Introduction

Understanding the causes of natural variations in atmospheric CO2 concentration requires an appreciation of the temporal distribution of carbon among the major reservoirs involved in carbon exchange on geological timescales. On long timescales, CO2 is added to the atmosphere via volcanic degassing, metamorphic decarbonation, and oxidation of organic matter, and is removed via chemical weathering and organic carbon burial1,2. The primary mechanisms that set this delicate balance of carbon inputs and outputs, which ultimately regulates long-term climate, are strongly debated2,3,4,5. Silicate weathering and organic carbon burial are proposed as key feedbacks in setting and/or mediating Neogene atmospheric CO2 levels3,6,7. Beyond understanding the geochemical carbon cycle, constraining the outcome of anthropogenic carbon addition requires an understanding of not only climate and ecological sensitivity to rapid carbon evasion8,9, but also the nature of different carbon cycle feedbacks. This latter requirement arises, because carbon cycle feedbacks can either enhance atmospheric CO2 trends or play a crucial part in their recovery. Numerous studies of Pleistocene climate records identified several carbon cycle feedback processes that operate or are enhanced when the Earth is in colder than modern climate modes10,11. In contrast, the feedback processes that operated during past warm climates are less well-studied and hence remain poorly understood, despite these being arguably more relevant for our warmer future. Carbon-cycle feedbacks are therefore currently not well represented in Earth System models and highlight a gap in our ability to predict future climate change12,13.

An ideal time interval to explore the controls on long-term atmospheric CO2 and pinpoint such carbon cycle feedbacks is the Miocene Climatic Optimum (MCO)—a period of sustained global warmth ~3 °C warmer than modern with reduced continental ice volume occurring ca. 14.7–17.0 Myrs ago and sitting within the Monterey carbon isotope excursion (MCIE)14,15. The MCIE was a prolonged (~3.5 Myrs) ~1.0 ‰ positive carbon isotope excursion (δ13C) of the global oceans, documented in planktic and benthic foraminifera records centered on ~15 Ma16,17 (Fig. 1), contemporaneous with basin-wide deposition of organic-rich deposits, as evident in the California Monterey Formation and elsewhere in the circum-Pacific Ocean16,18,19. The MCIE is usually described as comprising six carbon isotope maxima (CM events), which are paced by the ~400 kyr long-eccentricity cycle17,20 (Fig. 1b, c). Following the MCO, the middle Miocene Climate Transition (MMCT)ushered in a cooler and apparently more stable icehouse climate mode in a series of cooling or ice growth steps, as evidenced by benthic oxygen isotope records17,21(Fig. 1a) and a transition from a wet-based to dry-based Antarctic ice sheet22. This climate transition was associated with an overall decrease in the carbon isotopic composition of the ocean, although individual cooling steps have been associated with carbon maxima events, suggestive of the operation of positive carbon-cycle climate feedbacks23.

Fig. 1: Isotopic and trace metal records across the middle Miocene.
figure 1

Previously published benthic foraminiferal (a) C. mundulus δ18O14,21 and (b) δ13C records from site ODP 761; (c) ODP 761 T. trilobus δ13C record from this study; (d) ODP 761 (Indian), DSDP 588 (Pacific), and ODP 926 (Atlantic) T. trilobus B/Ca µmol/mol, and (e) T. trilobus δ11B-pH estimates29,35 from site 761 for one δ11Bsw scenario35,66 (G17) with Monte Carlo uncertainty estimates (66% CI); (f) SST estimates derived from T. trilobus Mg/Ca at site 761. Yellow vertical bars highlight the timing of the Miocene Climatic Optimum (MCO) and Middle Miocene Climate Transition (MMCT). Carbon isotope maxima (CM) events 1–6 are highlighted. Monterey carbon isotope excursion duration is highlighted by the above arrow.

The classic interpretation, known as the Monterey Hypothesis19, proposed that long-term cooling increased the latitudinal gradient, invigorating upwelling, which led to increased organic carbon burial rates, lowering atmospheric CO2. This reinforced the initial cooling, while also increasing mean ocean δ13C composition and acting as a positive carbon cycle feedback. Subsequent hypotheses using strontium and osmium isotope records, invoked increased silicate weathering and enhanced nutrient delivery as a mechanism for CO2 draw down via enhanced organic carbon deposition and mid-Miocene cooling6. Others proposed the addition of volcanic CO2 emissions as a mechanism to maintain the MCO warmth24,25 and thereby stimulating high rates of chemical weathering alongside enhanced organic carbon burial. With the availability of the first continuous Miocene atmospheric CO2 records26 and identification of a minimum in CO2 concentrations during the MCO, it was hypothesized that the positive δ13C anomaly were caused by increased rates of organic carbon burial19,26 with the warmth of the MCO sustained by its particular oceanographic configuration and low continental ice volume rather than by greenhouse gas forcing26. However, these interpretations relied on limited and, in some cases, unreliable, available indices of middle Miocene atmospheric CO2, silicate/organic carbon weathering, and climate variability6,19,24,25,26,27. In particular, the early alkenone-based CO2 reconstructions were biased by diagenetic imprints on the original paleotemperature records28 that were used to calculate CO2.

Recent efforts to reconstruct CO2 using the boron isotope and alkenone-based paleobarometers, with more robust organic proxy-based temperature estimates, instead suggest that the warmth of the MCO was sustained by relatively high CO2 (470–630 ppm) and the MMCT was associated with falling CO2 concentrations29,30,31,32,33. It is now thought that CO2 emissions associated with the Columbia River Flood Basalts (CRB) may have been at least in part responsible for the elevated CO2 during the climatic optimum24,29,31,32. Furthermore, geochemical box modeling suggests that increased concentrations of CO2 may also account for part of the MCIE itself, by increasing the photosynthetic isotopic fraction of carbon to form organic matter more enriched in 12C on a global scale25. However, attributing the MCIE solely to a volcanic carbon source does not explain the ~400 kyr pacing of the carbon maxima events or the organic-rich sedimentation typified by the Monterey Formation and modeling produces a ~0.4‰ change in mean whole ocean δ13C, whereas the MCIE itself has a far greater amplitude (~1‰)25.

The association of the warmth of the MCO with elevated CO2 levels, organic-rich sedimentation, and positive carbon isotope excursions is reminiscent of the Cretaceous carbon cycle perturbations marked by widespread deposition of organic-rich sediments. This is thought to represent a key negative feedback in Earth’s ice-free greenhouse climate mode, helping to draw-down atmospheric CO2, but is one which came at a cost to Cretaceous marine ecosystems—widespread oceanic anoxia33. Although similar in many respects, the Miocene intervals of organic-rich sedimentation were never associated with widespread anoxia34. Records of ocean carbon chemistry are therefore required to understand the nature of the carbon cycle feedbacks in this warmer-than-present icehouse world.

