa3 type cytochrome c oxidases and the bo₃ type ubiquinol oxidases, both belonging to family A, pump 4 protons per O₂ reduced, while the cbb₃ type cytochrome c oxidases (family C) usually pump 2 protons per O₂ [35,75,132]. Among the growing number of oxidases belonging to the B family of HCO [33], there are subfamilies that pump 4 protons per O₂, various subfamilies that pump 2 protons per O₂ as the prototypic B-family oxidase of Thermus [132] and five subfamilies that do not pump protons, also because they do not react with oxygen for the lack of an active-site Tyr, which is involved in the formation of the His-Tyr cofactor in oxygen reductases. Conversely, the bd type ubiquinol oxidases and AOX pump 0 protons per O₂ [35,51], although the bd oxidases generate protonmotive force [34,35]. The sub-units of non-pumping bd type oxidases are among the most widespread O₂-dependent enzymes in the present genome sample, apparently more widespread than those of the terminal oxidases that pump 2 or 4 protons per O₂ reduced (Table 2) (see references [33,34] for recent analyses of the HCO and bd oxidase superfamilies).

As with the Crabtree effect (2 ATP per glucose via cytosolic glycolysis in the presence of O₂), this spectrum of energy conservation within O₂-dependent terminal oxidases from 0 to 4 protons pumped per O₂ reduced suggests that the immediate physiological role of O₂ in evolution was not improved energy yield. Despite the highly exergonic nature of reactions between O₂ and organic compounds, the absence of proton-pumping capabilities in many terminal oxidases [34] and the noteworthy absence of O₂-dependent substrate-level phosphorylation suggest that the microbes that learned to harness O₂ in biochemical evolution were not limited in energy efficiency, but were instead limited in their ability to grow in Earth’s increasingly oxic habitats because of enzymatic inhibition imposed by O₂.

Lateral gene transfer is the default mode of prokaryotic genome evolution [133]. It has impacted the distribution of O₂-dependent terminal oxidases across prokaryotic lineages [76,134]. Most genes in prokaryotic genomes have undergone LGT, with 97% of all genes having undergone at least one case of LGT between bacteria and archaea [135]. No gene family present in prokaryotic genomes has been completely immune to LGT between prokaryotic phyla during evolution [20]. We found that O₂-dependent enzymes underwent LGT more frequently than O₂-independent enzymes (Figs 2 and 3) and that this is true across almost all functional categories (Fig. 3 and Fig. S7, Table S12). The frequent spread of genes for O₂-dependent enzymes indicates that they conferred a physiological advantage to organisms that retained them. The nature of that advantage falls into three categories: (a) the breakdown of stable bonds in aromatic and nitrogenous compounds thereby mobilizing recalcitrant substrates in oxic environments, (b) the replacement (or supplementation) of enzymes that are inhibited by O₂, and (c) terminal oxidases (Table 2), but terminal oxidases can only be used by an organism once O₂-tolerant pathways of essential intermediate biosynthesis (or auxotrophy) have been incorporated into the genome (Fig. 4).

In primary metabolism, O₂-dependent enzymes have not generated fundamentally new biosynthetic or assimilatory traits. The central pathways of primary metabolism were in place and have remained largely unchanged since the time of the last universal common ancestor LUCA [136]. O₂-dependent enzymes presented alternative, O₂-tolerant routes to preexisting O₂-independent pathways that involved O₂-sensitive intermediates or O₂-sensitive cofactors, in particular solvent-exposed low-potential FeS clusters [64–71,73]. O₂-dependent enzymes are O₂-tolerant by nature, they allowed cells to colonize O₂-containing habitats from which they otherwise would have been excluded [73]. This factor underlies the higher rates of LGT we observe for O₂-dependent reactions. The colonization of oxic habitats not only required the presence of O₂ detoxification mechanisms [9,63], it required the presence of O₂-tolerant pathways for essential cofactor biosynthesis and substrate mobilization. O₂-tolerant enzymes that provide alternatives to the older O₂-sensitive counterparts in essential cofactor biosynthesis had to evolve and be expressible in the genome before the use of O₂ as a terminal acceptor in respiratory chains became an option for any microbe. Stated another way, O₂ is useless for a cell if the cell cannot grow in the presence of O₂ for lack of essential cofactors—NAD⁺ [91], PLP [92], thiamine [95,96], quinones [93,94], chlorophyll [73], cobalamin [88,89], or heme [88]—whose anaerobic synthesis pathways are inhibited by O₂.

This selective pressure for the origin of O₂-tolerant and O₂-dependent enzymes, first suggested for chlorophyll biosynthesis [73], appears to apply very generally to O₂-dependent enzymes. Are there glaring exceptions to the rule? There is one, a big one—nitrogenase. Nitrogenase is rapidly inactivated by O₂ through damage to its solvent-exposed FeS clusters [128,137]. Yet in 4 billion years, life has never brought forth an O₂-dependent or O₂-tolerant alternative to Mo, Fe or V-dependent forms of nitrogenase. As a consequence, nitrogenase remained inhibited by atmospheric O₂ at ca. 1% PAL throughout the boring Pasteurian billion [12]. The inhibition of FeS clusters in nitrogenase by O₂ is well studied [128,137]. Nitrogenase inhibition by O₂ is a special case among O₂-dependent enzymes: it is the most important O₂-dependent enzyme that never evolved. Nitrogenase inhibition might have been the limiting factor for the increase in oxygen levels throughout the boring billion, through a negative feedback loop [12]. The essence of that negative feedback loop is that prior to the origin of cellulose synthesis by land plants, O₂ production by cyanobacteria and algae was stoichiometrically linked to synthesis of cell mass with a roughly 5 : 1 C : N ratio. Inhibition of nitrogenase brought biomass synthesis (hence O₂ production) to an immediate halt until environmental O₂ levels once again fell below 1% and nitrogenase could resume activity [12]. Even though nitrogenase inhibition by O₂ acted locally on FeS clusters in a protein, it exhibited an impact as global as cyanobacterial O₂ production itself. Our survey of O₂-dependent enzymes uncovers surprising generality of O₂ inhibition [73] and identifies it as a key mechanism that drove metabolic evolution, enabling prokaryotes to colonize oxic habitats, governing the composition of Earth’s atmosphere [12] for almost 2 billion years.

Abbreviations

GOE

great oxidation event

LGT

lateral gene transfer

OEC

oxygen-evolving complex

PAL

present atmospheric level

PLP

pyridoxal phosphate

ROS

reactive oxygen species

SLP

substrate-level phosphorylation

SOD

superoxide dismutase

V

verticality.

Data accessibility

All data are available in the main text or the Supporting Information. The protein families used for this analysis are available under https://www.molevol.hhu.de/resources.