Introduction

The Paleocene-Eocene Tfhermal Maximum (PETM) occurred around 55.9 million years ago (Ma) and lasted for 150–220 kyrs. It is probably the best documented hyperthermal event of the Phanerozoic and serves as a geologic analog for ongoing climate change1,2,3. The large release of carbon from methane hydrates or volcanic activity to the Earth’s surface drove a 4–8 °C increase in global surface temperature and carbon isotope excursions (CIE) of 3–6‰ at that time2,3. Rapid carbon emission and climate warming resulted in large perturbations of the marine environment and ecosystem, including global ocean acidification4,5, deoxygenation6,7 and biological turnover8. However, patterns of environmental change during the PETM are complex and far from the global-scale deterioration that punctuated other warming-induced crises from Earth’s geological past. First, due to the relatively short duration of the PETM, the environmental consequences of global climate change are proposed to be spatially variable9,10. In addition, the environmental and biological responses during the PETM are likely weaker than those during other hyperthermal events; the PETM is marked by an overall limited expansion of global seafloor anoxia11,12 and the appreciable extinction of only benthic foraminifera8, which is notably less than the changes in anoxia and biodiversity loss reported during e.g. the Permian–Triassic transition13. Marine redox changes also exhibit complex spatial patterns, with oxygen concentrations even rising in the tropical upper ocean14.

Rapid increases in continental chemical weathering in response to rising CO2 concentrations (pCO2) and surface temperatures have been typically considered to promote the delivery of nutrients to the ocean, enhancing marine productivity and oceanic deoxygenation during the PETM12,15,16. Yet, causal relationships among carbon emission, chemical weathering and marine redox conditions remain poorly constrained, due to the lack of quantitative reconstructions of the coupled evolution of continental weathering regimes and marine productivity during the PETM.

Here, we report coupled lithium isotope (δ7Li) data—a proxy for continental silicate weathering—in shallow-marine carbonate and siliciclastic successions deposited within the epicontinental sea in the Paratethys during the PETM (Fig. 1a). Combining these δ7Li data with previously published ones from both pelagic settings and a terrestrial floodplain17,18, we document the impact of global warming on continental silicate weathering regimes and the formation of detrital materials at both global and regional scales. Our geochemical data, together with Earth system model experiments, suggest that chemical weathering of fresh igneous rocks decreased globally while physical erosion simulatenously increased in many regions during the PETM, respectively explaining the limited expansion of anoxia during the event and the rapid recovery through organic carbon burial.

Fig. 1: The integrated isotopic records across the Paleocene-Eocene Thermal Maximum (PETM) (in kyr relative to PETM onset).
Fig. 1: The integrated isotopic records across the Paleocene-Eocene Thermal Maximum (PETM) (in kyr relative to PETM onset).
Full size image

a Paleographic map and sample locations during the PETM (PALEOMAP Project by Scotese and Wright, 2018 https://www.earthbyte.org/paleodem-resource-scotese-and-wright-2018/). b Carbonate (δ13Ccarb) of the Tingri section (this study) and Ocean Drilling Program (ODP) Site 69019 and organic carbon (δ13Corg) records of the Kuzigongsu section. c Carbonate δ7Li records (δ7Licarb) from the Tingri section (this study), and ODP Sites (865C, 1210, 1051)17. d Detrital δ7Li records (δ7Lidet) from the Tingri section and Kuzigongsu section (this study), Fur and Svalbard sections17, and the floodplains in the Bighorn Basin18. e δ15N of formainifera-bound organic matter (δ15NFB) records from ODP Sites (865, 690, 213, 1210, 1263)14. f Carbonate δ238U records from ODP Sites (865, 690, 401)12. Age correlations of each profile and definitions of PETM peak and recovery are modified based on δ13Ccarb curve and age model from the ODP Site 690—the interval of PETM peak initiated at the onset of negative δ13Ccarb excursion and terminated prior the recovery of δ13Ccarb excursion, and the PETM recovery initiated at the onset of negative δ13Ccarb excursion until the δ13Ccarb returned to the baseline value19,86,87. The blue zones in (f) represents the average δ238U values of modern inorganic carbonates and seawater97.

