Introduction

Because oxygen is a fundamental requirement for animal life and because oxygen levels in the ocean-atmosphere system have clearly changed over Earth history, the availability of oxygen has come to be regarded by many geologists and paleontologists as a master variable governing animal diversity through time. Understanding the relationship between oxygen and animal evolution is critical to understanding the rise and trajectory of complex life on Earth, and it also bears important lessons for modern and future global change. The fossil record offers valuable opportunities to investigate biotic responses to changes in Earth’s climate because these responses are not skewed by human interference (i.e., overexploitation or habitat loss), span a greater temporal scale than can be studied in laboratory experiments, and represent our only source of data on actual outcomes of changes in climate-related stressors (Harnik et al., 2012; Calosi et al., 2019; Clapham, 2019). Most critically, the fossil record is our only source of data for understanding whether most species’ extinctions are driven by physiological or ecological processes. Although physiological stressors (e.g., temperature-dependent hypoxia, ocean acidification) are the pertinent environmental factors for marine organisms, it is an open question whether most species-level extinction has occurred or will occur as a direct result of these stressors or via physiologically driven stress on keystone species, followed by trophic cascades and ecologically driven extinction for other species. The main challenge in learning and applying these lessons is moving beyond parallel and correlative studies of the fossil record and the geochemical record of ancient oceans. This is ultimately necessary if we are to ask questions such as how extinction selectivity during ancient hypoxic events corresponds to phylogeny, biogeography, or functional morphology, or whether extinction is likely to be mediated directly by physiology or ecological cascades. The answers to such questions will sharpen our ability to use the geologic record to predict biological responses in the future warming and deoxygenating ocean.

In this article, we first aim to provide an accessible introduction to how geochemists track redox changes in ancient oceans (which ultimately leads to inferences of global oceanic and atmospheric oxygen). Next, we review current knowledge of Earth’s oxygen history, moving from the anoxic oceans and atmospheres of the Archean eon (4.0–2.5 billion years ago) to the modern well-oxygenated state, where deoxygenation (and global warming) is largely driven by massive carbon inputs from large igneous province (LIP) super-eruptions (Fig. 1). We then discuss how animals evolved and responded to oxygen changes throughout this history, particularly aiming to use ecological space-for-time substitutions and modern ecophysiological approaches to link environmental change with biotic responses. We argue that such approaches have high promise for attaining more mechanistic understanding of how oxygen changes affected animals across timescales. Considering the past ∼400 million years of Earth history, we emphasize the role of LIPs in driving low-oxygen events in the ocean. We specifically dissect the evidence for environmental change resulting from the eruption of the Siberian Traps LIP, which caused the most devastating mass extinction in Earth’s history at the Permian-Triassic boundary. Finally, we discuss the lessons for modern anthropogenic global change that can be drawn from ancient low-oxygen oceans, as well as highlight future research directions that will help build on our current knowledge.

Reading the Record of Oxygen in Earth’s Past

When discussing atmospheric oxygen in Earth’s past, geologists and modelers commonly describe these levels by using percentage of present atmospheric level (% PAL; present = 100%). For marine redox state or dissolved oxygen levels in ancient oceans, geologists commonly use a descriptive scheme that differs slightly from that currently in use in modern global change studies. First, anoxic conditions are recognized to be either “euxinic,” with free sulfide, or “ferruginous,” with free ferrous iron, at least enough to titrate out available sulfide (Poulton and Canfield, 2011). In contrast to modern anoxic settings, which are almost all euxinic, it is now recognized that, especially in the Proterozoic and early Paleozoic (∼2000 to 400 million years ago [Ma]), most anoxic settings were ferruginous (Canfield et al., 2008; Planavsky et al., 2011; Guilbaud et al., 2015; Sperling et al., 2015b, 2021). Similar to modern global change biologists, geologists use the term “suboxic” to refer to the feather edge of anoxia, where very few animals aside from chemosynthetic organisms can live. Here, we define suboxic as ∼1.5% air saturation, ∼5 μmol L⁻¹, ∼0.1 mL L⁻¹ O₂ (note that oxygen concentrations expressed in μmol L⁻¹ or mL L⁻¹ will be dependent on temperature, pressure, and salinity; Hofmann et al., 2011). In contrast to global change biologists, geologists and paleontologists commonly use the term “dysoxic” instead of “hypoxic” to define oxygen levels between suboxic and fully oxic. In our opinion, using the term hypoxic, which is a species- or organism-specific physiological state, to define the geochemical characteristics of a water mass is incongruent. Further, a main point of this paper is understanding the combined effects of changing oxygen and temperature. Because hypoxia is the synergistic interaction of these two environmental parameters (e.g., Deutsch et al., 2015), even for a single organism a water parcel can be hypoxic (or not) across a range of oxygen levels, depending on the temperature. Thus, we will use dysoxic in this paper to describe low-oxygen conditions and define it as O₂ levels between suboxia (less than ∼1.5% air saturation) and normoxia (greater than ∼31% air saturation, >88 μmol L⁻¹, >2 mL L⁻¹ O₂). Sperling et al. (2015a) and Cole et al. (2020) have recently reviewed oxygen nomenclature across and within fields in geology and modern biology and oceanography.

