Capturing the global signature of surface ocean acidification during the Palaeocene–Eocene Thermal Maximum

Abstract

Geologically abrupt carbon perturbations such as the Palaeocene–Eocene Thermal Maximum (PETM, approx. 56 Ma) are the closest geological points of comparison to current anthropogenic carbon emissions. Associated with the rapid carbon release during this event are profound environmental changes in the oceans including warming, deoxygenation and acidification. To evaluate the global extent of surface ocean acidification during the PETM, we present a compilation of new and published surface ocean carbonate chemistry and pH reconstructions from various palaeoceanographic settings. We use boron to calcium ratios (B/Ca) and boron isotopes (δ¹¹B) in surface- and thermocline-dwelling planktonic foraminifera to reconstruct ocean carbonate chemistry and pH. Our records exhibit a B/Ca reduction of 30–40% and a δ¹¹B decline of 1.0–1.2‰ coeval with the carbon isotope excursion. The tight coupling between boron proxies and carbon isotope records is consistent with the interpretation that oceanic absorption of the carbon released at the onset of the PETM resulted in widespread surface ocean acidification. The remarkable similarity among records from different ocean regions suggests that the degree of ocean carbonate change was globally near uniform. We attribute the global extent of surface ocean acidification to elevated atmospheric carbon dioxide levels during the main phase of the PETM.

This article is part of a discussion meeting issue ‘Hyperthermals: rapid and extreme global warming in our geological past’.

1. Introduction

A hallmark of the Palaeocene–Eocene Thermal Maximum (PETM, approx. 56 Ma) is a negative carbon isotope excursion (CIE) of 3–5‰, signifying a large injection of isotopically light carbon into the atmosphere–ocean reservoirs [1–4]. Although the exact source of carbon is a matter of debate, the event provides a test for our understanding of the ocean's response to the rapid invasion of carbon and heat. The uptake of carbon dioxide (CO₂) in seawater should result in a decrease in ocean pH and calcium carbonate (CaCO₃) saturation state (Ω), commonly termed ocean acidification [5]. In the modern ocean, these environmental changes (i.e. warming, deoxygenation and ocean acidification) are connected by a single driver: an increase in atmospheric pCO₂ [6]. Current and predicted anthropogenic changes in ocean biogeochemistry inform us that the coastal regions and high latitudes are particularly vulnerable. Changes in continental run-off and nutrient fluxes along with seasonal stratification could amplify pH changes in coastal waters. Southern Ocean waters are inherently low in Ω, making them more susceptible to undersaturation with respect to CaCO₃ during ocean acidification events. The geological record can potentially provide insight into the extent to which such environments and their biotas can be impacted by ocean acidification.

The PETM represents a natural experiment of the global ocean response to multiple, covarying environmental stressors. The degree of ocean warming is now well constrained to be 5–8°C [2,7], whereas deep-ocean acidification is largely inferred from widespread dissolution of seafloor carbonate sediments [4,8,9]. Evidence of surface ocean acidification, however, has only recently emerged from planktonic foraminiferal boron isotope (δ¹¹B) and boron/calcium (B/Ca) proxy records [10–12]. Model studies provide ancillary support for whole ocean acidification [13–15]. Marine ecosystems experienced dynamic changes during the PETM but the relative impact of warming, deoxygenation and acidification on organisms remains unclear [16,17]. With improved spatial coverage of boron proxy records, we aim to better constrain the global magnitude of surface ocean acidification.

Here we quantify ocean acidification and its spatial extent within the upper ocean (surface to thermocline) during the PETM, synthesizing new and published records of planktonic foraminifera δ¹¹B and B/Ca proxies. The study sites span pelagic and coastal environments, and low and high latitudes in the Atlantic and Pacific Ocean basins (figure 1). We generated planktonic foraminifera B/Ca records at high-latitude Southern Ocean (Atlantic Sector, Ocean Drilling Program (ODP) sites 689 and 690) and continental shelf site ODP Ancora, as well as δ¹¹B data from thermocline-dwelling species at continental shelf sites (ODP sites Bass River and Millville). These new records complement published records from the pelagic, low-latitude Pacific ODP site 1209 [10], continental shelf ODP site Bass River [11] located along the eastern coast of North America and Deep Sea Drilling Project (DSDP) site 401 in the Northeast Atlantic [12]. Seawater properties (temperature, salinity, nutrients and oxygen concentration) and foraminiferal preservation state vary between the sites, and we exploit these differences to determine the global extent of surface ocean acidification across a diverse range of palaeoceanographic and depositional settings.

Figure 1.

Palaeocene–Eocene (approx. 56 Ma) palaeogeographic reconstruction and core study sites. Pelagic ODP sites include 1209, 1263, 865, 689, 690 and DSDP 401. The map is modified from an Ocean Drilling Stratigraphic Network (ODSN)-generated map using www.odsn.de. New Jersey coastal plain ODP sites include Bass River (BR), Ancora (AN) and Millville (MV).

2. Methods and material

(a). Site locations and depositional environments

High-latitude pelagic sites 689 (64°31.009′ S, 03°05.996′ E) and 690 (65°9.629′ S, 1°12.296′ E) are located on Maud Rise in the Weddell Sea (figure 1). The PETM palaeodepths of site 689 and site 690, composed of calcareous ooze, are estimated to be 1100 m and 1900 m, respectively [18]. A detailed study has shown that the CIE and its ensuing recovery are preserved within the site 690 PETM section, whereas a coring gap appears to have omitted the upper half of the CIE and the recovery at site 689 [18]. Acarinina soldadoensis (250–355 µm size fraction), A. praepentacamerata (180–250 µm size fraction) and Subbotina spp. (250–355 µm size fraction) were picked from washed sediment samples taken at 1–5 cm resolution at these sites. Because of foraminiferal species assemblage changes during the PETM [19] at sites 689 and 690, A. praepentacamerata was measured in the Palaeocene and A. soldadoensis was analysed within the CIE interval.

