Understanding cold bias: Variable response of skeletal Sr/Ca to seawater pCO2 in acclimated massive Porites corals

Coral skeletal Sr/Ca is a palaeothermometer commonly used to produce high resolution seasonal sea surface temperature (SST) records and to investigate the amplitude and frequency of ENSO and interdecadal climate events. The proxy relationship is typically calibrated by matching seasonal SST and skeletal Sr/Ca maxima and minima in modern corals. Applying these calibrations to fossil corals assumes that the temperature sensitivity of skeletal Sr/Ca is conserved, despite substantial changes in seawater carbonate chemistry between the modern and glacial ocean. We present Sr/Ca analyses of 3 genotypes of massive Porites spp. corals (the genus most commonly used for palaeoclimate reconstruction), cultured under seawater pCO2 reflecting modern, future (year 2100) and last glacial maximum (LGM) conditions. Skeletal Sr/Ca is indistinguishable between duplicate colonies of the same genotype cultured under the same conditions, but varies significantly in response to seawater pCO2 in two genotypes of Porites lutea, whilst Porites murrayensis is unaffected. Within P. lutea, the response is not systematic: skeletal Sr/Ca increases significantly (by 2–4%) at high seawater pCO2 relative to modern in both genotypes, and also increases significantly (by 4%) at low seawater pCO2 in one genotype. This magnitude of variation equates to errors in reconstructed SST of up to −5 °C.

Scientific RepoRts | 6:26888 | DOI: 10.1038/srep26888 The influence of pH and DIC on coral skeletal Sr/Ca is poorly constrained. Culture studies investigating the effect of reduced seawater pH across a range of ocean acidification scenarios yield mixed results: seawater pH and Sr/Ca were negatively correlated in Acropora digitifera 26 and newly settled recruits of Favia fragum 27 but were unrelated in Montipora capitata 13 . No systematic variation in skeletal Sr/Ca was found over a wider pH range (both lower and higher than present day) in Stylophora pistillata 28 . With the exception of A. digitifera 26 , these studies used addition of HCl or NaOH to manipulate seawater pH at constant seawater pCO 2 , consequently shifting other carbonate system parameters to unrealistic values 29 . Additionally, the short acclimation times (5-35 days) of three [26][27][28] of the four studies are probably insufficient to enable coral physiological responses e.g. changes in zooxanthellae density, to adapt to changes in environment.
We test the direct impact of variations in seawater pCO 2 on skeletal Sr/Ca in massive Porites spp. corals, the coral genus most commonly used for palaeoclimate reconstruction. Heads from 3 genetically distinct corals (2× Porites lutea and 1× Porites murrayensis; Fig. 1) were divided into smaller sub-colonies (each > 8 cm diameter) and cultured in a mixture of natural and synthetic seawater. Corals were housed in a purpose-built large-volume aquarium system constructed of low CO 2 permeability materials. The seawater in each treatment was bubbled with gas mixes set to reach the target seawater pCO 2 compositions. Corals were maintained at ambient pCO 2 conditions for 2 months, adjusted to pCO 2 treatment conditions over another 2 months and then acclimated at the final treatment pCO 2 for 5 months. Final seawater pCO 2 target levels ranged from the last glacial maximum (LGM; ~180 μ atm), through present day (~400 μ atm) to levels projected by the year 2100 (~750 μ atm). Actual seawater pCO 2 levels (198, 416 and 750 μ atm) in the reservoirs were calculated from total alkalinity and DIC using the CO2SYS program 30 . Seawater Sr/Ca, temperature, salinity and DIC system parameters were maintained within narrow limits throughout the study ( Table 1).
The skeleton deposited in the 5-week period, following the acclimation, was identified by alizarin red staining ( Fig. 1), and skeletal Sr/Ca of this region was analysed by secondary ion mass spectrometry (SIMS). Skeletal deposition is concentrated at the skeleton surface in massive Porites corals, although subtle thickening of the trabeculae may occur over a period of weeks 31 . The high spatial resolution of SIMS allows the selective analysis of skeleton deposited at the skeletal surface, avoiding the centres of calcification where Sr/Ca is anomalously high, and any material which thickens the trabecula and may be deposited a considerable time later 32 . The accuracy of our SIMS estimates is hindered by uncertainty in the elemental composition of the NaHaxby2 standard, but our analytical precision, which is critical to palaeothermometry, is good (within 0.6%, or 0.018 mmol mol −1 , (2σ ), equivalent to ± 0.4 °C).