For studies of the Holocene (the past 12,000 years) or Pleistocene (the past 2.6 million years), geologists are able to utilize small bubbles of ancient air trapped in ice cores. This has allowed for direct measurement of atmospheric chemistry, most notably the dramatic changes in CO₂ related to orbitally forced climate cycles (e.g., Petit et al., 1999) but also atmospheric O₂ (Yan et al., 2021). For deeper Earth history, however, atmospheric oxygen or marine redox changes—with a few exceptions detailed below—must be reconstructed through indirect proxy methods derived from sedimentary archives. Our two major archives for understanding paleoenvironmental conditions on ancient Earth are carbonates (composed of sedimentary minerals, such as calcite, aragonite, and dolomite, that are precipitated from seawater as both inorganic precipitates, such as ooids, and the organic biominerals of organisms, such as corals or bivalves) and shales or mudstones (composed of fine-grained particles that represent the transported remnants of previously weathered rocks). Carbonate proxies tend to rely on the abundance or isotopic ratios of elements that are incorporated as trace constituents into the carbonate crystal lattice, whereas shale proxies generally rely on elements that associate with organic matter or pyrite under reducing conditions. Each of these sedimentary archives can record local proxies, which reveal information about the bottom waters directly above a given sedimentary locality but with no global context, and global proxies, which reveal the mean state in the ocean but with no spatial or bathymetric context (for a comprehensive review of various proxy approaches, see Kendall, 2021). Local proxies for shales include iron speciation (Poulton, 2021) and the enrichment of redox-sensitive trace metals such as molybdenum and uranium (Tribovillard et al., 2012), whereas carbonates can be studied using iodine to calcium (I/Ca) ratios (Lu et al., 2018) or cerium anomalies (Bellefroid et al., 2018; Liu et al., 2021). Critically, I/Ca may be able to fingerprint dysoxic conditions (Lu et al., 2016), in contrast to most local proxies in shales, which can robustly distinguish oxic versus euxinic conditions but may have difficulty distinguishing fully oxygenated from dysoxic or ferruginous conditions (discussed by Sperling et al., 2018). The global marine redox landscape can be studied in shales by using molybdenum and uranium isotopes (Kendall et al., 2017; Stockey et al., 2020), which track variations in the extent of anoxic seafloor, or thallium isotopes (Them et al., 2018), which track variations in the extent of oxic seafloor.

In carbonates, the global redox landscape is increasingly studied using uranium isotopes (Fig. 2), which we will focus on here, particularly as they pertain to mass extinction events. Under anoxic conditions in the oceans, the reduction of soluble U(VI) to insoluble U(IV) at the sediment-water interface is associated with an isotopic fractionation because ²³⁸U is reduced more rapidly than ²³⁵U (due to the nuclear volume effect). Although there can be local differences, the average isotopic offset between seawater and anoxic sediments is about +0.6‰. Thus, expansion of ocean anoxia is expected to cause not only a reduction in ocean uranium concentrations but also a reduction in the oceanic ²³⁸U/²³⁵U ratio (typically reported in delta notation as δ²³⁸U; reviewed in Zhang et al., 2020). The long residence time of uranium in seawater (∼500 thousand years [kyr]) relative to the ocean mixing time (∼1–2 kyr) leads to the oceans being well mixed in terms of the ²³⁸U/²³⁵U ratio, meaning this ratio is the same anywhere in the ocean. Because carbonates trap minute quantities of uranium in their crystal lattice during precipitation with minimal or at least understandable offset from seawater (Chen et al., 2018), and because the ocean is well mixed with respect to uranium, this means that any carbonate deposited in an ancient ocean should be recording the U isotope composition of its seawater. And since the isotopic composition of seawater is primarily set by the ratio between seafloor covered by oxic versus anoxic bottom waters, geochemists can then quantitatively back-calculate (using increasingly sophisticated modeling techniques) the global redox landscape.

Atmospheric and Marine Oxygen Levels through Earth History

At the broadest level, Earth appears to have experienced four stages of atmospheric oxygen levels. There were likely important oscillations within these stages, especially during the middle part of Earth’s history; but from the perspective of changes that were relevant to animal life in the oceans, these stages are the early part of Earth history in the Archean and early Paleoproterozoic (∼4000–2200 Ma; when the oceans and atmosphere were anoxic), the mid-Proterozoic (∼2200–540 Ma; when oxygen was stably present but at relatively low levels, perhaps 1%–10% of present levels), the early Paleozoic (∼540–420 Ma; when oxygen levels were perhaps 10%–40% of present levels), and the later Paleozoic to present (∼420 Ma to present; when atmospheric oxygen was similar to present) (Figs. 1, 3). Despite the considerable study into the question of atmospheric oxygen through time and general agreement on the broad history, geologists and geochemists can recognize only two absolute values with clarity and a lack of controversy. Photochemical reactions in the atmosphere between incoming UV and SO₂ generate sulfur compounds wherein the sulfur isotopes are fractionated in a manner independent of their mass (mass-independent fractionation [MIF-S]). Such reactions occur today; but in the presence of even minute amounts of atmospheric oxygen, as low as 0.001% PAL (10⁻⁵ PAL), the products of these atmospheric reactions are oxidized and pass through the oceanic sulfur cycle, and any mass-independent fractionation effects are erased (Pavlov and Kasting, 2002). The MIF-S signals are recorded in the Archean and early Paleoproterozoic sulfur record (first discovered by Farquhar et al., 2000), providing an unambiguous signal of an essentially anoxic atmosphere on early Earth. The MIF-S then disappears, marking the Great Oxidation Event, or GOE. Current controversies include the timing of oxygenation (and whether it was protracted or sudden; Poulton et al., 2021; Hodgskiss and Sperling, 2022), but MIF-S and other markers of an anoxic atmosphere never return after ∼2.2 billion years ago, marking the irreversible oxygenation of Earth’s atmosphere.