Coastal sites Bass River (BR; 39°36.70′ N, 74°26.20′ W), Millville (39°24.2778′ N, 75°05.3332′ W) and Ancora (39°41.5329′ N, 74°50.9410′ W) are located in New Jersey in the mid-Atlantic coastal plain [20–22] (figure 1). Ancora was closest to the palaeo-coastline, followed by Millville, and the most distal and down-dip site is Bass River, with PETM palaeodepths estimated at 50–150 m, indicative of middle to outer shelf environments [23,24]. These sections are composed of uppermost Palaeocene age micaceous silts and glauconitic sands (Vincentown Formation) transitioning to lower Eocene kaolinitic clay (Marlboro Formation) which yields well-preserved foraminifera tests with glassy textures. This is in contrast to the pelagic sites used in this study, which display a ‘frosty’ texture indicative of some degree of post-depositional recrystallization.

At Ancora, trace element to calcium ratios were analysed on multi-specimen monogeneric samples of surface-dwelling photosymbiont-bearing Acarinina and Morozovella spp., and thermocline-dwelling Subbotina spp. from the 250–355 µm size fraction to obtain approximately 200–400 µg of CaCO₃ material. The gap in trace element data at the CIE onset (BR: 357.42–357.09 m, AN: 171.59–170.98 m) is due to an absence of foraminifera attributed to carbonate dissolution. At Ancora, resolution is low for the upper Palaeocene, due to low foraminifera abundances. Closely spaced samples were often combined to achieve sufficient sample material for boron isotopes (δ¹¹B) at sites Bass River and Millville. δ¹¹B measurements were carried out on the few species with adequate abundance, Subbotina triangularis (Palaeocene) and S. roesnaesensis (Eocene), using the 180–300 µm size fraction to obtain 1–4 mg of CaCO₃ material.

(b). Age models

All the sites used in this study have published planktonic foraminifera and bulk carbonate δ¹³C records. However, for the purpose of refining age models between sites, we correlate the bulk carbonate records, which benefit from a higher temporal resolution (figure 2d). Although the bulk carbonate records differ in magnitude between sites, their temporal δ¹³C trends parallel the planktonic foraminiferal records. We provide more details on age constraints for New Jersey and Southern Ocean sites generated in this study in the electronic supplementary material.

Figure 2.

Comparison of planktonic foraminiferal B/Ca (a–c) and bulk carbonate δ¹³C records (d). ODP 1209 (solid black circles) [10], 689 (open light green squares), 690 (solid green squares) and DSDP 401 (solid orange squares) [12] are pelagic sites (a–c). ODP sites Bass River (solid blue triangles) [11] and Ancora (light blue triangles) are continental shelf sites (a–c). Bulk carbonate δ¹³C records (d) are labelled in the same colour scheme as B/Ca records (a–c). Sites 1209 [9], 401 [12], 690 [25], 689 [18], Bass River [26] and Ancora [27] are plotted versus age in kiloyears relative to the onset of the CIE. A detailed explanation of the age model construction for each site can be found in the Methods section. Analytical reproducibility of B/Ca based on repeated analysis of in-house laboratory consistency standards is 6% and 7% (2 s.d.) at Rutgers University and University of California Santa Cruz, respectively [10,11]. Vienna Pee Dee Belemnite (VPDB) carbon isotope standard. (Online version in colour.)

(c). Analytical methodology

Trace element analyses were conducted on a Thermo Element XR inductively coupled plasma mass spectrometer at Rutgers University (sites Bass River and Ancora) and the University of California Santa Cruz (ODP 1209, 689 and 690) (figure 2a–c). Boron isotopes were measured on a Thermo Triton Thermal Ionization multi-collector mass spectrometer at Lamont-Doherty Earth Observatory (ODP 1209, Bass River and Millville) (figure 3a). Foraminiferal cleaning and instrument protocol carried out for records generated in this study (trace elements at sites Ancora, 689 and 690; boron isotopes at sites Millville and Bass River) follow previous work and detailed methodologies are provided therein [10,11].

Figure 3.

Relative changes in ocean pH derived from planktonic foraminiferal boron isotope (δ¹¹B) measurements (a). Surface dweller M. velascoensis was used to generate the 1209 (black circles), 1263 (red crosses), 865 (green triangles) [10] and M. subbotinae 401 (orange squares) [12] δ¹¹B records (a). Deep dweller Subbotina spp. were used in δ¹¹B reconstructions at Bass River (blue diamonds) and Millville (purple diamonds) (a). All records were normalized to a pre-event ocean pH = 7.75 by adjusting the intercept of the assumed δ¹¹B_foram to calibration following a similar approach to previous studies [10,12] (b,c). Ocean pH anomalies (ΔpH) were estimated by computing a relative δ¹¹B change compared with an average of pre-CIE values (c). A constant seawater δ¹¹B value of 38.9 ± 0.4‰ for the duration of the PETM was assumed [12] and no vital effect was considered to estimate ΔpH (c). Age is denoted in kiloyears relative to the CIE. Error envelopes on ocean pH estimates include the analytical uncertainty of 2 s.e. of repeat sample analyses (n = 2–3), and conservative uncertainty estimates for temperature (±2°C) and salinity (±2 units) (b). (Online version in colour.)