Although the development of the boron isotope (δ11B) proxy has contributed to the recognition that climate and CO2 were coupled across the MCO and the MMCT23,29,30,35, pH is just one aspect of the ocean carbonate system. Directly reconstructing relative changes in dissolved inorganic carbon (DIC) content of seawater would allow us to more firmly quantify changes in oceanic carbon storage, which would enable the carbon cycle feedback processes associated with the MCO to be investigated.

Here we use planktic foraminiferal boron to calcium (B/Ca) ratios along with complementary published δ11B records to reconstruct shifts in the concentration of [B(OH)4/DIC] and [B(OH)4/HCO3] and relative changes in surface ocean DIC across the early Miocene, MCO, and MMCT. We present a moderately high resolution (one sample every ~23 kyr) multi-proxy record of B/Ca, Mg/Ca, and carbon isotope (δ13Cp) variability in the planktic foraminifer Trilobatus trilobus spanning 16.5 to 11.5 Ma from Ocean Drilling Program (ODP) Site 761 located in the tropical eastern Indian Ocean. We supplement this record with lower resolution T. trilobus B/Ca reconstructions from the Pacific (Deep Sea Drilling Program (DSDP) Site 588; ODP Site 872) and Atlantic Ocean basins (ODP Site 926). We then apply ad hoc calibrations, based on modern culture and core-top datasets, along with existing records of pH to approximate shifts in surface ocean \(\left[ {{\mathrm{B(OH)}}_4^ - /{\mathrm{DIC}}} \right]\) and [B(OH)4/HCO3], and ultimately DIC (see Methods). We then pair these records of ocean carbon storage with planktic Mg/Ca and δ13Cp records and previously published benthic δ18O records to provide a novel perspective on the MCIE and drivers of the carbon cycle across the middle Miocene. Our records show a positive relationship between DIC and the broad Monterey δ13C excursion, but a negative relationship between DIC and δ13C for the individual CM events. This suggests that the MCIE is not the sum of the carbon maxima events as originally supposed by the Monterey Hypothesis and instead that these two features of the carbon isotope record reflect different carbon cycle processes. We suggest that the broad MCIE represents a negative feedback process akin to Cretaceous Oceanic Anoxic Events (OAEs), which operated in a warmer-than-present icehouse world while the positive carbon isotope excursions (CM events) within the Monterey carbon excursion likely reflect a positive carbon cycle feedback process associated with orbital scale variability in marine productivity.

Results

B/Ca record of T. trilobus from the mid-Miocene

ODP Site 761 T. trilobus B/Ca, Mg/Ca, and δ13Cp records span a 5-million-year time slice beginning from the onset of the MCO, through its main body and into the MMCT (Fig. 1). At the onset of the MCIE, the planktic δ13Cp record from Site 761 shows an overall increase, following the benthic δ13C record, with average MCO δ13Cp values of 2.5 ‰ decreasing to ~2.0 ‰ by 12 Ma (Fig. 1). Superimposed upon this background, long-term trend are the globally recognized CM events14 associated with glacial events and positive shifts in δ18Ob during the middle Miocene.

During the onset of the MCO, T. trilobus B/Ca ratios decreases from ~70 µmol/mol to a minimum of 59 ± 1 µmol/mol ( ± 2 SE; 14.7 to 16.0 Ma). This decrease in B/Ca ratios is associated with increasing tropical Indian Ocean temperatures by ~1.8 °C (Fig. 1). Over the course of the middle Miocene, T. trilobus B/Ca increases across the MMCT from the minimum values of the MCO to a maximum B/Ca value of 76 ± 1 µmol/mol (±2 SE; 13.5–11.5 Ma), an average increase of 17 µmol/mol. This long-term increase in B/Ca records occurs in tandem with a small long-term temperature decrease of ~0.7 °C. The long-term increase in B/Ca values through the MMCT bears a remarkable similarity to the benthic δ18O (δ18Ob) records and is expressed in two steps with the first B/Ca increase occurring at 14.7 Ma, with B/Ca values reaching 66 ± 1 (±2 SE) μmol/mol, before the final increase at 13 Ma to ~76 μmol/mol.

Superimposed on this long-term B/Ca increase are numerous short-term variations in B/Ca with an amplitude between 10 and 15 µmol/mol. Earlier CM events show smaller scale peaks in B/Ca corresponding to the peak foraminiferal δ13C values. Peak positive planktic and benthic δ13C values during CM6 correspond to an increase B/Ca by ~10 μmol/mol. Further, the minimal long-term cooling in sea surface temperatures (SST) is punctuated by a sharp 2.4 °C cooling from 14.0 to 13.8 Ma, associated with the positive δ18Ob excursion and CM-6 event (Figs. 1 and 2). The overall SST amplitude at ODP 761 between warm, deglaciated and cool, glaciated conditions ranges from 3 °C to 2 °C, with a minimum temperature around 13.2 Ma.

Fig. 2: Isotopic and trace metal records across CM6.
figure 2

a Planktic foraminifera T. trilobus B/Ca from Indian (ODP 761), Pacific (DSDP 588), and Atlantic Oceans (ODP 926 alongside (b) planktic foraminiferal carbon isotope record from ODP site 761. c Benthic foraminiferal carbon isotopes from ODP site 76114,21. d Mg/Ca-SST estimates from ODP site 761. e Benthic foraminiferal oxygen isotopes from ODP site 76114,21.

We also present lower resolution T. trilobus B/Ca records from the Pacific (DSDP Site 588; ODP Site 872) and Atlantic (ODP Site 926) Ocean (Fig. 1 and Supplementary Figs. 1 and 2). These records show similarities in their general trends and absolute values to ODP Site 761, despite their different site locations, paleo water depths, and sediment burial depths. DSDP Site 588 B/Ca increases from 62 ± 4 µmol/mol (±2 SE) during the MCO to 80 ± 1 µmol/mol ( ± 2 SE), whereas ODP 926 increases from 65 ± 1 µmol/mol (±2 SE) during the MCO to 81 ± 2 µmol/mol ( ± 2 SE). This corresponds to a long-term increase in B/Ca of 18 and 16 µmol/mol in the Pacific and Atlantic, respectively, similar to Indian Ocean. ODP 872 B/Ca values decrease from the early Miocene into the MCO similar to the orbital-scale record from ODP 761, although our ODP 872 record is of lower resolution (Supplementary Fig. 2).