The second unambiguous constraint on atmospheric oxygen concentration comes from the fossil record of charcoal (burned plant material). A series of elegant calculations and experiments have demonstrated that plant material will not ignite without ∼15% O₂ in the atmosphere (by volume, equivalent to 73% PAL, given 21% O₂ by volume in the atmosphere today), and globally distributed runaway wildfires will occur at greater than ∼35% O₂ (=170% PAL) (Jones and Chaloner, 1991; Wildman et al., 2004; Belcher and McElwain, 2008). Note, however, that this upper bound has not been experimentally determined with certainty and that many commonly cited secondary sources in the literature (e.g., Kump, 2008) cite earlier studies on paper ignition (Watson et al., 1978) that are not realistic for natural fuels (Wildman et al., 2004). Nonetheless, the presence of both charcoal and unignited plant fossils in the rock record since the late Silurian (∼420 million years ago) has allowed geologists to robustly infer that atmospheric oxygen has been between these approximate levels since this time (Scott and Glasspool, 2006). Thus, from the middle Paleozoic onward, surface oceans would have been in equilibrium with an atmosphere with near-modern oxygen. The deep ocean would have been fully ventilated (i.e., Fig. 3C), excepting any vagaries of polar deep-water formation as related to plate tectonics and climate, which do not show any directional trends over this time. Any changes in the strength of the biological pump through time could also have impacted the distribution of oxygen in the ocean’s interior (Meyer et al., 2016), but the degree to which such changes occurred is debated (Fakhraee et al., 2020). Importantly, for the past 420 million years, anoxic or dysoxic settings in the oceans would have had the same general oceanographic drivers as in the modern ocean (Fig. 3C, F). Specifically, deoxygenation would have developed beneath areas of higher primary productivity (in oxygen minimum zones [OMZs] and during times with higher nutrient inputs to the oceans) in enclosed seaways with limited deep-water renewal and, more broadly, in the open ocean because of stratification (Breitburg et al., 2018; Levin, 2018).

Between the initial onset of oxygen in Earth’s atmosphere at the GOE (∼2200 million years ago) and the unquestioned onset of near-modern atmospheric oxygen levels (∼420 million years ago), the situation is murkier. Here, geochemists must use observations of marine redox conditions and try to back-calculate atmospheric oxygen. Oxygen levels during the mid-Proterozoic, or Earth’s Middle Age, between ∼1800 and 800 million years ago, have generally been hypothesized to be between 1% and 10% PAL (Kump, 2008; Lyons et al., 2014). However, this oxygen estimate represents to some extent the gut feeling of Earth historians. This is derived from observations that many shallow mid-Proterozoic waters remained oxic (e.g., Shen et al., 2003; Hardisty et al., 2014) and from modeling results demonstrating that the invariably anoxic conditions in the deeper oceans would not have been present at higher than ∼40% PAL (Canfield, 1998; Ozaki and Tajika, 2013; Cole et al., 2022). The tricky question remains, though: exactly how low was atmospheric oxygen in Earth’s Middle Age? Certainly, recent papers with large numbers of geochemical observations from I/Ca (Lu et al., 2018), uranium isotopes (Gilleaudeau et al., 2019), and iron speciation (Planavsky et al., 2011; Sperling et al., 2015b) have supported the view of a very weakly oxygenated mid-Proterozoic Earth system. But converting such observations into hard numbers is difficult. For instance, Bellefroid et al. (2018) modeled cerium anomaly data to indicate a vanishingly low <0.1% PAL O₂, whereas modeling of similar data by Liu et al. (2021) resulted in estimates of 1%–2% PAL. While both of these estimates unambiguously point to lower levels than likely existed in the later Neoproterozoic or Paleozoic, as discussed below (see Oxygen and the origin of animals), this order-of-magnitude difference has important implications for the role of oxygen in the origin of animals. Thus, the current aim is distinguishing whether mid-Proterozoic oxygen levels were simply low or rather exceedingly low. There has also been increasing emphasis on treating this billion years of Earth history with more temporal nuance (Canfield, 2014; Diamond and Lyons, 2018; Planavsky et al., 2018).

The second canonical step in Earth’s oxygenation history was generally believed to have occurred in the late Neoproterozoic, but whether this resulted in full oxygenation of the ocean-atmosphere system to modern levels has recently been thrown into question. Although there is clear geochemical proxy evidence for oxygenation in the Neoproterozoic, there is also proxy evidence for deoxygenation, and much of the prior evidence for a state change at this time is less clear (discussed by Cole et al., 2020). Most likely, oxygen levels did rise around the base of the Cambrian, and the oceans were perhaps temporarily well oxygenated with respect to the remainder of the early Paleozoic (e.g., Chen et al., 2015; Wei et al., 2018). However, persistent and widespread anoxic conditions in the deeper waters of early Paleozoic oceans (Berry and Wilde, 1978; Lu et al., 2018; Dahl et al., 2019; Gill et al., 2021; Sperling et al., 2021) suggest that the broad constraint provided by modeling for atmospheric O₂ (e.g., <40% PAL) continues to apply to this interval. This inference is consistent with the latest generations of carbon cycle models (Krause et al., 2018; Lenton et al., 2018). The presence of Cambrian animals themselves does provide a new minimum estimate for Paleozoic oxygen levels, which was estimated by Sperling et al. (2015b) as ∼10% PAL. While this does provide an increase in minimum oxygen levels (Fig. 1), because late Neoproterozoic levels are only loosely constrained, the magnitude of any late Neoproterozoic-early Cambrian oxygen increase (and whether it represented a step change) is unclear. Full oxygenation of the ocean-atmosphere system to 100% PAL likely occurred later in the Paleozoic (Dahl et al., 2010; Wallace et al., 2017; reviewed by Reinhard and Planavsky, 2022). Current research is focused on the magnitude of Paleozoic oxygenation (and deoxygenation) events and possible links between plant evolution and later Paleozoic oxygenation (Dahl and Arens, 2020).