Results and discussion

Lithium isotope signals in carbonates and siliciclastic rocks reveal changes in continental weathering regimes and sources

The newly analyzed and compiled geochemical data in this study are temporally integrated based on the typically established age model and δ13C curve from ODP Site 6901,19, delineating the interval of PETM peak between the initiation of the CIE and the onset of its recovery (Fig. 1). Using a sequential leaching approach, we obtained the separated δ7Li records of carbonate and carbonate-hosted detrital components from a shallow-marine carbonate succession in the Tingri County, south Tibet (Fig. 1). After discarding potential impacts of detrital contamination and diagenetic alteration on our results (see Supplemental information for details), we report an overall negative δ7Licarb excursion of ~3–4‰ from shallow-marine carbonates in the epicontinental sea of the Paratethys during the PETM peak. The trend and extent of this negative δ7Licarb excursion are comparable to those from deep-marine carbonate sediments from the North Pacific and North Atlantic17, although the magnitude of δ7Licarb may slightly differ between regions due to the potential effects of local depositional environments (Fig. 1C). The analogous trends of δ7Licarb records in deep-marine and shallow epicontinental settings bolsters the case that the seawater δ7Li excursion during the PETM peak is of global significance, reflecting rapid and notable perturbations of global climate and environment. Meanwhile, although the oceanic Li residence time (~1 Myr in modern ocean)20 was likely much longer than the duration of the PETM (~150–220 kyr), such discernable excursions point towards large-magnitude changes in both Li fluxes and δ7Li values of marine Li budgets. Global seawater δ7Li is generally determined by Li isotope composition and flux of sources (river waters, high-temperature hypothermal fluids) and sinks (marine authigenic clays and oceanic crust alteration). Under very short time-scale (e.g., 150–220 kyr across the PETM), Li isotope and flux of hypothermal fluids and oceanic crust alteration do not significantly vary due to limited isotopic fractionations at high temperature and low mid-ocean ridge spreading rates17. Extents of marine authigenic clays and oceanic crust alteration may also affect global seawater δ7Li values, while these influences are commonly significant over million-year time scales, such as the Early Triassic and late Cenozoic21,22,23,24. Further, the PETM is not marked by notable turnover of marine siliceous organisms and decreased biogenic Si burial25, thus variations in Li isotopic fractionation during authigenic clay formation associated with secular evolution of the biogenic Si production21,23,24, cannot account for rapid changes in global seawater δ7Li during the PETM. In contrast, riverine Li flux and δ7Li may largely vary accompanying the shift in silicate weathering regimes26, even under seasonal scales27. Taken together, we suggest that changes in continental weathering regimes (or intensity) are fundamentally responsible for abrupt decreases of ~3–4‰ in global seawater δ7Li during the PETM.

Based on the interpretation framework for Li isotopic behavior during silicate weathering in modern rivers17,26 (Fig. 2a), substantial decreases in global seawater δ7Li under short-time scales require synchronously and rapidly decreased riverine δ7Li values and increased Li fluxes, indicating lower weathering intensity and a more kinetically-limited regime (referred to as lower W/D, the ratio of chemical weathering rate to total denudation rate, and total denudation rate is the sum of chemical weathering rate and physical erosion rate)26,28,29. Accordingly, it can be suggested that the PETM is characterized by relatively less clay formation relative to primary rock dissolution on continents, and accordingly greater increases in physical erosion rates relative to the chemical weathering rates17. The baseline of weathering intensity before the PETM may have been higher than the present day (high W/D side in Fig. 2a) due to relatively low topography and warm climate, indicative of a supply-limited weathering regime during the early Cenozoic30,31. This evaluation of weathering regime before the PETM is also similar to that of the Eocene average levels30. Under such a regime of high weathering intensity before the PETM, rapid decreases in global seawater δ7Li require significantly enhanced physical erosion and drecreased weathering intensity during the PETM peak. Alternatively, a regime of relatively lower weathering intensity than present day (low W/D side in Fig. 2a) for the pre-PETM baseline suggests a moderately increased physical erosion and decreased weathering intensity during the PETM peak, which likely reflects combined increases in both physical erosion and chemical weathering17. Despite uncertainties on overall weathering intensity before the PETM, abrupt decreases of ~3–4‰ in global seawater δ7Li provide clear evidence for enhanced physical erosion and decreased weathering intensity (i.e., more kinentic-limited weathering regime) during the PETM (Fig. 2b). The notable response of global physical erosion to climate change during the PETM is also supported by notably elevated sedimentation rates in many proximal marine locations9,32,33, and increased distributions of mud clasts in global continent-marginal and deep-marine sites, which denote enhanced terrestrially detrital inputs to the ocean2,34,35.

Fig. 2: Potential changes in weathering intensity or regime across the Paleocene-Eocene Thermal Maximum (PETM), and comparisons with other geological events during the Phanerozoic.
Fig. 2: Potential changes in weathering intensity or regime across the Paleocene-Eocene Thermal Maximum (PETM), and comparisons with other geological events during the Phanerozoic.
Full size image

a The relationship between δ7Li and weathering intensity (W/D) (i.e., the boomerang shape) is obtained from compiled data from the modern river waters26,28,29 (gray data points, see Supplement Information for details). The estimated weathering intensities for the Hirnantian glaciation and the Middle Eocene Climate Optimum (MECO) are derived from refs. 30,98. b The schematic relationship between physical erosion and chemical weathering is modified from ref. 99.