Overall, T. trilobus B/Ca records document a substantial decrease at the start of the MCIE, followed by an increase from the MCO through the MMCT in the Indian, Pacific, and Atlantic oceans. The similar trend and magnitude revealed by these B/Ca records, despite their diverse oceanographic settings suggests a global driver as the primary cause. In addition, the Site 761, Site 926, and Site 872 B/Ca records track the planktic δ11B-derived pH (i.e., MCO: low pH, low B/Ca) in the same cores supporting the idea that B/Ca is responding to global changes in surface ocean carbonate chemistry across the middle Miocene29,30,35 (Fig. 1 and Supplementary Fig. 2). Furthermore, the similar timing of change in the δ13Cp and B/Ca records across the middle Miocene suggests that the MCIE was associated with a change in the surface ocean carbon cycle and carbonate chemistry (Fig. 1).

The moderately resolved Indian Ocean T. trilobus B/Ca record shows a ~10 μmol/mol increase in B/Ca associated with CM6 (Fig. 2 and Supplementary Fig. 3). As the resolution of the Pacific and Atlantic B/Ca is too low to resolve short-term variations in surface ocean carbon storage, we look to the T. trilobus B/Ca record across CM6 from the Blue Clay Formation of Malta at Ras il-Pellegrin23,36 for comparison. This record shows an ~17 μmol/mol increase in B/Ca with a corresponding increase in planktic δ11B-derived pH. Together, the Indian Ocean and Malta records indicate surface ocean carbon reservoir changes associated with the short-term positive δ13Cp variations (i.e., CM events) in the global ocean (Supplementary Fig. 3).

Reconstructing Ocean Carbon Storage across the middle Miocene

Using surface ocean pH estimates derived from boron isotope reconstructions35 and ad hoc B/Ca-B(OH)4/DIC and B/Ca-B(OH)4/HCO3 calibrations (see Methods), we estimate changes in DIC and B(OH)4/HCO3 across the middle Miocene (Figs. 3 and 4).

Fig. 3: Sensitivity of pH vs. [B(OH)4-/HCO3]sw relationship to DIC and ALK.
figure 3

pH estimates derived from boron isotope reconstructions35 and ad hoc B/Ca-B(OH)4/DIC and B/Ca-[B(OH)4/HCO3]sw calibrations (see Methods) are plotted for a range of scenarios with changes in the dissolved inorganic carbon (DIC) and alkalinity (ALK) inventory. Blue and red line indicate scenarios with constant DIC and ALK, respectively. Purple and green lines indicate changes with a ALK:DIC ratio of 1:1 and 2:1. We plot one δ11Bsw35,66 (G17) scenario to highlight the overall relationships present and provide full carbonate system estimates in the Methods. Monte Carlo uncertainty estimates (95% CI) are plotted for pH35 and [B(OH)4/HCO3]sw estimates.

Fig. 4: Paleoclimate records across the middle Miocene.
figure 4

a Planktic foraminifera T. trilobus B/Ca ratios from Indian (ODP 761), Pacific (DSDP 588), and Atlantic Oceans (ODP 926) alongside (b) benthic foraminiferal δ13C14,21 records from ODP site 761. c Estimates of changes in global ice volume as derived from δ18Osw at ODP site 761. Gray circles are uncorrected BWT estimates, whereas dark gray circles are corrected21. d Eustatic sea-level change from the Marion Plateau40. Intervals of peak warmth in Antarctica as identified from ANDRILL15 are highlighted with pink bars (PW3-5). e Boron isotope derived atmospheric CO2 record35using three δ11Bsw scenarios66,89,92 (LO2; RH13; G17). f mid-Miocene DIC estimates with Monte Carlo uncertainty envelope using ad hoc B/Ca-B(OH)4/DIC calibration. We plot one δ11Bsw scenario35,66 (G17) to highlight overall changes in DIC (DIC) changes and provide full range of DIC estimates in the Methods. Miocene DIC values were subtracted from average MCO DIC levels to calculate DIC. g Estimated eruption rate of the Columbia River Basalt93. Top bars specify marine (yellow) and land based (green) organic carbon deposition events evident across the Miocene19,94,95.

First, this approach allows us to identify the likely drivers of ocean carbonate system change during the middle Miocene through examining the relationship between Miocene surface ocean B(OH)4/HCO3 and pH (Fig. 3). Changes in the ocean’s biological and carbonate pumps lead to alterations of the vertical gradients of DIC and alkalinity (ALK), but not their overall inventory. Whereas long-term shifts in carbon fluxes can alter ocean inventories of DIC and ALK with the stabilizing response of carbonate compensation processes (i.e., CaCO3 dissolution and formation). Specifically, if the change in the carbonate system reconstructed was due to modification of the DIC or ALK gradient within the Miocene ocean (i.e., constant DIC or ALK), then the Miocene surface water pH vs. B(OH)4/HCO3 would display a relationship with a relatively shallow slope (red and blue lines in Fig. 3). Similarly, if a change in these variables was due to whole-scale change in ALK:DIC with a 2:1 ratio due to carbonate compensation in response to changing the water column gradients, a shallow slope would result (green line in Fig. 3). DIC removal or addition due to the long-term geological processes (e.g., CO2 invasion, chemical weathering, and/or Corg burial/oxidation) where carbonate compensation operates to maintain a stable CaCO3 saturation state can be approximated by the purple line on Fig. 3 where ALK and DIC change with a 1:1 ratio. This results from the addition/removal of 1 mol of CO2 and the full carbonate compensation response that adds/removes ALK and DIC with a 2:1 ratio to maintain a steady state of CaCO3 burial37. The Miocene B(OH)4/DIC and pH data show a relationship which closely follows the 1:1 (ALK:DIC) line, suggesting the changes in carbonate system we reconstruct (i.e. pH and B(OH)4/HCO3) were mainly due to the addition/removal of DIC through slow geological processes, coupled with the expected CaCO3 system response (Fig. 3).

This approach allows a semi-quantification of ocean DIC through the Miocene. However, given the inherent uncertainties involved, particularly in the [B]sw and the sensitivity of the B/Ca-[B(OH)4/DIC]sw, only relative changes in Miocene DIC are considered here (Fig. 4). Using this semi-quantitative approach, we are able to show that the low B/Ca and B(OH)4/HCO3 values of the MCO suggest higher DIC levels in this warm interval, and that DIC had declined by ~200 to 400 μmol/kg by the end of the MMCT glaciation (see Methods and Fig. 4). Superimposed on this overall change in DIC on millions of years timescales are higher resolution changes in DIC that were likely orbitally paced. For example, paired δ11B and B/Ca records from Malta provide strong evidence for a DIC decline across the CM6 event (Supplementary Fig. 3). Although the absolute values remain poorly constrained we estimate this decline to be on the order of ~300 μmol/kg.