Large Igneous Provinces and Their Effects

After the atmosphere reached 100% PAL, most deoxygenation events developed as a result of massive inputs of carbon to the Earth system as supplied by LIP eruptions. The LIPs and their extrusive flood basalts are voluminous outpourings of basaltic magma that occur over a relatively short time period (geologically speaking, <5 million years [Myr] and generally <1 Myr) (Kasbohm et al., 2021). They occur in the middle of both oceanic and continental tectonic plates and are likely sourced from mantle plumes (making them fundamentally different in style of volcanism and size from any volcanic process currently operating on Earth). The flood basalt deposits from LIPs are often called trap rocks (the Siberian Traps, Deccan Traps, etc.) after the Swedish word trappa (“staircase”). This is due to the characteristic stairstep morphology of the successive basaltic lava flows (Fig. 4B). In addition to the aboveground flood basalts, LIPs have a complex underground plumbing system of dikes and sills that can often intrude into and volatilize organic- or sulfur-rich sedimentary rocks, thereby dramatically increasing the CO₂ or sulfur aerosols that are emitted to the atmosphere (Ganino and Arndt, 2009; Svensen et al., 2009).

High-precision geochronology has demonstrated that LIPs are precisely associated with multiple mass extinctions, including the three most recent Big Five extinctions (the five largest mass extinctions of the past 540 million years). The LIPs with precise links to extinctions include the second pulse of the North Atlantic Igneous Province (LIP)/Paleocene-Eocene Thermal Maximum extinction (PETM) (55.8 Ma), the Decan Traps/Cretaceous-Paleogene extinction (66.0 Ma), the Karoo-Ferrar/Pliensbachian-Toarcian boundary (182.7 Ma), the Central Atlantic Magmatic Province/Triassic-Jurassic boundary (201.5 Ma), and the Siberian Traps/Permian-Triassic boundary (251.9 Ma) (Ernst et al., 2020; Kasbohm et al., 2021). In these best-dated instances, the LIP erupted several hundred thousand years prior to the extinction (though note that both LIPs and extinctions may have multiple pulses). There are also LIPs likely associated with the Cretaceous ocean anoxic events at the Cenomanian-Turonian boundary (OAE-2) (93.9 Ma) and in the early Aptian (OAE-1a) (∼121 Ma), the end-Guadalupian extinction (259.1 Ma), the Frasnian-Fammenian mass extinction (372.2 Ma), and the early-middle Cambrian extinction (∼509 Ma), although the radiometric constraints are not as precise (Ernst et al., 2020; Kasbohm et al., 2021). Still other mass extinctions may be associated with LIPs, but this is based on indirect geochemical proxy evidence rather than a directly dated LIP (e.g., for the Late Ordovician extinction, Jones et al., 2017; Hu et al., 2020). Known LIP-extinction couplets are associated with geochemical evidence for both warming and deoxygenation (Fig. 4A; Ernst and Youbi, 2017). Reviewing the geochemical evidence for deoxygenation associated with LIPs/mass extinctions is beyond the scope of this article, but to focus on our highlighted proxy of uranium isotopes, a negative δ²³⁸U shift—indicating widespread marine anoxia—has been documented in association with the Frasnian-Fammenian extinction (Song et al., 2017; White et al., 2018), Permian-Triassic extinction (Brennecka et al., 2011; Lau et al., 2016; Zhang et al., 2018), Triassic-Jurassic extinction (Jost et al., 2017), Cenomanian-Turonian boundary (Clarkson et al., 2018), and PETM (Clarkson et al., 2021). All other LIP-associated extinctions show evidence for widespread anoxia via other proxy methods (e.g., thallium isotopes at the Pliensbachian-Toarcian boundary) (Them et al., 2018). Current research focuses on understanding the nexus of eruption magnitude, rate, and aerosol release, combined with Earth system boundary state, that causes some LIPs to be deadlier than others (e.g., Fig. 5B).

The Fossil Record

To understand the role of oxygen in animal evolution and biodiversity dynamics, it is necessary to understand how paleontologists read the fossil record and the nature of that record. For almost the entirety of Earth history (∼85%), the fossil record was dominated by microbes, and even those fossils were relatively rare. Consequently, quantitative studies of Precambrian diversity have been difficult (discussed by Cohen and Macdonald, 2015). However, increased scrutiny of older rocks has revealed that the first total-group eukaryotes appeared around 1650 Ma (represented by larger cells with cell wall protrusions and possible nuclei) and the first unambiguous crown-group eukaryotes (the red algae Bangiomorpha) around 1050 Ma (reviewed by Agić, 2021). Molecular clocks suggest that the animal crown group originated around 800 Ma (Dohrmann and Wörheide, 2017), although undoubted animal fossils are not present until the end of the Neoproterozoic, in the so-called Ediacaran fauna. The ∼50 million years spanning the Precambrian-Phanerozoic boundary witnessed the appearance of essentially all fossilizable animal phyla and classes (Erwin et al., 2011; the Cambrian explosion; Fig. 1). The explosive aspect of this event is with respect to the appearance of high-level animal body plans (disparity) rather than species diversity and with respect to the three billion years of preceding geological record with no animal fossils. Focusing on the ∼50 million years from 570 to 520 Ma, however, it rather reflects a protracted series of evolutionary innovations and radiations (Wood et al., 2019).

Following the Cambrian radiation, paleontologists have been able to better tabulate and statistically analyze the diversity of life on Earth. The most recognizable and influential Phanerozoic diversity curve of fossil animals is the Sepkoski curve (Sepkoski, 1981). This curve is based on a compendium that consists of first and last known fossil occurrences with a temporal resolution down to the geologic stage (the smallest geologic time subdivision) of marine animal orders, families, and genera from the published literature. Sepkoski demonstrated that there has been an increase in diversity across the Phanerozoic that follows a logistic curve (Fig. 5A). The general shape of the curve starts with a sharp increase during the Cambrian (Cambrian explosion), followed by a short plateau that is then followed by a larger increase during the Ordovician (Great Ordovician Biodiversification Event) before plateauing during much of the Paleozoic. This plateau is ultimately followed by an unabated increase to the Modern, starting during the early Mesozoic (Sepkoski, 1981). Recently, the compilation of fossil occurrences in the Paleobiology Database (2022) has enabled quantitative paleobiologists to address sampling biases in reconstructions of biodiversity, extinction, and origination through the Phanerozoic (Fig. 5; Alroy et al., 2008; Alroy, 2010; Kocsis et al., 2019).