3. Results

New B/Ca data from sites 689, 690 and Ancora (electronic supplementary material, tables S1 and S2) are presented with published B/Ca records from sites 1209, Bass River and 401 (figure 2a–c) [10–12]. At site 690, using A. praepentacamerata and A. soldadoensis to represent the mixed layer, B/Ca declines by 30% (from 53 to 37 µmol mol⁻¹) within the CIE, then recovers to pre-CIE values (figure 2b). Subbotina B/Ca, representing the thermocline, is slightly lower throughout but shows a similar 25% decline (from 45 to 34 µmol mol⁻¹) and recovery (figure 2c). At site 689, the B/Ca patterns recorded by A. praepentacamerata, A. soldadoensis and Subbotina are similar to those of 690, though slightly offset (figure 2b,c). At Ancora, average B/Ca ratios for Subbotina spp. decline by 25% (from 51 to 38 µmol mol⁻¹) within the CIE (figure 2c). Although no data were generated for the Palaeocene, within the CIE Morozovella spp. and Acarinina spp. values average 50 µmol mol⁻¹ and increase to 70 and 60 µmol mol⁻¹, respectively, in the recovery interval (figure 2a,b).

δ¹¹B data from the Bass River and Millville sites are plotted along with published δ¹¹B data from sites 1209, 1263, 865 [10] and 401 [12] (figure 3a; electronic supplementary material, table S3). Below the CIE onset at site Bass River, average δ¹¹B values for Subbotina spp. are 13.6‰ and decrease to average minimum δ¹¹B values within the CIE core of 12.5‰ and 12.7‰ at sites Bass River and Millville, respectively (figure 3a). To test a potential size bias, δ¹¹B data were measured in multiple size fractions but no significant offset was found over the size range covered in this study (electronic supplementary material, table S3). This is consistent with data from modern deep dwellers, the ecological parallel of Subbotina spp., which show only a negligible δ¹¹B-test size relationship [28].

4. Discussion

(a). Global trends in boron proxy records during the Palaeocene–Eocene Thermal Maximum

At all locations, and in each of the three foraminiferal taxa measured, the records follow a common pattern (table 1). Coincident with the onset of the CIE, B/Ca and δ¹¹B decrease, indicating acidification of upper ocean waters. In all the boron proxy records that span the duration of the event, this sharp decline is followed by a plateau of consistently depressed B/Ca and δ¹¹B values that persist for approximately as long as the CIE itself (approx. 70 kyr in our age models). Coeval planktonic foraminiferal δ¹³C values closely track B/Ca records through the CIE onset but δ¹³C appears to recover faster than B/Ca and δ¹¹B (figures 2 and 3). This could be indicative of decoupling between the key ocean carbonate system parameters (i.e. pH, dissolved inorganic carbon (DIC) and ) driving each individual proxy [10]. This divergence pattern can be explained by shifting the relative contributions of organic carbon burial versus silicate weathering to carbon sequestration during the CIE recovery. While both processes elevate pH, organic carbon burial is more efficient in raising δ¹³C because silicate weathering only accelerates dilution of this excess light carbon while organic carbon burial preferentially removes light ¹²C from exogenic reservoirs. Indeed, geochemical model studies invoke an intensified organic carbon burial feedback to facilitate carbon sequestration during the CIE recovery [12,29]. Finally, near the start of the CIE recovery, B/Ca and δ¹¹B begin to increase coincident with a recovery and temporary overshoot of %CaCO₃ in sediments globally [30], including an overshoot in the calcite compensation depth (CCD) [31] and/or a build-up of alkalinity in deep waters [32], suggesting a restoration of carbonate chemistry. Taken together, our foraminiferal boron proxy records support rapid PETM surface ocean acidification from pelagic to coastal, and tropical to polar settings, lasting tens of thousands of years.

Table 1.

Average planktonic foraminiferal B/Ca and δ¹¹B data during the PETM.

		average B/Ca µmol mol−1a
core site	species	below CIE onset	CIE core	post-CIE	ΔB/Cab
1209	M. velascoensis	70.4	44.5	62.7	25.9
1209	A. soldadoensis	71.0	43.9	53.5	27.1
401	M. subbotinae	51.9	51.4	52.6	0.5
689	Acarinina spp.	49.3	29.8	49.2	19.5
690	Acarinina spp.	53.4	37.0	52.7	16.4
Bass River	Acarinina spp.	65.0	40.0	48.0	25.0
Bass River	Morozovella spp.		42.0	45.0
Ancora	Acarinina spp.		38.0	65.0
Ancora	Morozovella spp.		51.0	58.0
1209	Subbotina spp.	46.0	31.3	37.5	14.8
689	Subbotina spp.	37.3	24.9	39.2	12.4
690	Subbotina spp.	45.3	33.8	42.6	11.5
Bass River	Subbotina spp.	57.0	40.0	42.0	17.0
Ancora	Subbotina spp.	51.0	38.0	49.0	13.0
		average δ11B (‰)a			Δδ11B (‰)b
1209	M. velascoensis	15.5	14.7	15.3	0.9
401	M. subbotinae	14.7	14.0	14.3	0.7
Bass River	Subbotina spp.	13.5	12.5		1.1
Millville	Subbotina spp.		12.7

^aThe onset of the CIE is identified as the initial decrease in δ¹³C, core CIE is an interval of sustained low δ¹³C values (less than 70 kyr) and the post-CIE interval represents the recovery of δ¹³C to values below the CIE onset (70–200 kyr) (figure 2d).