Discussion

Our planktic foraminiferal δ13Cp records show the slow rise and peak during the MCO, directly followed by a recovery during the MMCT, typical of the globally recognized MCIE. Given the original Monterey Hypothesis, one would expect the broad increase in δ13C associated with the onset of the MCIE to be associated with a decrease in seawater DIC. This is in contrast to our boron-based DIC reconstructions, which show a broad low in B/Ca values, low pH, and hence high DIC throughout the duration of the MCO (Figs. 3 and 4, and see Methods), whereas Mg/Ca-SSTs estimates indicate a ~1.8 °C warming (Fig. 1). These findings are therefore not consistent with an interpretation of increasing organic carbon burial in a cooling climate at the onset of the MCIE. Instead, it is consistent with a volcanic driver initiating the warmth of the MCO, which pushed the climate system into a less glaciated state with higher sea level. The majority of volcanic emissions from the CRB erupted between 16.7 and 15.9 Ma, which is approximately coincident with the onset of the MCO31 (Fig. 4). Several studies propose that CRB volcanism supplied the necessary carbon emissions to trigger and sustain higher than modern atmospheric CO2 concentrations (~470–630 ppm), and warm MCO climate sufficient to inhibit reglaciation of Antarctica24,25,29,30,31,32,33,34,35,36,37,38,39 (Fig. 4).

A rise of atmospheric CO2 and positive δ13C excursion (MCIE), despite the release of presumably isotopically depleted volcanic carbon (mantle value of −5‰) may at face value seem implausible. However, the higher CO2 levels could act to increase ocean δ13C by changing the fractionation factor associated with photosynthetic fixation of carbon, offsetting the depleted volcanic CO2 contribution25. Alternatively, a relatively slow emission rate or a carbon isotope composition heavier than the traditional −5‰ mantle value, possibly through metamorphic degassing of country rock, could more easily reconcile a positive MCIE with increasing CO2 levels. Separation of the δ18Ob signal into temperature and δ18Osw shows a decrease in seawater δ18O (δ18Osw)21 heading into the MCO suggesting a rise in sea level (Fig. 4). This is supported by numerous records of ice volume and sea-level variability based on sedimentological and oxygen isotope evidence from the early to mid-Miocene, which indicates that the Antarctic ice sheet was highly variable during this interval with equivalent to 30–80% loss of modern Antarctic ice sheet15,21,30,40,41,42. It is therefore conceivable that this glacio-eustatic expansion of shelf area would lead to an enhanced capacity for continental organic carbon burial. As modern continental shelves produce three times the mean primary productivity of the ocean43 this can lead to organic-rich low-oxygen sediment conditions ideal for increased preservation of organic carbon during the MCO. Furthermore, shelf facies are typically associated with relatively high sedimentation rates, which also favors organic matter preservation and burial44. Despite this proposed propensity for enhanced organic carbon burial, elevated surface ocean DIC levels over millions of years could be sustained by the supply of carbon emissions via Columbia River Basalt eruptions. This mechanism would not be expected to cause an increase in productivity or organic carbon burial in open ocean sites, consistent with evidence from deep-water palaeoproductivity proxies45. The MCO sea level rise likely also stimulated enhanced shallow water carbonate deposition in suitable locations, representing an additional source of CO2 to the atmosphere46. The net effect on the carbon cycle is difficult to quantify as it would depend on the balance between the volcanic and chemical weathering fluxes, and the proportion of organic carbon to carbonate burial fluxes. Nevertheless, our records provide the first evidence to suggest that the MCIE was associated with high oceanic DIC that sustained enhanced organic carbon burial over millions of years (Fig. 4f). Overall, this implies a role for CO2 induced glacio-eustatic sea level rise in modulating the organic carbon subcycle and producing the largest carbon isotope excursion of the Neogene. The associated burial of marine organic carbon therefore represents a negative feedback, mitigating further atmospheric CO2 rise. We note that prior studies suggest that marine organic carbon accumulation rates within the Monterey Formation (eastern Pacific) are too low to account for the entire magnitude of the MCIE and additional light carbon enriched depocenters were proposed to reconcile the full magnitude of the carbon isotope excursion45,46,47. Here we look at this question with an ocean-based lens and note that additional studies are needed to explore the relative contribution of shelf vs. continental organic carbon sequestration (e.g., enhanced peat formation) in driving the MCIE.

Overall, our results suggest that during the MCO there is an increase in organic carbon burial as seen for the Early Eocene Climatic Optimum48 and Cretaceous OAEs33,34,49. This played a role in mitigating the volcanically driven rise in CO2, acting as a negative feedback. However, in contrast to the MCIE, the hallmark of OAE events are widespread anoxic and euxinic conditions. The greenhouse state during the Cretaceous had virtually no continental ice and widespread warmth50,51. Furthermore, more sluggish ocean circulation due to the lack of polar ice and lower ocean oxygen content due to overall warmer ocean temperatures may have led to an increased capacity to produce an OAE in the early Cenozoic/Late Cretaceous. We conclude therefore that the MCIE was likely a failed OAE due to the presence of an Antarctic ice sheet and relatively well-ventilated ocean in an icehouse time interval.

Superimposed on the long-term broad MCIE and its recovery are short-lived (~400 kyr) carbon maxima events (CM1–6) associated with positive excursions in δ18Ob, glaciation/cooling and the long-term component of eccentricity (~400 kyr), with CM6 having the largest amplitude16,17,18,20,52. δ18Ob leads benthic δ13Cb in the eccentricity band in Miocene isotopic records indicating a lag of carbon cycle over climate variability19. Our planktic B/Ca record at ODP Site 761 exhibits similar short-term variability as δ18Ob and δ13Cb that is superimposed on the broad decrease in B/Ca, although the resolution of the record limits our ability to examine its possible orbital signature (Figs. 1, 2, and 4). An increase in B/Ca associated with CM-6 evident from ODP Site 761 and Malta indicates a DIC decline (Supplementary Fig. 3). This suggests that although the broad positive MCIE is associated with relatively high DIC, the individual carbon maxima events are associated with higher B/Ca (lower DIC), implying that different processes are responsible for the two features of the carbon isotope record.