Perhaps the most influential outcome of Sepkoski’s compendium is the first quantitative recognition of major mass extinction events. Extinctions in the fossil record were previously known, but Sepkoski’s compendium made it possible to estimate the magnitude and temporal resolution of these events (Raup and Sepkoski, 1982). Raup and Sepkoski (1982) revealed the occurrence of five major mass extinction events (colloquially named the Big Five) as having occurred during the Late Ordovician (∼445 Ma), Late Devonian (∼370 Ma), end-Permian (∼252 Ma), end-Triassic (∼201 Ma), and end-Cretaceous (∼66 Ma). The recognition of these extinction events spurred a flurry of new research agendas dedicated to uncovering the causal mechanisms. While there are open questions associated with all mass extinction events, the general consensus is that, for extinctions after atmospheric oxygen levels reached ∼100% PAL, the most common ultimate cause is the activity of LIPs, and the most likely proximate cause is the downstream environmental impacts including warming and deoxygenation (Bond and Grasby, 2017; Clapham and Renne, 2019; Fig. 4). The primary exception is the end-Cretaceous extinction event, which is primarily attributed to a bolide impact (Alvarez et al., 1980; Hull et al., 2020), although the temporal coincidence with the Deccan Traps LIP complicates causal scenarios (Schoene et al., 2015).

In addition to identifying five major mass extinction events, Raup and Sepkoski (1982) also noted that background extinction has gradually decreased across the Phanerozoic. Specifically, the total extinction rate (family extinctions per million years) was greatest during the Cambrian but gradually decreased over the next 500 million years. Subsequent studies that better account for possible sampling biases have also observed this decrease in extinction rate (e.g., Foote, 2003; Alroy, 2008; Kocsis et al., 2019). Various causal mechanisms have been proposed to explain the decrease in extinction rates, including increased ecological complexity, ecological resistance to extinction, taxonomic structure (increase in the number of species per clade over time), lineage durations, variability of the rock record, and stabilization of the carbon cycle as a result of decreased shallow-marine anoxia (reviewed by Payne et al., 2020). These observations remain largely correlative, and no community-wide consensus has yet been established.

Connecting Geochemistry and Fossils through Physiology

Traditionally, geochemical data have been compared stratigraphically with paleontological diversity data to assess the environmental impact of a given oxygenation or deoxygenation event. Here, we attempt to interpret these coupled records in light of information from modern physiology and ecology (Knoll, 2013). In particular, our interpretation of possible links between oxygen and the development of animal multicellularity are based on observations from respiration experiments on body plan analogs of the earliest animals, whereas our interpretation of the Cambrian radiation is based on space-for-time substitutions (e.g., Blois et al., 2013) from modern OMZs. Moving into the Phanerozoic, we focus on the metabolic index (Deutsch et al., 2015) as an ecophysiological link between environment and organismal evolution. The metabolic index (Φ) incorporates both environmental parameters (the partial pressure of oxygen [pO₂] and temperature [T]) and species-specific physiological parameters (A_o, E_o, and n) to predict the balance between organismal oxygen supply and demand (Equation 1), which must be greater than 1 for habitat to be aerobically viable. The parameter A_o represents an organism’s hypoxia tolerance (at a reference temperature in some formulations), E_o represents how hypoxia tolerance changes with temperature, and n represents the allometric scaling of an organism’s oxygen supply and demand balance with its body size (B). These physiological parameters are calibrated using respirometry experiments in which the lowest pO₂ that can sustain an organism’s standard metabolic rate, or P_crit, is measured across a range of temperatures. Note that a number of minor variations of the metabolic index equation have been applied to different ecophysiological problems since its development by Deutsch et al. (2015), but all are solving the same problem of viable aerobic habitat:

Simply put, Φ is the ratio of environmental oxygen supply to metabolic oxygen demand of a species, given the combination of local temperature and oxygen conditions. At Φ = 1, there is exactly enough oxygen in a given area for the species to maintain standard metabolic rate (SMR). At Φ > 1, the species has that many more times the environmental oxygen available to it than it needs to maintain SMR. Conversely, at Φ < 1, the species does not have enough environmental oxygen supply to meet its standard metabolic oxygen demand; thus, that given area would be aerobically unsustainable. Marine organisms do not, however, occupy all areas of the ocean where Φ > 1; this is because organisms are performing physiological functions, such as growth, reproduction, or feeding, that require a greater amount of oxygen than what is needed to maintain SMR. The term Φ_crit is the lowest Φ value where a given species is documented to exist and is considered the minimum Φ value required for the persistence of that species in a particular area (Deutsch et al., 2015, 2020).