^bDifference between average values prior to the CIE onset and the CIE core interval.

A partial exception to the above trends occurs with the geochemical proxy records based on the surface dweller M. subbotinae at DSDP site 401 [12]. At site 401, one extreme minimum δ¹³C point and one extreme minimum δ¹¹B point occur approximately 25 kyr after the CIE onset and are directly followed by the CIE recovery phase (figure 3a). Unlike all the other sites, B/Ca at site 401 shows only a very slight decrease at the CIE onset (figure 2a and table 1). Furthermore, the foraminifera δ¹³C records at site 401 exhibit a relatively small CIE magnitude compared with regional and global records [33]. Incomplete recovery by rotary drilling (RCB in International Ocean Discovery Program (IODP) terminology) at site 401 probably resulted in a truncated or disturbed PETM interval [34], which helps explain the discrepancies with other sites [35]. During the recovery of core 401–14R, which contains the PETM, drilling advanced 9.5 m but only 7.2 m of sediment was recovered, indicating the loss of over 2 m of sediment within this interval [34]. Given the abrupt lithological change at the Palaeocene–Eocene boundary, it is likely that at least some of this sediment loss occurred at the base of the PETM, truncating the CIE and associated geochemical records. A similar problem occurred during attempts to recover the clay layer at ODP 1263 by rotary coring [36]. We include the site 401 boron proxy records in the global compilation while recognizing that the CIE is only partially represented (figures 2a and 3).

(b). Quantifying changes in seawater carbonate chemistry

Following the approach of Penman et al. [10], we use the boron isotope data from mixed-layer and thermocline-dwelling foraminifera to quantify the decrease in surface seawater pH during the onset of the PETM. We focus on the relative magnitude of acidification rather than estimating absolute pH, and use the more established δ¹¹B proxy to corroborate B/Ca trends. Knowledge of past seawater δ¹¹B (δ¹¹B_sw) as well as a species-specific calibration are required to derive absolute ocean pH using foraminiferal δ¹¹B data. An estimation of temperature, salinity and major seawater ion concentration is also required to calculate the dissociation constant of boric acid (K*_B). The experimentally determined boron isotopic fractionation factor of 1.0272 was used [37]. We applied an estimated δ¹¹B_sw value of 38.9 ± 0.4‰ [12], which is consistent with δ¹¹B_sw estimates over the Eocene [38]. To account for potential laboratory offsets, pre-event δ¹¹B values were normalized to pH = 7.75, following the approach of previous PETM boron proxy model studies [10,12]. K*_B was calculated with the MyAMI model assuming Eocene seawater (approx. 50 Ma) [Mg] = 30 mM and [Ca] = 20 mM [39]. Surface and thermocline temperatures are estimated using Mg/Ca palaeothermometry with a modern multi-species calibration [40]. In addition to temperature, variation in past seawater Mg/Ca composition affects the Mg distribution coefficient and possibly Mg/Ca proxy sensitivity to temperature. A nonlinear correction scheme to account for the influence of variable past Mg/Ca_sw developed for the modern planktonic foraminifer Trilobatus sacculifer was applied to compute ocean temperatures [41]. We used previously generated Mg/Ca records in M. velascoensis at site 1209 [10,42], M. subbotinae at site 401 [12] and Subbotina spp. at site Bass River [11]. Pre- and post-event salinity values of 37 (ODP 1209) and 35 (sites Bass River, Millville and ODP 401) were used and are similar to climate model regional estimates [43]. A salinity increase of 1.5 units was previously determined at site 1209 [42], whereas a 2 unit decrease is estimated for coastal New Jersey sites and a 1.5 unit decrease at site 401 (further discussion on estimation of salinity anomalies in §4c Contribution of regional salinity shifts on B/Ca).

Culture calibration studies with modern planktonic foraminifera document species-specific δ¹¹B_foram to relationships that are variably offset from seawater borate values [44]. These isotopic offsets are thought to be caused by the physiological processes of symbiont photosynthesis, respiration and calcification that act to modify pH within the calcifying microenvironment around foraminifera and, subsequently, alter the δ¹¹B recorded in foraminiferal calcite [44]. To estimate ocean pH from the extinct species used in this study, we need to approximate their relationships. The first option is to assume equilibrium between δ¹¹B of foraminiferal calcite (δ¹¹B_c) and δ¹¹B of seawater borate , which is equivalent to assuming the absence of any vital effects . Alternatively, habitat ecology (i.e. surface dwelling and photosymbiont bearing) determined by previous studies on the basis of carbon and oxygen isotope offsets can be used as a means to assign a modern calibration relationship to an analogous extinct species. However, ascribing such a modern analogue to an extinct species assumes modern vital effects were similar during the Palaeogene, which is currently poorly understood.