Previous work has shown that CM-6 is associated with an invigoration in ocean circulation and increase in marine export productivity45,53,54 and pCO2 draw-down23, following a major period of Antarctic ice sheet expansion55. This is a similar situation to the glaciation and positive CIE which occurs at the Oligocene–Miocene boundary (~23 Ma)56. As with the MCO, these δ13C carbon isotope excursions are associated with enhanced carbon burial and potentially enhanced marine productivity.

Modeling studies have linked the origin of the long-period eccentricity pattern in the δ13C record to insolation driven changes in a marine biosphere productivity57 and increased weathering and nutrient input and marine productivity58. Data-model comparisons show that the magnitude of δ13C change and the timing and frequency of CM events depends on a balance between orbitally driven changes in ice volume and hence sea-level and shelf deposition/weathering and marine productivity53. Overall, this suggests that CM events during the MCIE were linked to high nutrient delivery sustained by high CO2-driven weathering alongside insolation-driven marine productivity events. However, further work is needed to fully consider the global nutrient budget and role of insolation in setting the stage for Miocene CM events.

Our records therefore indicate that increased marine organic carbon burial exhibited contrasting time-scale dependent behavior during the middle Miocene: on short-term orbital timescales it acted as positive climate feedback during the orbitally paced CM cooling events, and on million year timescales it acted as a negative climate feedback, mitigating volcanic CO2 emissions during the MCO.

Overall, our records suggest planktic B/Ca is a powerful proxy to explore the oceanic carbon cycle during large carbon cycle perturbation events. Our new DIC record represents the dynamic interaction between volcanic driven CO2 input, organic carbon burial, and weathering. The broad δ13C increase of the MCIE was associated with high surface ocean DIC concentration and the sustained warmth of the MCO. We propose that elevated carbon emissions associated with Columbia River Basalt Flood volcanism induced global warming and sea level rise, which enhanced organic carbon burial via the drowning of continental shelves. We suggest that the broad MCIE itself is therefore a consequence of this key negative feedback process. The icehouse climate state of the middle Miocene, with its continuous ocean ventilation likely prevented the regime of enhanced organic carbon preservation from developing into a full-blown OAE as seen in the greenhouse world of the Cretaceous. During the CM events of the Miocene, climatic cooling episodes triggered corresponding periods of increased marine productivity, which likely reduced surface ocean DIC and atmospheric CO2 and acted as a positive carbon cycle feedback process consistent with the original Monterey Hypothesis. The nature and expression of carbon cycle feedbacks in response to C-addition are therefore state-dependent, and such feedbacks may be critical when exploring future carbon emission scenarios in Earth System models.

Methods

Age model and study sites

Samples from DSDP Site 588 and ODP sites 761 Hole B, 926 Holes A/B, and 872 Hole C were used to reconstruct SST and \(\left[ {{\mathrm{B(OH)}}_4^ - /{\mathrm{DIC}}} \right]_{{\mathrm{sw}}}\) across the early to middle Miocene (22.0–11.5 Ma; Supplementary Fig. 1). Planktic foraminifera Mg/Ca, B/Ca, and δ13Cp records from ODP site 761B were sampled at an average temporal resolution of 23 kyr. In contrast, our other sites were sampled at lower resolution for B/Ca: ODP 926 (~200 kyr) and DSDP 588 (~100 kyr) and ODP 872 (~1 Myr). ODP 761 (16°44.23’S, 115°32.10’E, water depth of 2179 m) is situated in the Indian Ocean on the Wombat Plateau. Site ODP 926 (3°43.148’N, 42°54.507’E, water depth 3598 m) in the eastern equatorial Atlantic Ocean on Ceara Rise. Site DSDP 588 A (26°06.7’ S, 161°13.6’ E, water depth 1548 m) in the Southwest Pacific Ocean on Lord Howe Rise. ODP Site 872 (10°N, 162°E, 1287 m water depth) is in the tropical north Pacific gyre on the sedimentary caps of flat-topped seamounts.

The age model for Site 761B is based on a fourth-order polynomial fit through the biostratigraphic and isotopic datums14,35. For ODP site 926, we use a polynomial fit through nannofossil and planktic foraminifer biostratigraphic datums adapted to the CK95 timescale35. The published age model for DSDP 58859 is based on compiled biostratigraphy, magnetostratigraphy, isotope stratigraphy calibrated to CK95. Ages for ODP Site 872 were calculated by linear interpolation between reliable biostratigraphic datums60 from the early to mid-Miocene adapted to the CK95 timescale35. Core sites29,30,35,61 used in this study were selected because currently they are located in regions where modern surface waters are close to equilibrium with respect to atmospheric CO2 (Supplementary Fig. 1). We use δ11B-pH estimates35 from ODP sites 761, 872, and 926 in addition to B/Ca presented in this study to estimate surface ocean \(\left[ {{\mathrm{B(OH)}}_4^ - /{\mathrm{DIC}}} \right]_{{\mathrm{sw}}}\) and\(\left[ {{\mathrm{B(OH)}}_4^ - /{\mathrm{HCO}}_3 - } \right]_{{\mathrm{sw}}}\).

Planktic foraminiferal taxonomy and ecology

We generated trace element and isotope records using T. trilobus, a multi-chambered, photosymbiont bearing species, which is predominantly a mixed layer dweller calcifying at 0–50 m and is abundant in subtropical and tropical oceans. T. trilobus is a morphospecies of Trilobatus sacculifer62,63 present throughout the Neogene and Quaternary periods. T. trilobus has been used extensively in previous Miocene climate and boron isotope studies29,23,30,35 due to its narrow habitat range, well-defined δ11B-pH calibration, calcification close to equilibrium conditions (i.e., minimal vital effects), and abundance29,35,63,64. Further, previous studies demonstrated that temperatures derived from T. sacculifer are most suitable for estimating annual mean SST65 in tropical waters (between 20° N/S) within ±1 °C. Visual inspection of preservation (Supplementary Fig. 4) of T. trilobus from ODP site 761B show moderately good preservation with no visible signs of infilling or dissolution. Visual inspection of foraminiferal specimens from ODP site 926 shows evidence of a recrystallized nature, however the Mg/Ca and boron isotope signature has remained intact29,35,66. Visual inspection of foraminiferal specimens from ODP site 872 show a glassy nature indicative of well-preserved specimens and exceptionally good carbonate preservation supported by previous work using scanning electron microscopic (SEM) images60,67. Visual inspection of foraminiferal specimens from DSDP 588 have been shown to be pristine in nature with no evidence of dissolution and have been previously used in oxygen isotope reconstructions52.