Oxygen and Animals in the Precambrian and Paleozoic

Oxygen and early Paleozoic extinctions

Although the dawn of the Paleozoic witnessed these two major evolutionary radiations, the early Paleozoic was also characterized by anomalously high extinction rates relative to the remainder of the Phanerozoic. A new ecophysiological explanation for high early Paleozoic extinction rates was recently proposed that requires only commonplace climatic perturbations in the context of lower atmospheric oxygen levels (Stockey et al., 2021; Figs. 5, 6). Our understanding of the synergistic effects of oxygen and temperature predicts that marine ectotherms are likely to have smaller thermal safety margins at lower environmental oxygen levels (Pörtner, 2010; Deutsch et al., 2015). Marine animals are therefore expected to be more vulnerable during ocean warming events in low-oxygen Earth system states. Stockey et al. (2021) generated 3D ocean simulations using the cGENIE Earth system model (Ridgwell et al., 2007) at different atmospheric O₂ levels, continental configurations, and climate states. These ocean models were then combined with the metabolic index framework (Deutsch et al., 2015; Penn et al., 2018; Equation 1) to estimate the magnitude of warming-driven extinction for a given climate perturbation under different levels of atmospheric oxygenation. Specifically, the modeled ancient oceans were populated with ecophysiotypes having physiological parameters drawn from the distributions measured in modern species (namely, A_o, E_o, and Φ_crit). The ocean was then perturbed with a CO₂-driven warming event, and the percentage of ecophysiotypes that could find viable aerobic habitat before the warming event was compared to the post-warming ocean. The modeled marine extinction events resulting from standard climate variability under early Paleozoic levels of oxygenation were much more severe than those predicted for the same climate perturbation under modern atmospheric oxygen levels (Fig. 5). Notably, these analyses illustrate an inflection point in simulated extinction magnitude around the upper limit for reconstructed early Paleozoic oxygenation. This modeling approach further demonstrates that atmospheric oxygen is a more important factor in controlling the extinction vulnerability of marine ectotherms than other Earth system boundary conditions such as continental configuration, pre-warming climate state, and the strength of the biological pump.

Anatomy of a Mass Extinction: Permian-Triassic Mass Extinction

The end-Permian mass extinction was the most severe biotic crisis in the history of animal life, whether measured in terms of the proportion of taxa lost (Raup and Sepkoski, 1982; Benton, 1995) or the impact on ecosystem structure (Droser et al., 2000). Many causes have been proposed for this global catastrophe (reviewed in Erwin, 1993; Payne and Clapham, 2012), but emerging data from sedimentary geology and geochemistry and new insights from animal physiology demonstrate that stresses impacting respiratory physiology can account for both taxonomic and biogeographic selectivity patterns in the extinction (Knoll et al., 1996, 2007; Clapham and Payne, 2011; Kiessling and Simpson, 2011; Penn et al., 2018; Song et al., 2020; Foster et al., 2022).

Most researchers now agree that the eruption of the Siberian Traps, the largest LIP of the past 500 million years, was the geological event that triggered the environmental changes responsible for the biodiversity crisis (e.g., Renne and Basu, 1991; Campbell et al., 1992; Reichow et al., 2009; Svensen et al., 2009; Burgess and Bowring, 2015; Burgess et al., 2017; Fig. 7). The global environmental impact of the Siberian Traps eruptions resulted from the release of volcanic gases, dominated by carbon dioxide but also including sulfur volatiles and halocarbons (Svensen et al., 2009; Sobolev et al., 2011; Black et al., 2012; Broadley et al., 2018; Elkins-Tanton et al., 2020). The eruptions of this specific LIP likely had an outsized environmental impact because the gas content of the large volume of basaltic magma was enhanced via heating and degassing of volatile-rich sedimentary rocks (i.e., coal) in the Tunguskaya Basin (Ganino and Arndt, 2009; Svensen et al., 2009; Ogden and Sleep, 2012; Elkins-Tanton et al., 2020). The discovery of coal fly ash at the Permian-Triassic boundary in sites far distant from the Siberian Traps provides further direct evidence for massive coal combustion and concomitant CO₂ release (Grasby et al., 2011).

This massive injection of CO₂ into the Earth system caused substantial greenhouse warming and marine deoxygenation (Fig. 7). Seawater temperatures in the surface of the tropical Tethys Ocean increased ∼12 °C within ∼800 kyr, from ∼25 °C to ∼37 °C, with potentially 7 °C to 10 °C of warming occurring in ∼39 kyr (Sun et al., 2012). Accurate temperature estimates are unfortunately not available from most areas worldwide across the Permian-Triassic, although other temperature proxy records do concur there was substantial warming (e.g., Joachimski et al., 2012, 2019). It is essential to consider both magnitude and timescale of warming when comparing the end-Permian crisis with modern global change, because the late Permian warming was substantially larger but occurred over a much longer duration. With respect to deoxygenation, the series of oceanographic linkages between CO₂ injection and deoxygenation remain controversial. For instance, Penn et al. (2018), using the community earth system model (CESM), suggested that warming drove persistent stratification of the ocean and, thereby, anoxia, whereas Hülse et al. (2021), using the cGENIE Earth system model (Ridgwell et al., 2007), suggested that widespread anoxia was largely the result of temperature-driven microbial respiration. Despite these uncertainties regarding the oceanographic mechanism of deoxygenation, it is now clear from geologic and geochemical evidence that ocean deoxygenation coincided with the end-Permian mass extinction and persisted for several million years into the Triassic.

Evidence of end-Permian ocean deoxygenation emerged first from analyses of sedimentation patterns, as many shallow-marine records exhibit shifts from light-colored, bioturbated sediments below the extinction horizon to dark-colored, unbioturbated sediments above the extinction horizon (Wignall and Hallam, 1992; Wignall and Twitchett, 1996). Similar sedimentological shifts occur in deep-marine stratigraphic sections as well (Isozaki, 1997). Iron speciation and pyrite framboid sizes demonstrate euxinic conditions developing in places at mid-shelf or even inner shelf depths during the end-Permian extinction, with ferruginous conditions present at deeper depths (Bond and Wignall, 2010; Clarkson et al., 2016; Schobben et al., 2020). Finally, in addition to these inorganic proxies, fossil molecules (biomarkers) derived from the pigments of green sulfur bacteria peak in abundance near the Permian-Triassic boundary (Grice et al., 2005; Cao et al., 2009). These bacteria photosynthesize using hydrogen sulfide as an electron donor, thus requiring anoxic, sulfide-bearing water within the marine photic zone. The occurrence of these fossil molecules at widely distributed sites indicates an unusual prevalence of shallow-marine euxinia associated with the mass extinction event.