Several lines of evidence argue for diminished vital effects of Eocene age planktonic foraminifera relative to modern species. Reduced δ¹³C test size relationships in muricate PETM species (acarininids and morozovellids) compared with modern species (Globigerinoides) suggest overall lower photosymbiont activity rates [45–47]. If we assume that lower photosynthetic rates translate to reduced vital effects, we would expect foraminiferal δ¹¹B to be closer to seawater values. Indeed, δ¹¹B values analysed for a suite of planktonic foraminifera occupying the upper water column show a narrower isotopic range in the Eocene relative to modern day, implying reduced vital effects [38,48]. This is further corroborated by similar δ¹¹B values obtained in time-equivalent Eocene samples of S. roesnaesensis and benthic foraminifera Cibicidoides [38], genera which are interpreted to have minimal δ¹¹B-related vital effects in modern ocean sediments [44]. When modern ecologically equivalent calibration equations are applied to Eocene surface (morozovellids) and thermocline (subbotinids) species, an unrealistic reverse ocean pH depth profile is obtained [38]. This suggests that modern δ¹¹B to calibrations overestimate Eocene planktonic foraminifer vital effects. For these reasons, we consider no vital effect with a slope of 1.0 to be a closer approximation of surface and thermocline ocean pH change (ΔpH) at the CIE relative to pre-event values (figure 3c). We calculate an average pH decrease for M. velascoensis of 0.22 units (extrema 0.12–0.30) at site 1209, 0.15 units (0.04–0.34) for M. subbotinae at site 401 and for Subbotina spp. 0.22 units (0.23–0.52) in New Jersey (figure 3c). By contrast, if we apply the pH sensitivity of the modern photosymbiont-bearing T. sacculifer (slope = 0.82) (refitted in [28,49]) at site 1209, a larger average pH decrease for M. velascoensis of 0.26 units (extrema 0.14–0.37) is estimated (electronic supplementary material, figure S1c). Whether we include or exclude any vital effects, our multi-site pH estimates produce a uniform degree of acidification in the coastal ocean compared with pelagic sites (figure 3c; electronic supplementary material, figure S1c) in agreement with model predictions, lending support to our observations [12]. This spatial similarity is perhaps unsurprising, as an increase in atmospheric pCO₂ should be absorbed by the surface ocean everywhere, leading to a relatively uniform pH decrease globally. A decline in the degree of surface ocean acidification from the mixed layer to the thermocline is also consistent with model predictions of attenuated acidification with depth [14]. Regardless of the calibration uncertainties, both δ¹¹B records are consistent in providing evidence for upper ocean acidification in support of our interpretation of complementary B/Ca records.

Culturing experiments established a carbonate system control on B/Ca in planktonic foraminiferal calcite [50–53]. A first-order approximation is to simply relate modern B/Ca proxy systematics to interpret PETM records. However, when modern B/Ca calibrations are applied to B/Ca data from extinct PETM taxa, unrealistic negative values are estimated [10]. In general, the PETM B/Ca values are significantly lower, extending beyond the modern calibration range. We instead consider new empirical culture calibrations that relate B/Ca to of modern photosymbiont-bearing species Orbulina universa under simulated Palaeocene seawater chemistry (high [Ca], low [Mg] and low total boron concentration [B_T]) [53] (figure 4). We translate B/Ca excursion anomalies (average pre-CIE baseline compared with core CIE values) to estimate as in previous studies [10,11] and apply ‘Palaeocene’-derived calibration sensitivity [53] (figure 4). We only consider B/Ca excursion anomalies of mixed-layer (acarininids) and thermocline (subbotinids) dwellers, for which there are pre-CIE values recorded at all locations used in this study. We estimate a decrease for M. velascoensis of 0.023 at site 1209 and essentially no change for M. subbotinae at site 401 (figure 4a). For Acarinina spp., all sites (NJ, 1209, 689 and 690) show an average decrease of 0.022 (extrema 0.021–0.024) (figure 4b) and Subbotina spp. show an average decrease of 0.015 (extrema 0.013–0.018) (figure 4c). Using the new Palaeocene calibration yields anomalies are closer to previous estimates based on calibrations conducted in modern seawater [10] but still larger than predicted carbonate and borate chemistry estimated by model outputs [53]. This model–data difference could be a result of the Palaeocene calibration being too insensitive to fully resolve the B/Ca anomalies across the PETM [53]. Another explanation is that a larger carbon release (greater than 3000 Pg C) is required to better approximate the B/Ca-based estimates of carbonate chemistry change. Both B/Ca and δ¹¹B proxies estimate a shift in surface ocean carbonate chemistry towards lower pH and saturation state. Within the calibration uncertainties, both boron-based proxies support a uniform pattern of ocean acidification during the PETM. Coincident biogeochemical changes, such as salinity, dissolution and symbiont loss, could influence B records and we evaluate these factors in addition to seawater carbonate chemistry.

Figure 4.

Ocean carbonate chemistry reconstructions derived from planktonic foraminiferal B/Ca records (a–c). B/Ca records are presented for ODP sites 1209 (black circles) [10], 689 (open light green squares), 690 (solid green squares), Bass River (solid blue triangles) [11], Ancora (light blue triangles) and DSDP 401 (orange squares) [12]. Seawater anomalies are relative to average pre-CIE values and are based on sensitivity determined in modern photosymbiont-bearing planktonic foraminifera species O. universa under simulated Palaeocene seawater chemistry [53]. (Online version in colour.)