Trace metal and isotopic analysis

Between 30–40 tests of the planktic foraminifera T. trilobus were picked from the 300–355 μm size fraction from ODP site 761, 926, and DSDP 588. ODP 872 samples were prepared for boron isotope analysis and an aliquot of the foraminiferal sample was used to collect trace element data (see ref. 35). Picked specimens were weighed, gently crushed between glass plates and homogenized for chemical cleaning and geochemical analysis. In samples where T. trilobus abundance was low (~30 samples), fewer specimens (10–20 individuals) were analyzed. After sample homogenization, an aliquot of the foraminiferal sample was used to collect trace element and δ13Cp data. It is noteworthy that only δ13Cp from ODP site 761 are presented here.

Test fragments for Mg/Ca and B/Ca analyses were cleaned using a protocol to remove surficial clays, adhered organic matter, and potential secondary carbonate overgrowths68. Clay removal consists of repeated rinses with low boron Milli-Q water and methanol. The oxidative cleaning step consists of a sodium hydroxide and hydrogen peroxide solution to remove any present organic matter. The reductive cleaning step was not included in the cleaning process as this step is unnecessary for B/Ca as the cleaning does not alter the B/Ca ratio of foraminifera or calcite bound boron69. Lastly, a weak acid leach using 0.001 M hydrochloric acid was conducted to remove any ions re-absorbed during the cleaning process on the test surface. Following the clay removal and oxidative steps, samples were examined under a binocular microscope and visible non-carbonate particles were removed using a fine paintbrush. Cleaned, treated samples were dissolved in trace metal pure 0.065 M HNO3 and diluted with trace metal pure 0.5 M HNO3 to a final volume of 350 μl, to achieve a target calcium concentration of 4 mM. Trace element analysis (B, Mg, Al, Ca, Mn, Fe) for ODP 761, ODP 926, and DSDP 588 was carried out at Cardiff University on a Thermo Scientific Element XR Sector Field Inductively Coupled Plasma Mass Spectrometer (SF-ICP-MS), whereas samples from ODP site 872 were analyzed at the University of Southampton, using a Thermo Scientific Element 2 XR SF-ICP-MS. In both instances, calcium concentrations of bracketing standards were matched to foraminifera samples to reduce matrix effects70,71.

To monitor the possibility of clay contamination Mg/Ca and B/Ca data were rejected if Al/Ca exceeded 80 μmol/mol for all sites. Additional cleaning effectiveness was supported by observing no significant correlation between Mg/Ca, and Fe/Ca and Mn/Ca (Supplementary Figs. 5-6). There is no correlation between B/Ca and Fe/Ca at any of the investigated sites. There is however a positive correlation between Mn/Ca and B/Ca at ODP site 926 (R2 = 0.4) and 761 (R2 = 0.6), and a slight negative correlation (R2 = 0.3) between Mn/Ca and B/Ca at ODP site 872 (Supplementary Fig. 6). However, at site 761, correlations across 500 kyr windows from 16.5 to 11.5 Ma show no significant correlation (R2 < 0.3; p > 0.05) between these variables in 8 out of the 10 intervals (Supplementary Table 1). We therefore propose here that the Mn/Ca record reflects another aspect of the system (e.g., oxygenation, productivity), which would be correlated on geological timescales with B/Ca as in the Miocene72. Long-term precision at Cardiff University and University of Southampton was determined by analyzing an independent consistency standard during each run for 1 year. Reported values are 0.5% and 4% (r.s.d.) for Mg/Ca and B/Ca, respectively, for analyses at Cardiff University and 2% and 4% (r.s.d.) for Mg/Ca and B/Ca, respectively, for analyses at University of Southampton.

Stable carbon isotope ratios were measured at Cardiff University on a Finnigan MAT 252 micro-mass spectrometer Kiel III Carbonate Device when sample weights were <100 μg and measured on a Delta isotope ratio mass spectrometer when samples were greater than 100 ug. Long-term precision based on replicate measurements of a laboratory standard (NBS 19) are 0.08‰ for δ13C.

Mg/Ca-paleotemperature calculation and constraining Mg/Ca seawater changes

Planktic T. trilobus is the morphotype of the modern T. sacculifer, we consider it appropriate to calculate mean annual SSTs from T. trilobus Mg/Ca data using the T. sacculifer calibration65:

$$\frac{{Mg}}{{Ca}} = 0.347\ \left( {\!\pm \! 0.011} \right)e^{0.090\left( {\! \pm \! 0.003} \right)T}$$
(1)

Variations in Miocene foraminiferal Mg/Ca are driven by changes in temperature; however, on longer timescales, both changes in temperature and in seawater Mg/Ca must be considered. Recently, it has been shown that a power function73 best describes this relationship.

$$\frac{{Mg}}{{Ca}}_{foram} = \left[ {\frac{{\frac{{Mg}}{{Ca}}_{sw}\left( t \right)}}{{\frac{{Mg}}{{Ca}}_{sw}\left( 0 \right)}}} \right]^CBe^{AT}$$
(2)

where Mg/Casw(t) and Mg/Casw (0) are seawater Mg/Ca ratios for the Miocene and present, and A, B, and C are constants (A = exponential, B = pre-exponential, C = power constant), respectively.

Mg and Ca have relatively long residence times (~13 Myr and ~1.1 Myr, respectively) in the ocean. The modern day seawater Mg/Ca value is 5.2 mmol/mol and low-resolution fluid inclusions data74 show a rise across the Neogene to the modern values. Here we use the fluid inclusion value of 3.43 mmol/mol in the paleotemperature calculation. We use the T. sacculifer calibration65 (A = 0.09, B = 0.347) for T. trilobus and apply a power constant75 of 0.41 for T. Sacculifer based on the available data for this species76. T. trilobus SST estimates can be calculated from the following equation:

$$\frac{{Mg}}{{Ca}}_{foram} = 0.293\,e^{0.090T}$$
(3)

Using a range of Mg/Ca seawater estimates, calculated SST during the MCO period in Site 761 ranges from 1.4 °C warmer to uncorrected Mg/Ca-SST estimates being cooler by 0.5 °C than the modern (modern mean annual temperature is ~27.7 °C).

In addition to changes in seawater Mg/Ca, Mg incorporation in foraminifera has shown a species-specific dependency to changes in carbonate chemistry. A recent review77 proposed a pH correction was necessary to correct Mg/Ca records when estimating SST for some species. This study showed that Mg/Ca ratios in T. sacculifer are insensitive to changes in pH. As Mg/Ca ratios are derived from T. trilobus shells, the morphotype of T. sacculifer, we apply no pH correction in this study.