Over the past decade, inferences from sedimentary patterns and local redox proxies have been strengthened via geochemical analysis of global redox proxies. Specifically, studies demonstrate that the Middle and Late Permian were characterized by δ²³⁸U similar to the modern ocean, followed by a negative uranium isotope shift across the Permian-Triassic boundary in shallow-marine carbonate rocks. This uranium isotope excursion was first identified across the immediate Permian-Triassic boundary (Brennecka et al., 2011), with later studies demonstrating that light δ²³⁸U values characterized the entirety of the Lower Triassic and could be found in numerous locations worldwide, including south China, Iran, and Turkey (Lau et al., 2016; Elrick et al., 2017; Zhang et al., 2018). Numerical models of the uranium cycle accounting for uncertainty in model parameters indicate that the shift in δ²³⁸U requires a change in the areal extent of seafloor anoxia from <1% to >10% in less—and perhaps much less—than 50 kyr (Lau et al., 2016; Kipp and Tissot, 2022; Pimentel-Galvan et al., 2022; Fig. 7). Agreement between independent reconstructions of seafloor anoxia based on δ²³⁸U and uranium concentration data further suggests that both datasets contain a strong original redox signature rather than local or diagenetic overprints (Pimentel-Galvan et al., 2022).

It is now clear that the latest Permian ocean was similar to the modern in terms of both temperature and widespread oxygenation and that the mass extinction event occurred in association with a substantial increase in atmospheric carbon dioxide levels, climate warming, and ocean deoxygenation. The central challenge in using environmental paleo-proxy data to infer extinction causes is the mapping between proxy data (environmental stress) and physiological response. The uranium isotope modeling indicates that anoxia was widespread (a best guess of 18% of the global seafloor in Pimentel-Galvan et al., 2022). Anoxia was likely concentrated below storm wave base in outer shelf or slope environments, although truly anoxic and/or sulfidic conditions may have impinged into shallower water in some areas (Grice et al., 2005; Schobben et al., 2020). However, with near-modern atmospheric oxygen (i.e., 100% PAL) it is nearly impossible to drive the entire ocean anoxic (Fig. 3E). This is particularly true for shallow waters in direct connection with the atmosphere, and indeed, well-aerated shoreface environments in the earliest Triassic can record robust and healthy animal communities (Beatty et al., 2008). Given the difficulty of developing widespread anoxia in the locus of animal biodiversity (shallow shelf), most likely anoxia and/or euxinia was an important local or regional kill mechanism but not the global kill mechanism. Rather, geochemical evidence for increased anoxia in deeper-water settings (e.g., light δ²³⁸U carbonate values) are an indirect signal of the real killer: the development of widespread low-oxygen (dysoxic) conditions on the shelf combined with high temperatures.

Progress in understanding the relative biotic impact of these stressors has been made through a workflow incorporating (1) an Earth system model calibrated with available geochemical proxy data, (2) an ecophysiological model of temperature-dependent hypoxia (the metabolic index), and (3) comparison between model results and the fossil record. Application of this approach to the end-Permian mass extinction demonstrates that changes in aerobic habitat due to ocean warming and resulting ocean deoxygenation can account for the observed latitudinal gradient and extinction intensity in the initial major pulse of end-Permian extinction (Penn et al., 2018). Specifically, this study used a general circulation model with an end-Permian paleogeography to simulate a major climate warming event (via an increase in modeled partial pressure of carbon dioxide [pCO₂] from 150 to 5580 ppm). The resulting paleoceanographic model provided a good fit to available U isotope constraints and paleotemperature proxies (Sun et al., 2012; Lau et al., 2016). As discussed above with respect to early Paleozoic extinctions, the Penn et al. (2018) study also simulated “ecophysiotypes” by sampling from the distributions of physiological parameters measured in modern species. Similarly, the availability of aerobic habitat for these species within the model was assessed before and after the modeled CO₂ injection event.

The simulated pre-extinction ocean was habitable to at least some species in all regions, and aerobic habitat declined for nearly all species after the injection event as a result of the combined effects of deoxygenation and warming. Because of changes in the wind field, ocean currents, and primary productivity, the upper regions of the tropical ocean experienced substantial warming but little deoxygenation, whereas the polar regions experienced more substantial deoxygenation. Consequently, loss of aerobic habitat in the model was governed more by deoxygenation near the poles and more by warming near the equator (Fig. 8B). Overall, predicted loss of habitat and extinction was greater near the poles and lesser, albeit still substantial, in the tropics (Fig. 8B). Intriguingly, the ecophysiological model results also suggest that the latitudinal pattern of extinction intensity was driven as much by the spatial distribution of species with different physiological capacities in the pre-extinction Permian ocean as by the spatial pattern of environmental change during the event itself. Ecophysiotypes living in the tropics prior to the extinction were characterized by traits pre-adapted to warm, low-oxygen environments (low hypoxic thresholds and low ratios of active to resting metabolism, or Φ_crit) and were able to exploit such conditions when they arose globally. Conversely, higher latitudes in the pre-extinction ocean could support ecophysiotypes with high hypoxic thresholds and higher Φ_crit. These ecophysiotypes had no escape and were preferentially exterminated when the high latitudes warmed and lost oxygen.

Model predictions for the biogeography of the end-Permian mass extinction based on application of the metabolic index generally agree with evidence from the fossil record. Extinction was severe everywhere but eliminated a greater proportion of marine animal genera at temperate and high latitudes than in the tropics (Fig. 8A; Penn et al., 2018). This latitude gradient in extinction does not appear to reflect variation in the quality of the fossil record before and after the extinction and appears independently in analyses separating the phyla with more sophisticated respiratory and circulatory systems and greater motility (Arthropoda, Chordata, Mollusca) from other well-fossilized phyla (Bryozoa, Brachiopoda, Cnidaria, Echinodermata) (Penn et al., 2018, fig. S14A). The model thus does not currently explain the difference in extinction intensity across classes and phyla identified by Knoll et al. (1996), suggesting other physiological stresses, such as hypercapnia, or differences in metabolic index parameters across clades not yet captured by existing (and limited) experimental data. Differences in the latitude gradient of extinction when considering later Griesbachian taxonomic losses (Song et al., 2020) may also point toward the importance of other stresses in causing extinctions after the initial pulse at the Permian-Triassic boundary itself.