(c). Contribution of regional salinity shifts on B/Ca

Culture experiments reveal a small but significant influence of salinity on boron incorporation and, consequently, B/Ca ratios in planktonic foraminifera [50]. Regional shifts in salinity during the PETM could augment or dampen B/Ca anomalies and, if unaccounted for, bias our interpretation of the B/Ca component related to carbonate chemistry. We can disentangle the influences of salinity and carbonate chemistry by normalizing B/Ca records to a common salinity value. However, accurate estimates of salinity using the paired Mg/Ca-δ¹⁸O method [42] require accurate estimates of temperature and an appropriate δ¹⁸O_sw–salinity regional relationship, which are difficult to constrain. As an alternative, we chose to reconstruct salinity anomalies relative to pre-CIE baseline values [42]. Morozovellid and acarininid Mg/Ca and δ¹⁸O datasets at sites Bass River [11,26], Ancora [27], 401 [12], 1209 [10,42] and 690 [2,54] were used to estimate relative salinity anomalies. For each site, an average pre-CIE δ¹⁸O and Mg/Ca value was subtracted from each data point to estimate the relative changes across the event, Δδ¹⁸O and ΔMg/Ca. Ocean temperature anomalies were estimated using a multi-species calibration, as inter-species variability in Mg/Ca temperature sensitivity is quite small in modern subtropical planktonic foraminifera [40]. In this way, we can quantify uncertainty of non-thermal controls on Mg/Ca that are necessary to consider for absolute temperature reconstructions. The temperature component of Δδ¹⁸O was calculated with a Δδ¹⁸O/ΔT sensitivity of 0.21‰/°C, derived for modern symbiont-bearing O. universa [55]. Orbulina universa was selected as its temperature sensitivity is a mid-range value for modern planktonic foraminifer species and B/Ca salinity sensitivity is also known [50]. The salinity anomaly estimate assumes the discrepancy between δ¹⁸O and Mg/Ca temperature anomalies is solely from regional surface water δ¹⁸O_sw. The residual non-temperature δ¹⁸O component corresponds to Δδ¹⁸O_sw and was converted to ΔSSS (surface seawater salinity) by using the same slope as the modern regional Δδ¹⁸O_sw to ΔSSS relationship. The residual Δδ¹⁸O was converted to ΔSSS using a slope of 0.15‰/SSS at sites Bass River and Ancora [56], 0.38‰/SSS for site 1209 [42], 0.24‰/SSS for site 690 [57,58] and 0.55‰/SSS for site 401 [59].

At each site, the pre-CIE B/Ca average of each species was used as the baseline, and ΔB/Ca was calculated for each sample that also has a paired Mg/Ca and δ¹⁸O-based salinity anomaly estimate (figure 5). For Acarinina spp., the pre-CIE B/Ca average from site Bass River was applied to proximal site Ancora to compute ΔB/Ca because there are no Palaeocene B/Ca values at Ancora (figure 5b). B/Ca records were corrected for ΔSSS using the O. universa sensitivity of ΔB/Ca/ΔSSS (2.6 µmol mol⁻¹ per salinity unit) as determined in laboratory culture [50]. Both the uncorrected and salinity-corrected ΔB/Ca are shown in figure 5 with the exception of Morozovella spp. at sites Bass River and Ancora, as the B/Ca records do not extend into the Palaeocene (figure 2a). At the shelf NJ sites Bass River and Ancora, North Atlantic site 401 and Southern Ocean sites 689 and 690 regional salinity decreases amplify the total B/Ca anomaly, while the estimated salinity increase at North Pacific site 1209 dampens the total anomaly. These estimated salinity changes can account for a 13% increase in the total B/Ca anomaly at site 1209, and an 11% and 22% decrease in NJ and Southern Ocean sites, respectively. In our salinity correction on B/Ca, we assume that modern B/Ca salinity sensitivities apply to extinct PETM taxa. This should be considered an approximation because modern studies document species-specific B/Ca salinity responses [50], and it is unclear whether these modern salinity sensitivities might have been different under Palaeocene seawater composition. In light of these assumptions, we consider our salinity-corrected reconstructions as a first-order sensitivity test to evaluate the potential bias in B/Ca records. While regional salinity variations can account for some minor differences between B/Ca anomalies among studied sites, ultimately the trends remain robust and most consistent with ocean carbonate chemistry as the main driver of PETM B/Ca records.

Figure 5.

Estimation of the potential contribution of regional changes in salinity to B/Ca records. B/Ca anomalies (ΔB/Ca, µmol mol⁻¹) were computed by subtracting a pre-CIE average value from each point in the record (a,b). Paired planktonic foraminiferal Mg/Ca and δ¹⁸O were used to estimate δ¹⁸O_sw changes; regional δ¹⁸O_sw–salinity relationships were used to translate δ¹⁸O_sw to ΔSSS. Plotted here are ΔB/Ca records uncorrected (dashed line) and corrected (bold line) for regional changes in salinity using B/Ca salinity sensitivity of modern O. universa derived in laboratory culture experiments [50]. The shaded region represents the potential contribution of local salinity changes to ΔB/Ca records. Age is denoted in kiloyears relative to the CIE. (Online version in colour.)

(d). Impact of preservation

With multiple sites from a diverse range of palaeoceanographic settings, palaeodepths, sediment burial depths, lithology and diagenetic pathways, we can assess the potential impact of preservation on individual boron proxy records. Differential dissolution tends to lower B/Ca of surface planktonic foraminifera [60], but shows little impact on thermocline dwellers [61]. It was speculated that variable dissolution response between species is the result of heterogeneous B distribution within the test and relatively greater dissolution susceptibility of certain test textures [61]. δ¹¹B of T. sacculifer was found to lower with increasing water depth, again as a result of enhanced dissolution [62–64]. However, δ¹¹B in Globigerinoides ruber does not follow a similar dissolution pattern [63,64]. This variable dissolution susceptibility between different species is argued to be a result of differential δ¹¹B heterogeneity within the test or variable dissolution susceptibility of ontogenetic versus gametogenic calcite [65]. The proximity of sites 689 (1100 m) and 690 (1900 m) but with different palaeo water depths provides a scenario to test a depth-related dissolution influence on B/Ca. The close similarity of the absolute values and trends in sites 689 and 690 B/Ca records despite disparate palaeodepths suggests a minimal and consistent influence of dissolution (figure 2b,c).