Foraminiferal preservation

Several lines of evidence suggest that dissolution does not significantly affect the Mg/Ca or B/Ca values at Site 761 and suggest our data represent a climate signal. First, ODP Site 761 is situated well above the modern lysocline, above the critical 20 μmol/kg ΔCO32−, in a relatively shallow burial depth during the middle Miocene (<50 m) (GLODAP78). Average shell weight of T. trilobus, from the 300–355 μm size fraction, does not covary with the Mg/Ca or B/Ca record, supporting our argument that these values are not biased (Supplementary Fig. 7).

Despite the reasonable appearance of foraminifera, all tests appear frosty or opaque in contrast to exceptionally well-preserved translucent test shells from hemipelagic muds79. However, this preservation state is typical of most deep-sea carbonates and is caused by micro-recrystallization of calcite. Large-scale recrystallization is not evident in SEM images for ODP site 761 (Supplementary Fig. 4), suggesting that diagenesis did not drive prominent shifts in δ13C and trace elements. In addition, Sr/Ca ratios from ODP site 761 show high values (~1.1–1.2 mmol/mol; Supplementary Table 2), which are consistent across much of the record, suggesting that diagenesis did not have a major influence on Mg/Ca or B/Ca variations. Nonetheless, Mg/Ca values decrease by a negligible amount with initial diagenetic alteration79, thus temporal changes in Mg/Ca are less likely to be affected. Further work80 has shown a decrease in planktic B/Ca in recrystallized relative to well-preserved foraminifera; however, additional work is needed to isolate the primary diagenetic signal. For the reasons outlined above, we believe that diagenesis had a minimal effect on our records and we therefore interpret the geochemical records in terms of paleoceanographic conditions.

Estimation of surface ocean [B(OH)4 /HCO3 ]sw and [B(OH)4 -/DIC]sw from B/Ca

Similar to boron isotopes, the boron content (expressed as B/Ca) of planktic foraminifera has been suggested to be a proxy for ocean carbonate chemistry69,81. Boron exists in seawater primarily as two species borate (B(OH)4 and boric acid (B(OH)3), and the relative concentration of each boron species and their δ11B composition varies with pH in seawater. Boron isotopic evidence and other arguments suggest that B(OH)4 is the species predominantly incorporated into the foraminiferal calcite lattice82. B/Ca in planktic foraminifera is controlled by a combination of ocean carbonate chemistry parameters (pH, B(OH)4 \({\mathrm{HCO}}_3^ -\) and DIC) as demonstrated through empirical culture experiments and field studies71,81,83. Culturing efforts aimed to disentangle this co-varying carbonate system parameters show that B/Ca is governed principally by the ratio of seawater B(OH)4 to DIC or \({\mathrm{HCO}}_3^ -\)81,84. Application of the B/Ca proxy to reconstruct shifts in the concentration of [B(OH)4/DIC] in seawater across major climate transitions has shed light on past carbon cycle perturbations84,85 (e.g., Paleoceane-Eocene Thermal Maximum), where large shifts in the ocean carbonate system occurred. Similarly, in our study we aim to examine the carbon cycle perturbations of the early to middle Miocene, a time interval whereby a large carbonate system shifts likely occurred, suggesting some of the complexities that others have discussed with regard to the glacial–interglacial cycles may not be significant.

Culture experiments with living T. sacculifer demonstrate a positive relationship between B/Ca and oceanic carbonate system parameters, where B/Ca increases at higher pH and lower DIC and HCO3 concentrations (Supplementary Fig. 8)81. These experiments also provide calibrations that relate B/Ca to both [B(OH)4/HCO3]sw and [B(OH)4/DIC]sw; however, they argue for a better fit to the latter parameter. Recent culture work further explored the mechanistic understanding of B uptake, indicating that planktic B/Ca, Orbulina universa specifically, is driven by HCO3 in support of [B(OH)4/HCO3]sw rather than [B(OH)4/DIC]sw controlling B/Ca83. Here we present [B(OH)4/HCO3]sw to evaluate the main drivers of carbonate system change through the MCIE. Further, we consider both B/Ca carbonate chemistry sensitivities (e.g., [B(OH)4/HCO3]sw and [B(OH)4/DIC]sw) to estimate surface ocean DIC.

When we compile all available B/Ca-[B(OH)4/DIC]sw and [B(OH)4/HCO3]sw T. sacculifer calibration data, an offset is observed between culture81 and core-top86 datasets, where the core-top data sits lower and with a slightly steeper slope predicted by culture experiments, although this slope difference may largely be a function of the different ranges spanned by the datasets, as the available core-top data are limited by the range in surface ocean carbonate chemistry (e.g., [B(OH)4/DIC]sw) that span the depth habitat of wild-type T. sacculifer compared with conditions manipulated in culture work81. We account for differences in test size87 and normalize values to a common salinity (S = 35)81, but an offset and slope difference between the datasets remain (Supplementary Fig. 8). Here we suggest the offset could be related to the fact core-top foraminiferal specimens, during their life cycle, sink to water with lower pH and higher DIC/ HCO3 in contrast to the cultured specimens. This is supported by the lower core-top B/Ca values relative to culture. Furthermore, the core-top study used T. sacculifer specimens with the final sac-life chamber, which have lower B/Ca relative to non-sac chambers81, in contrast to the cultured specimens.

To address the B/Ca offset between datasets we retain the slope predicted in culture and normalize the culture to core-top values to generate new ad hoc B/Ca to [B(OH)4/DIC]sw and [B(OH)4/HCO3]sw calibrations (B/Ca = 33 + 773*[B(OH)4/DIC]sw;B/Ca = 39 + 561*[B(OH)4/HCO3]sw; Supplementary Fig. 8). Application of these ad hoc calibrations to ODP 761 B/Ca reconstructions requires additional consideration of past changes in seawater boron ([Bsw]) on calibration sensitivity. We can assume a constant seawater composition over the studied interval as it is considerably shorter than the oceanic residence time of B (10–20 million years)88. A positive relationship between seawater B concentration ([Bsw]) and B content in planktic foraminiferal calcite (O. universa) is observed in culture experiments81,83,84 (Supplementary Fig. 9). Following Mg/Ca convention and the observation that Mg/Ca sensitivity changes as a function of seawater Mg/Ca ratio, we adjust the calibration sensitivity to changes in B/Casw, to account for the influence of variable Mg/Casw on Mg/Ca thermometry70,75 by assuming a linear scaling:

$$\frac{B}{{C{\mathrm{{a}}}}}{\mathrm{{foram}}} = \frac{{\frac{B}{{C{\mathrm{{a}}}}}sw\,t = t}}{{\frac{B}{{C{\mathrm{{a}}}}}sw\,t = 0}}x\frac{{B({\mathrm{{OH}}})_4^ - }}{{{\mathrm{{HCO}}}_3^ - }} \ast {m} + {b}$$
(4)
$$\frac{B}{{C{\mathrm{{a}}}}}{\mathrm{{foram}}} = \frac{{\frac{B}{{C{\mathrm{{a}}}}}sw\,t = t}}{{\frac{B}{{C{\mathrm{{a}}}}}sw\,t = 0}}x\frac{{B({\mathrm{{OH}}})_4^ - }}{{{\mathrm{{DIC}}}}} \ast {m} + {b}$$
(5)

where B/Casw (t=t) for the age of the sample and B/Casw (t=0) is modern B/Casw and m and b are the slope and y-intercept for each calibration from. Miocene estimates of Bsw and Casw are derived from boron isotope modeling and fluid inclusion estimates, respectively74,88.