Regardless of whether details of the Penn et al. (2018) findings are maintained following the collection of additional paleo-proxy, fossil, modeling, and experimental respirometry data, the principle of the approach is essential to use ancient episodes of climate change and extinction to test hypothesized extinction mechanisms and to inform projected biological responses to climate change in the twenty-first century. Simple, quantitative measures of habitat viability are required to translate between environmental changes determined from paleo-proxy data and biological responses observed from fossils. Biological traits that are conserved at high taxonomic levels, such as basic respiratory and circulatory anatomy, or are directly measurable from fossils, such as body size, are most amenable to inclusion in measures of habitat viability; but the influence of other variables can be determined by sampling from distributions constrained by measurements of living species.

Lessons from the Geological Oxygen Record for Modern Global Change

When attempting to draw lessons or analogies from the geologic past to modern and future global change, the major stumbling blocks are magnitude and timescale. The magnitude of both environmental change and marine extinction during the Big Five mass extinctions is simply not comparable to modern events—except under extreme scenarios of anthropogenic global change, we do not expect ∼12 °C of tropical surface warming, near complete loss of oxygen in the ocean interior, and upward of 90% species extinction, such as characterized the Permian-Triassic boundary (Penn et al., 2018; Penn and Deutsch, 2022). Conversely, while high-precision geochronology can only provide upper estimates for the timescale of ancient environmental change when such change was very rapid (i.e., <50,000 years for the Permian-Triassic), and the geological record systematically under-estimates rates of climate change (Kemp et al., 2015), there is no evidence that geologic forces come anywhere close to humanity’s ability to rapidly transfer carbon from rock reservoirs to the atmosphere. Thus, modern organisms are likely facing a much lower-magnitude perturbation but a faster rate of change, which restricts the ability of natural feedback processes to mitigate environmental stresses. Studies of Pleistocene records (e.g., Moffitt et al., 2015; Salvatteci et al., 2022) may provide more appropriate scales of oxygen or temperature change in terms of time and magnitude than the eye-catching Phanerozoic extinctions, although even here the timescale of change was still slower than the modern. Despite these issues, the geological record of oxygen changes (and temperature-dependent hypoxia) offers several important lessons for modern global change.

It is clear from the early Paleozoic record of oxygen and life that animal communities living under low atmospheric oxygen levels had very little aerobic buffer and were prone to rapid extinction from relatively modest changes in oxygen, temperature, or both. This has implications for the modern analog environment, specifically animals living in or near OMZs along continental margins. The geological observations provide a historical precedent consistent with observations from physiology (Wishner et al., 2018) and sharp ecological thresholds in animal diversity or abundance with declining oxygen levels (Mullins et al., 1985; Wishner et al., 1990; Gooday et al., 2009; Levin et al., 2009; Sperling et al., 2016). Together, these data from the ancient and modern Earth suggest that animals living in these environments today are likely to be especially vulnerable to loss of aerobic scope. Although such environments are often characterized as “muddy deserts,” they host a mosaic of habitats and ecosystems with high regional diversity as well as increasingly important (and likely unsustainable) fisheries that have been developed as shallower-water species and stocks are overfished (Morato et al., 2006; Levin and Sibuet, 2012). Perhaps equally important to consider with respect to future oxygen changes are impacts on the bioturbation activities of slope denizens. Continental margins are important sites of nutrient and carbon cycling (Berner, 1982), mediated in large part by animal bioturbators. While there have been biogeochemical studies of bioturbators in the modern ocean (e.g., Aller and Aller, 1998), researchers have generally not directly investigated, at either the local scale or the global scale, the biogeochemical effects in a warming or deoxygenating ocean. In contrast, the action of bioturbators in the poorly oxygenated early Paleozoic has recently received increased scrutiny (albeit with varying modeling outcomes; Canfield and Farquhar, 2009; Boyle et al., 2014; Tarhan et al., 2015; van de Velde et al., 2018). Most likely, the effects of global change on bioturbation are heterogenous and require nuance in global modeling efforts; for instance, biodiffusion mediates phosphorus burial, whereas bioirrigation recycles P to the water column and enhances productivity (Tarhan et al., 2021). In any case, the lessons from and modeling approaches applied to the ancient low-oxygen world regarding bioturbation are likely to be useful in considering the biogeochemistry of our future oceans.

The most useful lesson from the ancient record (especially as a counter-point to climate change skeptics who claim that modern climate change could just be natural cycles or who question the link between CO₂ and climate change) is that over the past 400 million years, every time volcanic forces have injected massive amounts of CO₂ into the Earth system, we see evidence in the geologic record for global warming, ocean deoxygenation, ocean acidification, and widespread extinction of marine species. The timescales and magnitudes may be different, but the repeated response to natural carbon cycle perturbations is the same as that observed today in response to anthropogenic fossil fuel combustion. Notably, biological recovery from such events takes hundreds of thousands to millions of years, as long as or longer than the timescale of return to pre-eruption environmental conditions. From a stressor perspective, these major extinction events all involve both deoxygenation and warming. Ecophysiological modeling of the Permian-Triassic extinction (Penn et al., 2018) suggests that it was the synergistic combination of these two factors—with varying contribution by latitude—that best explains the pattern of extinction in the fossil record. This finding is consistent with recent meta-analyses suggesting that the combination of oxygen and temperature is by far the most detrimental stressor combination to marine invertebrates (Reddin et al., 2020; Sampaio et al., 2021) and highlights the importance of studying not just oxygen or temperature but, rather, temperature-dependent hypoxia.