Our δ7Lidet records of shallow-marine carbonate and siliciclastic successions exhibit similar temporal trends, with δ7Lidet shifting to lower values roughly prior to the onset of CIE in marine carbonates (Fig. 1). The temporal offset between detrital and carbonate δ7Li records may suggest that the terrestrial environment responded more rapidly to climate warming than the marine system during the PETM, or that the marine system is more buffered during the perturbations of global climate and environment. Comparing the δ7Lidet data in Kuzigongsu and Tingri sections to other regions (e.g., Bighorn Basin, Svalbard and Fur setions17,18), we observe largely variable δ7Lidet values in different regions (Fig. 1d). Ancient marine siliciclastic sediments may also consist of authigenic clays that drive δ7Li of bulk sediments towards higher values23,36,37. We first use Li/Th vs. Al/Th and δ7Lidet vs. K/Al ratios to examine any potential effects of marine authigenic clays on the δ7Lidet records during the PETM (Fig. 3), as authigenic clays are preferentially enriched in Li and K relative to Al and Th, due to additional incorporation of Li and K from seawater or sediment pore water37,38,39. Compared to typical aluminosilicates in marine sediments dominated by authigenic clays (e.g., the lower interval of Site U1366 from the South Pacific Gyre)40, the siliciclastic sediments during the PETM overall exhibit a lack of appreciable enrichments in Li and K (Fig. 3a) with relatively uniform and low K/Al ratios, except for several samples with relatively high δ7Li and K/Al (Fig. 3b), which provides clear evidence for little accumulaction of marine authigenic clays in studied samples. Mineralogical analyses by X-ray diffraction also support this result, which show dominance of continentally-derived detrital silicates (i.e., smectite, illite, chlorite, feldspar and quartz) in these samples (Supplementary Fig S4). All above results suggest that the siliciclastic sediments in this study are not dominated by marine authigenic clays, thus changes in δ7Lidet signals are ultimately driven by the evolution of continental silicate weathering, rather than marine reverse weathering. The siliciclastic sediments deposited on continental shelves typically reflect a mixture of signatures coming from newly-weathered products and from eroded older sedimentary and primary rocks (reflecting the cannibalistic nature of continental weathering and erosion), the latter experiencing sediment recycling in the river floodplain36,37,41,42. Hence, the observed heterogeneity in δ7Lidet likely reflects differences in provenance and chemical weathering intensity at the local to regional scale17.

Fig. 3: Evaluation of origins for siliciclastic components in sedimentary rocks across the Paleocene-Eocene Thermal Maximum.
Fig. 3: Evaluation of origins for siliciclastic components in sedimentary rocks across the Paleocene-Eocene Thermal Maximum.
Full size image

Cross-plots of a Li/Th vs. Al/Th, b δ7Li vs. K/Al, and c TiO2 vs. Zr for carbonate-hosted silicates in the Tingri section and mudstones in the Kuzigongsu section (this study), and the floodplain clays in the Bighorn Basin18. Different colored points in (a) represent the compostions of ancient detrital and authigenic clays preserved in marine sediments (Edicaran sedimentary rocks and aluminosilicate components in Integrated Ocean Drilling Program Site U1366 core)38,40. The dark green and orange areas in (b) represent the end members of igneous rock (K/Al = 0.17 ± 0.10 for average igneous rocks; δ7Li = +4.8‰ ± 1.4‰ for andesites, and +2.0‰ ± 2.3‰ for granites)26,45,100) and fine-grain sedimentary rock (referred to as shale) (K/Al = 0.33 ± 0.23; δ7Li = –0.5‰ ± 1.9‰ without the Precambrian data, and +0.48 ± 2.5‰ with recently published Precambrian data)26,37,46,47,100. The gray areas with dashed lines in (b) represents the K/Al values of authigenic and detrital clays documented in U1366 core (K/Al = 0.51 ± 0.09 for authgenic component, and 0.30 ± 0.05 for detrial component)40. The divisions of different types of igneous rocks in (c) are modified after ref. 43.