Calculation of DIC and HCO3from the B/Ca-[B(OH)4/DIC]sw and [B(OH)4/HCO3]sw calibrations requires an independent estimate of B(OH)4 across the Miocene, and here we use δ11B-pH estimates from ODP 761, 872, and 92635 to calculate B(OH)4. B(OH)4 is related to pH as described here:

$${\mathrm{{pH}}} = - \log [H^ + ]$$
(6)
$$B({\mathrm{{OH}}})_4^ - = B_{{\mathrm{{TOT}}}}/\left( {1 + \frac{{[H^ + ]}}{{K_{\mathrm{{B}}}^ \ast }}} \right)$$
(7)

where \(K_{\mathrm{{B}}}^ \ast\) is the stoichiometric equilibrium constant for boric acid89 at time equivalent T, S, and P, and Mg/Casw (t = X Ma). pH estimates used were derived from the three δ11Bsw scenarios35 (G17, RH13, L02). The pH derived from each scenario have a similar structure but differ in absolute pH values. This should minimally impact the DIC trend across the Miocene, however, we use pH estimates from all scenarios to estimate the likely DIC (Supplementary Fig. 10). Surface ocean pH estimates from ODP 872 are higher by 0.1 pH units compared with site 926 and 761 in the early Miocene (16.5–17.0 Ma) (Supplementary Fig. 2). To account for this offset, we adjust ODP site 872 by 0.1 pH units and use these adjusted pH values in the DIC estimates. SST estimates are derived from Mg/Ca in complementary samples35 and modern salinity is assumed for the entire duration of the Miocene record. We normalize planktic B/Ca values to a common salinity, assuming modern salinity values for each site (S = 35)81. Bsw varied in time according to marine boron isotope budget88 in the same manner as above for B/Casw estimates. Using the B/Ca-[B(OH)4/HCO3]sw calibration, we estimate DIC from HCO3 using the following equation:

$${\mathrm{{DIC}}} = \frac{{[{\mathrm{{H}}}^ + ][{\mathrm{{HCO}}}_3^ - ]}}{{K_1}} + \left[ {{\mathrm{{HCO}}}_3^ - } \right] + \frac{{K_2[{\mathrm{{HCO}}}_3^ - ]}}{{[{\mathrm{{H}}}^ + ]}}$$
(8)

where K1 and K2 are the carbonic acid dissociation constants89.

The principal uncertainties in calculating [DIC]sw and [HCO3] in this way were judged to be uncertainties in the B/Ca calibration and in the pH used to calculate [B(OH)4]. Here we used Monte Carlo method to propagate these uncertainties into our final estimates of DIC and HCO3. In order to highlight the long-term trends in DIC and HCO3, we then fit smoothing splines through the data with the degree of smoothing determined by generalized cross validation. Again, the uncertainty in these smoothed trends was determined using a Monte Carlo approach and a consideration of the error bars of each individual estimate.

DIC estimates derived using the range of pH scenarios (G17 v RH13 v L02) show a similar trend across the middle Miocene (Supplementary Fig. 11). Miocene estimates of surface ocean DIC derived from ODP site 761 using both B/Ca-[B(OH)4/DIC]sw and [B(OH)4/HCO3]sw calibrations, and pH estimates from all scenarios show higher DIC during the MCO (14.0–17.0 Ma; Supplementary Fig. 11 and Supplementary Tables 3 and 4; 68% CI; n = 15). Early Miocene (17.0–22.0 Ma) estimates are lower than the MCO (Supplementary Table 3; 68% CI; n = 7) increasing towards the onset of the MCO. MCO DIC estimates decline following the MMCT. MMCT DIC levels (13.5–11.0 Ma) are lower than the MCO (Supplementary Table 3; 68% Confidence Interval; n = 12). Further, the overall decline in DIC from the MCO to post-MMCT of ~220 to 370 μmol/kg (B/Ca-[B(OH)4/DIC]sw) and 280 to 430 μmol/kg (B/Ca-[B(OH)4/HCO3]sw) (Supplementary Table 5) is similar to change derived from a modeling study25, although absolute values are offset. Absolute Miocene DIC estimates here fall within the general estimates based on a range of approaches from modeling to geochemical reconstructions (Supplementary Fig. 12)35,90,91.

Regardless of the approach and location, our findings suggest the MCO DIC levels were higher than the early Miocene and late middle Miocene and DIC levels decreased across the MMCT by as much as 200–400 μmol/kg. Given the inherent uncertainties in these calculations, particularly in the [B]sw and the sensitivity of the B/Ca-[B(OH)4/DIC]sw and [B(OH)4/HCO3]sw, the absolute values should be taken with caution. The trends recognized, however, because they are largely determined by B/Ca can be considered robust.

Although the Miocene B/Ca record from ODP site 761 presented in this study encompasses the short-term variations in δ13C (i.e., CM events), due to the lack of complimentary δ11B-pH records at the same resolution, we are unable to estimate relative DIC on these timescales, based on records from this study (Supplementary Fig. 3). Here we look to the T. trilobus B/Ca record from Malta which includes both B/Ca and δ11B-pH estimates across CM6 and shows a similar increase in B/Ca (Supplementary Fig. 3). Using the same approach as above and equations 4-7, we estimate relative change in DIC across CM6. A 17 μmol/mol increase in planktic B/Ca and corresponding increase of 0.08 in pH (Supplementary Fig. 3) across the CM6 event equates to a lowering of DIC by ~300 μmol/kg, although in absolute terms this remains poorly constrained at present. We do note that to fully quantify and assess surface DIC changes on these timescales, a suite of orbitally resolved records of the carbonate system (i.e., B/Ca and δ11B) across mid-Miocene carbon isotope excursions are needed.