An additional opportunity to assess the foraminiferal preservation state comes with the strontium/calcium ratio (Sr/Ca) of biogenic CaCO₃, which is released into pore fluids during calcite dissolution and excluded during inorganic calcite precipitation. Indeed, detailed Sr/Ca measurements of diagenetically altered foraminiferal calcite show lower values, clearly demonstrating the effect of dissolution and recrystallization [66]. Sr/Ca values at Bass River and Ancora are generally higher than the value typical of diagenetic calcite (1.0 mmol mol⁻¹) [66], indicating that diagenesis did not compromise the geochemical records, with the exception of samples within two discrete depth intervals not included in the reconstructions [11]. Foraminiferal Sr/Ca at site 1209 is lower (near 1 mmol mol⁻¹ [42]), consistent with the ‘frosty’ texture of foraminifers at that site. However, Sr/Ca is near constant throughout the PETM interval, suggesting that if recrystallization has altered foraminiferal trace element compositions it has done so uniformly over this short time interval and, therefore, relative changes in B/Ca should be preserved.

Another strategy to evaluate the effects of preservation on geochemical signals is to compare sites with identical environmental conditions but demonstrably different diagenetic sediment histories. Edgar et al. [65] compared time-equivalent (Middle Eocene) planktonic foraminiferal B/Ca and δ¹¹B records from glassy foraminifers (Tanzanian Drilling Project, Indian Ocean) with a site that contains recrystallized foraminifera (ODP 865, central Pacific Ocean). Planktonic foraminiferal B/Ca are lower in the recrystallized (frosty) foraminifers from site 865, yet, by contrast, δ¹¹B values were similar, suggesting that exchange between pore fluids and recrystallized carbonate can result in a lowering of boron concentration without isotopic fractionation. This predicts that B/Ca in diagenetically precipitated inorganic calcite is lower than biogenic calcite, which is consistent with the idea of preferential B exclusion during recrystallization. While this raises some concern for the pelagic records, site 865 represents an extreme example of recrystallization as the entire sediment column consists of a high-porosity ooze dominated by foraminifera sand so that recrystallization is occurring in a completely or partially open system at high seawater to CaCO₃ ratios [67]. This contrasts with the lower porosity pelagic sites in this study, where pore water chemistry is mainly controlled by diffusion and diagenesis occurs in a relatively closed system at lower seawater to CaCO₃ ratios. Furthermore, the continental shelf sites (sites Bass River and Ancora) exhibit exceptionally well-preserved foraminifera with classic glassy shell textures similar to the preservation seen in Eocene Tanzania sections [68]. The relatively impermeable nature of clay-rich sediments limits or even prevents interaction of foraminiferal calcite with pore fluids, thus preserving primary chemical signatures. Visual evaluation of the site Bass River foraminifera by scanning electron microscopy supports this claim [11]. Remarkably, despite these different diagenetic histories and preservation states between the shelf and pelagic sites, a similar pattern of B/Ca is recorded, suggesting that any diagenetic bias on relative trends of individual records is minimal (figure 2a–c).

(e). Symbiont loss

A potential biological consequence of warming is loss of photosymbionts (bleaching) in planktonic foraminifera. Evidence of symbiont bleaching in planktonic foraminifera was previously suggested for the Middle Eocene [69,70]. Since the early stages of δ¹¹B proxy development, it was apparent that biogenic CaCO₃ is likely not to be simply a recorder of ambient seawater pH but also carries the signal of local pH variation within the calcifying microenvironment. Physiological processes including symbiont photosynthesis, respiration and calcification can modify local pH [71,72] and, in turn, B/Ca and δ¹¹B of foraminiferal calcite [73,74]. Detailed intra-test measurements in symbiont-bearing benthic foraminifera [75], light intensity in planktonic foraminifera culture experiments [53,73] and sediment-trap work [76] collectively substantiate the potential impact of symbiont activity on B/Ca and δ¹¹B.

Test size δ¹³C gradients in PETM planktonic foraminifera taxa are used as an indicator of photosymbiont activity [45] and loss of gradients as a proxy for bleaching in symbiont-bearing species [70]. Preferential use of light carbon (¹²C) during photosynthesis in combination with either increased photosymbiont abundance or activity in larger individuals can explain the positive relationship between test size and δ¹³C. The δ¹³C test size gradients in symbiont-bearing species (A. soldadoensis and Morozovella spp.) in shelf sites are maintained within the CIE, suggesting that, at temperate latitudes, bleaching did not occur [11]. Deeper dwelling Subbotina spp. also record a B/Ca and δ¹¹B change, corroborating the trends shown in symbiotic surface dwellers within the same cores, signifying that even if bleaching did occur it cannot explain the entire signal (figures 2a–c and 3). In addition, Southern Ocean sites (689, 690) were located in comparatively colder polar waters where foraminifera would presumably be less vulnerable to warming-induced bleaching. The close similarity of the B/Ca trends between the cold high latitudes (sites 689 and 690), the warm low latitudes (site 1209), and the temperate continental shelf (NJ margin) argues against symbiont bleaching as a major driver of the B/Ca trends at all sites.