Results and Discussion
Skeletal Sr/Ca varies with seawater pCO 2 in some genotypes. Coral aragonite Sr/Ca partition coefficients (K D Sr/Ca = skeletal Sr/Ca/seawater Sr/Ca) are estimated for each coral (Supplementary Table S1). K D Sr/Ca varies significantly in response to perturbations in seawater pCO 2 in both genotypes of P. lutea but is unaffected in P. murrayensis (Fig. 2). Seawater temperatures do not vary significantly between the treatments (Table 1) and we observe excellent agreement (within 0.3%) in the K D Sr/Ca of duplicate sub colonies of the same coral genotype within each treatment (Fig. 2), indicating that minor differences in coral positioning and lighting in each tank do not affect skeletal Sr incorporation. K D Sr/Ca is significantly increased at high seawater pCO 2 compared to ambient in both genotypes of P. lutea (by 4% in genotype 1, and 2% in genotype 2; p < 0.05, ANOVA and Tukey post-hoc). K D Sr/Ca is also significantly increased (by 4%; p < 0.05) in P. lutea genotype 2 at low pCO 2 compared to ambient. In contrast, K D Sr/Ca does not vary significantly between the P. murrayensis sub-colonies cultured over the full seawater pCO 2 range.
Two hypotheses have been proposed to explain the impact of seawater pCO 2 on coral skeletal Sr/Ca 26 . Coral aragonite precipitates from an extracellular calcifying fluid enclosed in a semi-isolated space between the coral tissue and underlying skeleton, and corals increase the pH of this fluid (upregulate pH) above that of ambient seawater to promote high fluid aragonite saturation states favourable for skeletal precipitation. The coral calcification fluid is derived from seawater which is transported paracellularly (between cells) to the calcification site and from additional Ca transported transcellularly (across cells) via L-type Ca channels 33,34 and the enzyme Ca-ATPase 33,35 . Enhancing transcellular Ca transport could decrease the Sr/Ca of the calcification fluid. This hypothesis is appealing as corals maintained at high seawater pCO 2 are able to upregulate calcification fluid pH more to partially offset the effect of lowered seawater pH 36 . Ca-ATPase is a Ca 2+ :H + antiport which increases the pH of the calcification fluid 37 and may be expected to have a higher activity under high seawater pCO 2 . This hypothesis suggests that coral skeletal Sr/Ca is decreased at high seawater pCO 2 . However, Sr 2+ has a similar ionic radius to Ca 2+ and inhibition of Ca channels and Ca-ATPase in the branching coral, Pocillopora damicornis, did not affect skeletal Sr/Ca, suggesting that transmembrane Ca transport does not fractionate Sr/Ca in the calcification fluid 38 .
Alternatively, skeletal Sr/Ca variations may reflect elemental partitioning during aragonite precipitation (Rayleigh fractionation). Rayleigh fractionation occurs when variable volumes of an isolated fluid reservoir are used for solid precipitation, and a trace element ion (in this case Sr) is either discriminated against or preferentially incorporated compared to the major precipitating ion (in this case Ca). The Sr/Ca partition coefficient in aragonite is > 1 39 , so Sr is preferentially incorporated into the growing crystals over Ca 40 . As precipitation proceeds, the Sr/Ca of the fluid remaining in the reservoir, and of the carbonate subsequently precipitated from it, decreases. The final Sr/Ca of the aragonite reflects the proportion of the reservoir used in precipitation 40 . The deposition rate of coral aragonite is strongly dependent on the saturation state of the calcifying fluid 41,42 . At high calcification fluid saturation states, a higher proportion of the fluid reservoir may be used for calcification resulting in a low skeletal Sr/Ca. Although corals at high seawater pCO 2 increase the pH of the calcification fluid more than their counterparts cultured at ambient pCO 2 , they do not ultimately attain such a high fluid pH 36 and calcification rates in these corals are usually reduced 43 . This hypothesis suggests that coral skeletal Sr/Ca will be increased at high seawater pCO 2 . Neither hypothesis immediately explains our observed pattern of skeletal Sr/Ca Scientific RepoRts | 6:26888 | DOI: 10.1038/srep26888 variations where skeletal Sr/Ca is increased in some corals at both low and high seawater pCO 2 . However it is unlikely that the two hypotheses operate independently as the saturation state of the calcification fluid is probably highly responsive to Ca-ATPase activity.