To better understand the origin of δ7Lidet perturbations during the PETM, we initially use a typical diagram of TiO2 vs. Zr contents43 to examine the primary weathering source of the detrial silicates in this study (Fig. 3c), which indicates a dominance of felsic igneous rocks as the original weathering source of the detrital silicates during the PETM. The relationship between δ7Lidet and CIA (chemical index of alteration) records of detrital silicates in this study and modern riverine sediments are further compared (Fig. 4a). Modern riverine fine-grained sediments exhibit a significant negative correlation between δ7Lidet and CIA, which can be related to regional W/D changes (Fig. 4a). Together, lower δ7Lidet and higher CIA values generally indicate higher W/D ratios (i.e., enhanced chemical weathering intensity, lower physical erosion rates, and low Li fluxes) in modern riverine systems44. Hence, if the negative excursions of δ7Lidet during the PETM represented changes in weathering regimes at the global scale, this would imply an increase in the W/D ratio44 (Supplementary Fig. S6), which would not be consistent with our δ7Licarb records. We further note that, although δ7Lidet and CIA correlate in the Kuzigongsu and Tingri sections, and the Bighorn basin, identified correlations strongly deviate from the one established for modern riverine sediments (Fig. 4a), which implies that weathering intensity may not be the sole controlling factor on δ7Lidet and CIA records in the studied samples. Given that continental silicate weathering and erosion are commonly cannibalistic through geological time37,41,42, changes in δ7Lidet signals documented in bulk marine sedimentary rocks may reflect the relative contributions of different weathering sources (e.g., chemical weathering products, unweathered rock fragments, and recycled sedimentary rocks), which may be also related to size sorting during their deposition41. Based on the modern riverine interpretative framework41,44, we constrain the evolution of weathering regimes during the PETM by using the diagrams of δ7Lidet vs. Li/Al and Na/Al vs. Li/Al of the samples, and comparing their compositions to those of typical rock endmembers (Fig. 4b, c). The endmember of igneous rocks generally have high δ7Li, Na/Al and low Li/Al values, while the endmember of sedimentary rocks (referred to as shales) show relatively low δ7Li, Na/Al and high Li/Al values41,44. During chemical weathering, Na and Li are preferentially released to dissolved loads relative to Al, leading to weathering products enriched in isotopically light Li with low Na/Al and Li/Al values relative to their bedrocks41 (Fig. 4b, c). In addition to the typical endmembers of igneous rocks (andesites) and shales used for riverine sediments from the modern Amazon basin26,41, we extend the δ7Li compilations for different endmembers, including the granites and Precambrian shales37,45,46,47. The granite endmember has overall lower δ7Li relative to privously used andesite endmember41, and partially overlaps with the shale endmember. Neverthless, distinctions among the endmembers in terms of δ7Li, Li/Al, and Na/Al are appreciable, particularly when examining the correlation trends within the sample data (Fig. 4b, c). Clear negative correlations of δ7Lidet vs. Li/Al and Na/Al vs. Li/Al of the detrital silicates in the Kuzigongsu, Tingri, and Bighorn sections indicate the mixing trend of igneous and sedimentary endmembers with distinct elemental and isotopic signatures, rather than the weathering trends of these endmembers, supporting that decreased δ7Lidet values in these sections reflect enhanced erosion and weathering of shales during the PETM. The δ7Lidet and Li/Al values of pre-PETM samples are overall higher than those of the granite endmember, likely reflecting additional Li incorporation from seawater or pore water during early diagenesis (Fig. 4b). In constrast, the decreasing trends of δ7Lidet during the PETM peak more likely reflects higher proportions of eroded fragments from previously formed shales (i.e., sedimentary rocks) with higher Li/Al and lower δ7Li values. This shift in weathering sources or regimes are consistent with the results from clay-sized floodplain sediments in the Bighorn Basin (USA)18, which clearly show the weathering of older sedimentary rocks during the PETM (Fig. 4b, c). Consequently, an increase in the erosion and weathering of previously-formed clay-rich sediments likely promotes the delivery of detrital clays (e.g., illite, kaolinite or smectite) to continental margins, and their preservation in multiple siliciclastic successions during the PETM peak48,49,50,51, which is also consistent with the results of X-ray diffraction for the samples in this study—the samples during the PETM peak is marked by more distributions of clay minerals relative to the pre-and post-PETM samples (Supplementary Fig. S4).