(f). Biological implications

Our global compilation of boron proxy (δ¹¹B and B/Ca) reconstructions reveals near-uniform upper ocean acidification (figures 3c and 4). The similarity in B/Ca values between records implies minimal spatial variability in absolute ocean carbonate chemistry (e.g. ) (figure 4). Since the degree of acidification was uniformly expressed in the surface ocean then there appears to be no refuge. The carbonate saturation state, on the other hand, could be spatially variable, which is suggested during the PETM by model output [77]. However, if we follow this logic, the high latitudes would be closer to undersaturated values with respect to CaCO₃ and thus are more vulnerable to acidification. There could be short-term responses to ocean acidification during the CIE onset that are not captured in our B proxy records, as dissolution gaps at the CIE onset are common in most sections, even in the shallow marine shelf sections. Another possibility is that the combined effects of ocean acidification and warming on marine biota could be more significant than ocean carbonate chemistry as an independent stressor.

Several major plankton groups experienced dynamic changes during the PETM, including evolutionary turnover and shifts in latitudinal distribution, probably in response to warming. Moreover, geochemical evidence indicates that tropical sea-surface temperatures probably exceeded 30°C, thereby pushing thermal thresholds observed in the modern ocean [78,79]. In the tropics, mixed-layer dwelling planktonic foraminifera (morozovellids and acarininids) exhibit a transient increase in morphological variation that included the appearance of distinctive (malformed) ecophenotypes, whereas populations of thermocline-dwelling taxa (subbotinids) temporarily collapsed at some locations [80]. A similar pattern of migration and elevated levels of taxonomic turnover is also observed in calcareous nannoplankton [81], Apectodinium spp. dinocysts [82] and radiolarians [83]. These lines of evidence are consistent with the heterogeneous latitudinal pattern of ocean warming [7].

There is a wealth of evidence implicating warming as an agent of change amongst plankton communities, yet the high incidence of malformed nannoplankton and thinning of coccolithophores found at numerous sites during the peak of the CIE suggests that ocean acidification also played a role [84–87]. Still, the impact of undersaturation on nannoplankton was exceeded by that of temperature; extracellular coccolithophore groups thought to be sensitive to ocean chemistry took refuge in cool waters at high latitudes [77]. Moreover, increased rates of coccoliths and foraminifera malformation in coastal sites suggest that the pH decline in combination with warming and eutrophication increased stress on planktonic calcifiers [88]. A similar combination of stressors was previously suggested for the collapse of framework-building corals during the PETM [89]. Disentangling the biotic effects of multiple abiotic stressors is proving to be a challenging task and should be a focus of future research. Nevertheless, the relative impacts of future ocean acidification are likely to be more severe as anthropogenic emission rates are considerably faster than estimated for the PETM [12,90,91].

5. Conclusion

The similarities between foraminiferal boron proxy records over the PETM, from a variety of palaeoceanographic settings and diagenetic histories, is consistent with globally uniform ocean acidification as the primary driver for the first-order feature common to all records: the large, rapid decrease in B/Ca and δ¹¹B at the onset of the CIE. Subtle differences in absolute values between sites can be explained by differences in local oceanographic conditions, including salinity and dissolution. The boron isotope records at several of our sites allow us to quantify the degree of ocean acidification and suggest approximately a 0.15–0.3 pH unit decrease in the upper ocean. These estimates overlap with the higher end of carbon emissions scenarios considered in carbon cycle model simulations of the event [14,15]. This degree of acidification is less severe than the 0.4 pH unit decrease predicted for the end of the twenty-first century due to anthropogenic carbon emissions [92] and probably occurred over several thousand years [90,91], an order of magnitude slower than current ocean acidification. This might explain the general lack of extinctions among mixed-layer dwelling calcifiers during the PETM.

Finally, all boron records indicate a prolonged state of moderate acidification lasting as long as the CIE, adding to the mounting evidence (including the ‘body’ of the CIE itself [15] and a delayed CCD overshoot [31]) that carbon emissions continued for several tens of thousands of years following the onset of the PETM.

Data accessibility

The geochemical datasets supporting this article have been uploaded as part of the electronic supplementary material, tables S1–S3.

Authors' contributions

T.L.B. collected data, constructed age models, performed pH and carbonate chemistry estimates and drafted the manuscript and figures. D.E.P. collected data, constructed age models and assisted in drafting the manuscript. B.H. collected New Jersey boron isotope data and assisted in drafting the manuscript. D.C.K. assisted in planktonic foraminifera taxonomy. D.C.K. and T.J.B. contributed ODP 689 and 690 sample materials. D.C.K., T.J.B. and Y.R. assisted in drafting the manuscript. J.C.Z. directed the study and assisted in drafting the manuscript. T.L.B., D.E.P., T.J.B., Y.R. and J.C.Z. conceived of and designed the project. All authors read and approved the final version of the manuscript.

Competing interests

We have no competing interests.

Funding

This research was supported by NSF grant no. OISE-1107787, Schlanger Ocean Drilling Fellowship and Joanna Resig Foraminiferal Research Fellowship awarded to T.L.B., a Flint Postdoctoral Fellowship granted to D.E.P., NSF OCE 12-20554 to B.H. and NSF grant no. OCE-1415958 awarded to T.J.B. and J.C.Z.