Growth rate and skeletal Sr/Ca. We explored the relationship between coral aragonite:seawater K D Sr/Ca and skeletal growth rate, inferring that Rayleigh fractionation is linked to skeletal precipitation rate. We observe significant differences in both the skeletal extension and calcification rate of some genotypes between treatments (Supplementary Table S2). However, correlation between K D Sr/Ca and either skeletal linear extension or calcification rate ( Supplementary Fig. S1) are insignificant in all coral genotypes (p > 0.3; Pearson's correlation). Our data suggest that variations in Sr incorporation between corals do not reflect Rayleigh fractionation.
Alternative controls on skeletal Sr/Ca. We observe significant variation in skeletal Sr/Ca between different coral genotypes cultured at the same seawater pCO 2 (Fig. 3) demonstrating that other factors also influence Sr incorporation. The origin of this is currently unknown. The mucus layer at the coral surface behaves as a Donnan matrix, concentrating Ca 2+ from seawater prior to transport across the coral ectoderm tissue 44 and matrix variations between corals could conceivably subtly influence the Sr/Ca of seawater transported to the calcification fluid. Organic molecules at the calcification site also play a major role in regulating aragonite nucleation and growth 45 . Skeletal organic materials differ between coral species 46 , and may influence skeletal trace element content either because biomolecules impact the incorporation of trace elements in the crystal lattice, or due to direct organic-metal complexing. Understanding the role of these organic templates in fractionating Sr/Ca during calcification, and quantifying differences between coral genotypes even within an individual species, will be a critical step forward in resolving biological influences on skeletal Sr/Ca from the thermal signature.
Implications for SST reconstructions. Our data show that seawater pCO 2 can be a significant factor affecting the Sr/Ca of massive tropical coral skeletons frequently used for palaeotemperature reconstruction. Decreasing seawater pCO 2 from ambient (416 μ atm) to values consistent with the LGM (198 μ atm) increases skeletal Sr/Ca by 4% in one of the three coral genotypes. The sensitivity of coral skeletal Sr/Ca to temperature is typically ~0.8-1% °C −1 and the effect of the decrease in seawater pCO 2 implies a cooling of 4-5 °C in this single genotype, although all colonies were cultured at the same temperature. The influence of seawater pCO 2 on skeletal Sr/Ca may explain the 'cold bias' observed in some fossil corals, which suggest that tropical SSTs at the last or penultimate deglaciations in the Atlantic 47 and Pacific 48 Oceans were ~6 °C cooler than at present. These estimates contrast with other marine proxies which indicate that SSTs typically cooled by 2-3 °C at most at this time 49,50 . Proposed variations in seawater Sr/Ca over glacial-interglacial periods are insufficient to explain the 'cold bias' observed in coral SST reconstruction 20 .
We find that the influence of seawater pCO 2 on skeletal Sr/Ca is inconsistent between corals, even of the same species, and this prevents the calculation of a correction factor to compensate for past variations in seawater pCO 2 . Skeletal Sr/Ca is unaffected by pCO 2 over the modern-LGM range relevant to paleotemperature reconstruction in two of our cultured genotypes, yet one genotype shows a substantial, and significant response that is observed across duplicate sub-colonies. We are unable to relate the seawater pCO 2 effect to coral growth rate indicating that coral extension cannot be used to correct skeletal Sr/Ca for an indirect influence of pCO 2 in Porites spp., despite successful application of a growth correction in other coral genera 51 . Further culture work is required to resolve the relationship between skeletal Sr/Ca and seawater pCO 2 across a wider range of coral genotypes, and to identify how seawater pCO 2 contributes to 'cold bias' . At present, the accuracy of absolute tropical past SST estimates from skeletal Sr/Ca of fossil corals is limited by this uncertainty. In contrast, relative changes in annual seawater temperature are usually well preserved in coral skeletal Sr/Ca 52 and fossil specimens may yet provide good records of Sr/Ca derived seasonal SST variations which can be used to infer the amplitude and frequency of past ENSO 53 and interdecadal climate events 54 . However, it will also be important to quantify in further studies whether skeletal Sr/Ca temperature sensitivity varies between different coral genotypes, and is affected by seawater pCO 2 .