Fig. 4: Changes in silicate weathering regimes and sources across the Paleocene-Eocene Thermal Maximum.
Fig. 4: Changes in silicate weathering regimes and sources across the Paleocene-Eocene Thermal Maximum.
Full size image

Cross-plots of a δ7Li vs. CIA, b δ7Li vs. Li/Al, and c Na/Al vs. Li/Al for carbonate-hosted silicates in the Tingri section and mudstones in the Kuzigongsu section (this study), and the floodplain clays in the Bighorn Basin18. Different colored circles in (a) are derived from modern riverine fine-sediments with different weathering intensities (W/D) shown in the color bar at the right side44. The dark green and orange areas in (b) and (c) represent the end members of igneous rock (Li/Al = 0.3 ± 0.11; Na/Al = 0.32 ± 0.03; δ7Li = +4.8‰ ± 1.4‰ for andesites, and +2.0‰ ± 2.3‰ for granites)41,45,100 and fine-grain sedimentary rock (referred to as shale) (Li/Al = 1.05 ± 0.1; Na/Al = 0.08 ± 0.03; δ7Li = –0.5‰ ± 1.9‰ without the Precambrian data, and +0.48 ± 2.5‰ with recently published Precambrian data)37,41,46,47. The black arrows in (b) and (c) denote the weathering processes of igneous and sedimentary rocks, while the dashed lines denote the mixing trends of different silicate components41.

Although the δ7Li records of siliciclastic rocks conservatively reflect regional signals specific to each sedimentary basin, the remarkably consistent trends observed in the Kuzigongsu, Tingri, and Bighorn sections—deposited in widely separated regions—suggest that these patterns may carry broader, potentially global, implications. Combined Li isotope records from marine carbonates and siliciclastic rocks suggest that the PETM was characterized by concomitant decreases in weathering intensity primarily caused by increases in physical erosion rates, which are likely related to increased storminess and vegetation degradation9,52,53. Meanwhile, the relatively low topography and warm climate from the late Cretaceous to the early Paleogene may have resulted in a transport-limited weathering condition with moderately high weathering intensity, which facilitated the accumulation of abundant clay-rich sediments in peneplained continents prior to the PETM31,54,55. Although additional work is needed to explain these changes, coincident changes in δ7Licarb and δ7Lidet may point towards a shift to more kinetically-limited weathering regimes56, with enhanced exhumation of older sedimentary rocks during the PETM.

Response of oceanic oxygen concentrations to changing weathering regimes

Abrupt carbon emissions and ensuing global climate warming events during the Phanerozoic have been considered to drive rapid environmental deterioration13. Rapid increases in chemical weathering in response to global warming have been commonly considered to increase the flux of nutrients to the ocean, enhancing marine productivity and triggering anoxia expansion15,16,57,58, hence destabilizing ecosystems13. However, the modulation of phosphorus supply by continental weathering changes depends on both the nature of weathering regimes and outcropping continental lithologies59,60. Our Li isotope records provide geochemical evidence for decreased weathering intensity and increased physical erosion and weathering of older sedimentary rocks during the PETM, which agrees with records of other warming events in the geological record because warming may accelerate the hydrological cycle, thus increase erosion rates relatively greater than weathering rates17,30. The responses of silicate-bound Li weathering and phosphorus weathering to changes in weathering source and regime may have been decoupled through geological time59,61,62. Most of Li flux in modern river systems is especially derived from the weathering of sedimentary rocks, even in catchments dominantly underlain by igneous rocks26,63. Thus, enhanced erosion and weathering of sedimentary siliciclastic rocks can lead to substantial release of Li with low isotopic values to surficial run-off, which notably influences the marine Li budgets, and then results in rapid negative δ7Li excusions of global seawater64. In constrast, the laboratory experiments of mineral dissolution showed that dissolution rates for igneous fluorapatite are significantly higher than those for sedimentary carbonate fluorapatite—~100 times at circum-neutral pH conditions65,66,67 that represent the overall range of soil pH at the global scale68. The modern natural observations that exposed igneous rocks cover ~15% of the land area but contribute ~38% of the total phosphorus release flux60, also suggest that igneous rocks are likely greater contributors to global phosphorus weathering59. Addtionally, phosphates formed via apatite weathering can be locally absorbed by clay minerals, inhibiting the release and transport of phophorus to oceans58. Taken together, considering the short duration of the PETM, intensified erosion and weathering of clay-rich sedimentary rocks may not have led to substantial increases in available phosphorus inputs to oceans, which is likely different from previous modeling estimations of increased phosphorus weathering flux under warming climate conditions12,58. We, thus, provide a revised scenario for the response of marine environment to climate warming during the PETM—changes in continental weathering regimes decreasing the delivery of nutrients, such as phosphorus, to the ocean.