In this study we hypothesise that inconsistencies in SST reconstructions of the LGM from coral Sr/Ca may be explained by a change in seawater pCO 2 between the modern and glacial ocean, equivalent to ~220 μ atm. Approximately 50% of this increase in pCO 2 towards present day has occurred in the last 150 years as a result of rising greenhouse gas emissions, overlapping with multi-centennial coral Sr/Ca records. Anthropogenic CO 2 emissions are isotopically depleted in 13 C, and a centennial-scale decrease in the δ 13 C of surface ocean DIC attributable to rising atmospheric CO 2 concentrations 55 is also observed in the skeletal aragonite of > 200 year-old corals collected at the end of the 20 th century 56,57 . Between 1850 and 2000, the ~85 μ atm increase in atmospheric CO 2 is clearly preserved in the carbon isotopic ratios of these corals, yet there is no discernible influence on skeletal Sr/Ca 5,6 . In our culture study, a significant increase in skeletal Sr/Ca was observed at low pCO 2 compared to ambient in one of three genotypes, and we may therefore expect some corals to show a decrease in Sr/Ca with rising CO 2 over the 20 th century. In practice, any CO 2 forcing on coral Sr/Ca is likely to be obscured by significant interannual variations in SST caused by ENSO against a global trend of increasing SST 5,6,58 . It is also important to   Table S1) across 2 or more skeletal units within individual colonies. Within each species/genotype, different letters indicate significant differences between treatments (p < 0.05; ANOVA and Tukey post-hoc). Error bars represent the combined 95% confidence limits of seawater and skeletal Sr/Ca measurements. bear in mind that 70% of the increase in atmospheric CO 2 between 1850 and 2000 occurred in the second half of the twentieth century, so the majority of these multi-centennial coral Sr/Ca records span very small (10-20 μ atm) changes in pCO 2 . Longer coral Sr/Ca datasets extending into the 21 st century are required to investigate the effect of > 100 μ atm changes in CO 2 on in situ skeletal Sr/Ca.
Our study demonstrates that the variations in published skeletal Sr/Ca-SST calibrations between corals of the same species do not necessarily reflect differences in physical (e.g. temperature, light levels and ocean currents) and chemical (e.g. nutrient levels and composition) conditions 22,52 . Rather skeletal Sr/Ca varies significantly between coral genotypes even of the same species cultured under the same conditions. Resolving how genotypic variations affect coral biomineralisation and impact skeletal geochemistry is a key target for future research.

Methods
Aquarium System. Culturing experiments were carried out in the marine culturing facility at the Department of Earth & Environmental Sciences, University of St Andrews, UK. Corals were maintained in 21 litre cast acrylic tanks, recirculated from high density polyethylene reservoirs containing ~900 litres of seawater. The reservoirs were bubbled (at 10 L min −1 ) with gas mixes set to reach the target seawater pCO 2 compositions. The modern day treatment was aerated with untreated ambient air. The high and low CO 2 treatments were aerated with ambient or low-CO 2 air, respectively, combined with high purity CO 2 (Foodfresh, BOC, UK). Flow rates of air and CO 2 were regulated by high-precision mass flow controllers (SmartTrak 50 Series, Sierra USA) controlled by purpose-written MATLAB ® programs. Low-CO 2 air was produced by bubbling ambient air through a caustic solution (0.9 M NaOH and 0.1 M Ca(OH) 2 ) and rinsing it by bubbling through deionised water 59 . The [CO 2 ] of the low-CO 2 air (before mixing with CO 2 ) was monitored every 2 hours by automated non-dispersive infra-red CO 2 analysers (WMA04, PP systems, USA) and ranged from 20-100 μ atm depending on the age of the caustic solution. The [CO 2 ] of the low, ambient and high CO 2 gas streams (after addition of any CO 2 ) was monitored automatically 3-4 times per day and were 180 ± 3, 400 ± 5 and 761 ± 6 μ atm (mean ± 1σ ) over the experimental period. Corals were maintained under LED lighting (Maxspect R420R 160w-10000k) on a 12 h light: 12 h dark cycle, with wavelength settings of 100% A and 20% B such that light intensity at coral depth was ~300 μ mol. Corals were fed weekly with rotifers.