As an independant means of evaluating the validity of our interpretative framework for environmental changes during the PETM, we quantified the impact of different changes in oceanic nutrient concentrations on dissolved oxygen concentration ([O2]) during PETM warming using the Earth system model of intermediate complexity cGENIE69 (see Materials and methods). We configured the model with an Eocene configuration providing reasonable global ocean circulation patterns, as testified by the correlation between simulated and observed benthic foraminiferal δ13C records70. The cycling of carbon and marine biological productivity were based on co-limitations by phosphate and iron71, and water-column organic matter remineralization was parameterized using a calibrated temperature-dependent scheme72. We imposed an atmospheric oxygen concentration of 1.2 times the modern value12. The model was integrated for 20,000 years until equilibrium under pCO2 values of 834 ppm and 2176 ppm, respectively representative of pre-PETM and PETM maximum warming (~40 kyrs after PETM CIE onset) conditions in an inverse modeling approach consistently conducted using the same Eocene model configuration5. Although the PETM was a transient warming event, the duration between the event onset and maximum warming (~40 kyrs) is an order of magnitude larger than the adjustment timescale of the global ocean circulation (of a few kyrs). Therefore, steady-state simulations constitute good approximates of the response of marine biogeochemistry to PETM maximum warming, in line with previous modeling investigations of other hyperthermal events73,74.

We quantified the impact of changing oceanic nutrient concentrations on oceanic [O2] during the PETM. The continental weathering module implemented in cGENIE represents continental weathering as a simple function of the mean annual continental surface air temperature75. It offers no representation of the interplay between changes in chemical weathering and physical erosion on the flux of nutrients delivered to the ocean76. This, combined with the lack of a global lithology map in the deep time, prevents us from mechanistically simulating the flux of nutrients delivered to the ocean and the respective contributions of igneous vs. sedimentary rocks on a global scale during the PETM. Instead, we tested three contrasting scenarios by changing the mean oceanic phosphate concentration ([PO4]) in cGENIE (+30%, no change and –20% relative to the modern value of 2.159 μmol kg–1). The strongest perturbation (+30% increase in [PO4]) is derived from the biogeochemical box model of the coupled C, P and U cycles of ref. 12 forced with carbon injections and constrained by δ238U proxy data (but ignoring our δ7Li data), which suggests a ~30% increase in oceanic P inputs during the PETM. Because the oceanic residence time of P, on the order of 40 kyrs77, is similar to the time elapsed between the PETM onset and maximum warming, this rate change directly translates into a +30% change in the total oceanic P inventory, or equivalently in global-mean oceanic [PO4], during PETM maximum warming. For comparison, this oceanic [PO4] change is smaller than the doubling invoked in cGENIE over a similar duration to explain redox proxy data during the Permian-Triassic extinction73. Conversely, our scenario of a decrease in the mean oceanic [PO4] by 20% would imply a mean decrease by 20% of the P flux to the ocean during the period of time between the PETM onset and PETM maximum warming. Only altering phosphate concentrations is aimed at specifically representing variations in nutrient fluxes resulting from continental weathering. Iron concentrations are kept constant in our simulations, as eolion dust and volcanic ash rather than runoff from land constituting the main source of iron to the surface ocean16.

All three scenarios of [PO4] change lead to relatively similar responses of the deep-ocean [O2] (Fig. 5), which below 2000 m depth is of 255 μmol kg–1 in our pre-PETM simulation (at 834 ppm) and decreases by 31 μmol kg–1, 27 μmol kg–1 and 21 μmol kg–1 at 2176 ppm when increasing [PO4] by 30%, 0% and –20%, respectively. This slight deoxygenation mostly results from the decrease in oxygen solubility in warmer waters, while changes in oceanic circulation, hence ventilation, are minor in our simulations. The small deep-ocean [O2] decrease is in agreement with the muted expansion of global anoxia reconstructed based on lack of the notable changes in carbonate U isotopes (Fig. 1e)12, but does not permit discriminating between our different scenarios.

Fig. 5: Changes in oceanic [O2] during the Paleocene-Eocene Thermal Maximum in an Earth system model.
Fig. 5: Changes in oceanic [O2] during the Paleocene-Eocene Thermal Maximum in an Earth system model.
Full size image

Changes in oceanic [O2] simulated in the carbon-centric Grid Enabled Integrated Earth system model (cGENIE) at depths of ca. 230 m and 2300 m below sea level (b.s.l.) in response to global warming during the Paleocene-Eocene Thermal Maximum and different changes in the mean oceanic phosphate concentration relative to the modern mean concentration: +30% (or equivalently ×1.3, top row), +0% (no change; middle row) and –20% (bottom row). In all simulations, global warming is simulated by increasing pCO2 from 834 ppm to 2176 ppm during the Paleocene-Eocene Thermal Maximum5. Eckert IV projection with latitudes shown every 30°. Grid points that are land at the depth level shown are shaded white.