Monitoring Seawater Composition. Dissolved inorganic carbon (DIC) was measured weekly in each
reservoir by LI-7000 CO 2 differential, non-dispersive, infrared gas analyser (Apollo SciTech; AS-C3). Samples were calibrated against a natural seawater certified reference material (CRM; A. Dickson, Scripps Institution of Oceanography). Internal reproducibility was calculated from the standard deviation of 8 replicate measurements of a single sample (σ /√ n), and was always < 0.1%. Multiple measurements of the CRM were analysed as unknown samples over the 4 week period to check the calibration, and these were in good agreement with the certified value (unknown = 2019 ± 6 (1σ ) μ mol kg −1 , n = 4; CRM = 2014 μ mol kg −1 ).
We adjusted addition rates to maintain total alkalinity within narrow limits (±≤14 μ mol kg −1  Seawater samples were collected weekly during the experimental period for Sr and Ca analysis by quadrupole ICP-MS (Thermo Scientific X Series) at the National Oceanography Centre, Southampton. Samples were diluted 1000-fold in 5% HNO 3 (with 5 ppb In as an internal standard) and calibrated against matrix-matched synthetic standards prepared from 1000 μ g ml −1 single-element stock solutions (Inorganic Ventures) in 5% HNO 3 . Based on replicate analyses (n = 4) of IAPSO standard seawater, external reproducibility (σ /√ n) was 0.6% for Sr/Ca. Measured values of Sr/Ca in IAPSO (8.78 ± 0.05 (1σ ) mmol mol −1 ) were in excellent agreement with certified values (8.77 mmol mol −1 ). Nutrients were measured in filtered (Whatman GFF; ~0.7 μ m) seawater samples from each reservoir by flow cell spectrophotometry (Lachat 8000; Scottish Association of Marine Science). Concentrations of NH 4 + (< 0.19 ± 0.01 μ mol L −1 ; mean ± 1σ ), PO 4 3− (< 0.06 ± 0.00 μ mol L −1 ), SiO 4 4− (< 1.3 ± 0.03 μ mol L −1 ) and NO 3 + NO 2 (< 0.95 ± 0.02 μ mol L −1 ) were within the range reported for pristine reef waters 60 . Coral sample processing and SIMS. Skeletal samples were prepared from the maximum growth axis and fixed in epoxy resin (Epo thin, Struers UK) in 2.5 cm diameter circular moulds, under vacuum. Polished, gold coated sections were analysed using a Cameca imf-4f ion microprobe in the School of Geosciences at the University of Edinburgh ( 16 O − ion beam accelerated at 10.8 kV; primary beam current = 8 nA; energy offset = 75 eV; field aperture 1; contrast aperture 2). Trace element data were collected over 8 days, and multiple analyses of a fossil coral aragonite standard, NaHaxby2 (Sr/Ca = 2.87 mmol mol −1 ), were made each day. Calculation of relative ion yields (RIYs) and standardisation of data was performed as in Allison et al. 16 . Internal reproducibility was calculated from ten cycles of a single SIMS analysis (2σ /√ 10) and was ~0.4% for Sr/Ca. External reproducibility (the precision of ~15 daily analyses on the standard) was 0.6%, or 0.018 mmol mol −1 , (2σ ) for Sr/Ca. Where significant variations in the RIY for NaHaxby2 were observed between days, instrument drift was distinguished from heterogeneity in the composition of the standard through additional analyses of the same area on a subsequent day. Samples were calibrated against mean standard RIYs, either from daily or weekly analyses depending on this assessment of instrument performance. See Allison et al. 32 for representative images of skeletal ultrastructure and SIMS analyses pits 32 .
Daily calcification rates (μ mol CaCO 3 cm −2 d −1 ) were estimated for each sub colony from the measured change in seawater TA over an isolation period in the light (5 hours), and the dark (7 hours), on three separate occasions over the experimental period. Linear extension (μ m) was measured as the distance between the alizarin stain lines in each analysed trabecula.