In comparison with the deep ocean, the upper ocean exhibits a stronger response to changing oceanic [PO4]. In our first scenario of unchanged [PO4], upper-ocean [O2] decreases in most regions as a result of the concomitant increase in export productivity and decrease in oxygen solubility in seawater in response to global warming (Fig. 5). The unexpected outcome of this simulation, however, is that [O2] increases over large portions of the low-to-mid latitude upper ocean. This regional oxygenation arises from the temperature-driven increase in microbial respiration of sinking organic matter, which renders remineralization more intense in a warmer world, hence shifting remineralization to shallower depths and reducing [O2] consumption deeper down – as can be seen here at ca. 230 m depth. This analysis is supported by the absence of such [O2] increase in additional simulations conducted with a less mechanistic representation of the biological pump featuring no dependence of remineralization on temperature (Supplementary Fig. S7). The simulated [O2] increase, however, does not reach the low-latitude middle Pacific, which conflicts with the decreases in foraminifera-bound nitrogen isotopes that were recently used to demonstrate an oxygenation of the tropical upper Pacific during the PETM (Fig. 1f)14. An alternative scenario considering an increase in oceanic [PO4] in response to global warming, such as usually invoked (here using a +30% increase after the biogeochemical box model of ref. 12), further reduces low-latitude Pacific [O2] (Fig. 5) and thus strengthens model-data discrepancies14. In contrast, a moderate decrease (–20%) in [PO4] allows [O2] to increase over the entire low-latitude Pacific (Fig. 5), permitting to reconcile our simulation results with proxy data (Fig. 1f)14. This conclusion is supported by previous ocean-only general circulation model experiments14. It stands when an alternative, phosphate-only export production parametrization is employed (Supplementary Fig. S8). While our simulations consider that iron-rich dust input to the ocean surface remained constant, magnetic and geochemical analyzes suggest that iron fertilization by eolian dust and volcanic ash may have fostered export production in the eastern equatorial Pacific Ocean during the PETM16. Accounting for such iron-driven enhancement of export production in the eastern Pacific would require a stronger decrease in oceanic [PO4] to simulate the increase in low-latitude Pacific [O2] in line with proxy data14. This would lend further support to our δ7Li-based hypothesis of a decreasing flux of terrestrially-derived nutrients to the ocean.

Implications for marine environmental and biological perturbations during the PETM

Multiple lines of isotopic evidence have supported notable response of continental silicate erosion or weathering to rapid warming during the PETM17,78. For instance, increased 187Os/188Os records suggested enhanced continental weathering inputs during the PETM78. By comparison, the detrital δ7Li data in this study specifically indicate a dramatic shift in silicate weathering regimes and sources during the PETM. Enhanced erosion and weathering of shales rather than igneous rocks is proposed to mark the continental silicate weathering response to global climate change during the PETM, which can also account for increased radiogenic Os inputs to the ocean. In particular, shifts to more erosion and weathering of shales fundamentally lead to much less P release and transport to the ocean at short-term scales, notably influencing marine biogeochemical cycles and redox states during the PETM. Earth system model simulations confirm that a (resulting) decline in oceanic nutrient concentrations is required to simulate an oxygenation of the low-latitude upper Pacific ocean in agreement with proxy data14. Decreases in the terrestrial nutrient inputs are further supported by the records of nannofossil and authigenic Ba isotopes that demonstrate declining marine nutrient reservoir and export productivity during the PETM peak79,80.

A depressed response of nutrient release via chemical weathering to global warming helps explaining the muted expansion of seafloor anoxia during the PETM12. It also has implications for the PETM recovery. The rate of carbon sequestration has been shown to be an order of magnitude more rapid than expected from changes in silicate weathering alone, suggesting a major contribution of the sequestration of organic carbon in the seafloor81,82,83. Carbonate and detrital δ7Li records through the PETM suggested decreased weathering intensity in step with global warming. We do not argue against the case that chemical weathering rates increase during the PETM, whereas suggest that decreased weathering intensity (or weathering efficiency), due to a transition of weathering sources from igneous rocks to shales, likely leads to lower efficiencies of atmospheric CO2 consumption via silicate weathering than previously estimated for a greenhouse climate condition (Fig. 2b). Thus, increases in organic carbon burial in marine sediments are particularly required to contribute to the rapid recovery of CIE. A significant increase in organic carbon production seems to conflict with the limited expansion of seafloor anoxia during the PETM12, due to a possible decrease in biological export production at that time79,81,82, and our finding of decreased chemical weathering intensity and P release, which in our preferred Earth system modeling scenario (i.e., ×0.8 [PO4]; Fig. 5) translates into a 21% decrease in the total particulate organic carbon rain flux from the upper ocean to the seafloor sediments. We here propose that enhanced physical erosion of pre-exiting clay-rich rocks and subsequent deposition on continental margins, supported by δ7Lidet data and observations of increased sedimentation rates of continental marginal sediments32,33, may have, in part, enhanced the burial efficiency of organic carbon producing from the upper ocean in the absence of any major increase in biological productivity or expansion of seafloor anoxia37,84,85. This mineral protection of organic carbon inhibits the re-oxidation of organic matter via microbial decay in water columns, ultamely reducing the comsuption of dissolved O2 in deep seawater and facilitating the recovery of CIE following the PETM. We hence suggest that the coupled responses of chemical weathering and physical erosion to carbon emissions and global warming explain the reduced marine environmental and biological perturbations during the PETM compared to other hyperthermal events (e.g., Permian-Triassic transition).

Methods

The Kuzigongsu section in the Tarim basin was deposited along an east-facing terrigenous-rich foreland basin, representing a sedimentary environment of shallow epicontinental sea86. The Kuzigongsu section mainly consists of limestone, calcareous mudstone and marlstone, deposited following the large transgression across the Paleocene-Eocene boundary. The Tingri section in the Tethyan Himalaya of south Tibet was deposited in the Tethyan Himalaya of the northern Greater Indian continental margin, representing an epeiric, shallow-marine environment, >300 km away from the shoreline87. The Tingri section mainly consists of nodular limestone, nodular marl and limestone. The biostratigraphy and C-isotope chemostratigraphy of these two sections can be correlated to other PETM depositional records in deep-marine sites86,87.

Lithium isotope analysis

The mudstone and marlstone samples in the Kuzigongsu section were fully digested with distilled HF-HNO3-HCl acids to obtain the Li isotope compositions of the detrital components (δ7Lidet), before which a 0.5 M HAc acid was used to eliminate the carbonate minerals88. The carbonate samples in the Tingri section were sequentially leached using 1 M ammonium acetate, 0.5 M acetic acid, and distilled HF-HNO3-HCl acids in order to obtain Li isotope compositions of the carbonate (δ7Licarb) and carbonate-hosted detrital (δ7Lidet) components38. The Li isotopes along with major and trace elements of different sample solutions were analyzed using Thermo Neptune XT MC-ICP-MS and Element XR ICP-MS at the Centre for Research and Education on Biological Evolution and Environment, Nanjing University. The standards OSIL Atlantic seawater, BHVO-2 basalt and GSR-12 carbonate were used to monitor the accuracy of elemental and isotopic results in this study (see Supplemental information for details).

Earth system modeling

cGENIE (carbon-centric Grid ENabled Integrated Earth system model) is an Earth system model of intermediate complexity69, combining a biogeochemistry-enabled 3D ocean circulation model with a 2D energy–moisture balance atmospheric component and a sea-ice model. The model was configured on a 36 × 36 equal-area grid with 16 unevenly-spaced vertical levels to a maximum of depth of 5000 m in the ocean. In our main experiments, the cycling of carbon and the representation of biological export production were based on a phosphate and iron nutrient co-limitation71, and we adopted the Arrhenium-type temperature-dependent scheme for the remineralization of organic matter exported through the water column72,73. A standard model configuration of the early Eocene was used, with a 0.46% reduced solar luminosity and 1% decreased seawater salinity in the absence of land ice5,70,89. For these experiments, a spatially-uniform dust flux was designed, providing iron to the surface ocean, the globally-integrated value being set to 97% modern to account for the proportionally lower Eocene ocean surface area. To best capture the pre-PETM climatic state and PETM maximum warming, we used atmospheric pCO2 values (imposed in the form of radiative forcing) of respectively 834 ppm (or equivalently 3 times the pre-industrial value of 278 ppm) and 2176 ppm (7.8 times 278 ppm), obtained by ref. 5 on the same Eocene model configuration when simultaneously inverting boron isotope and carbon isotope data. We imposed an atmospheric oxygen concentration of 1.2 times the modern value12. In our pre-PETM simulation ran at 834 ppm pCO2, we used a modern mean oceanic phosphate concentration [PO4] = 2.159 μmol kg–1. For our PETM maximum warming simulations ran at 2176 ppm pCO2, we tested three scenarios of nutrient change during the PETM by increasing the modern [PO4] by 30% (according to refs. 12, 58), 0% (no change), and –20% (hence representing a moderate decrease). cGENIE simulations were initialized with a sea-ice free ocean and homogeneous temperature and salinity of global seawater, then integrated for a total of 20,000 years to reach a steady state. The results of the last simulated year were used in our analyses. Despite its low resolution, cGENIE has been shown to satisfactorily simulate ocean biogeochemistry, and notably the oxygen cycle in the modern ocean69 and in the geological past73,90,91,92,93.