Coral calcifying fluid aragonite saturation states derived from Raman spectroscopy

Quantifying the saturation state of aragonite (Ar) within the calcifying fluid of corals is critical for understanding their biomineralization process and sensitivity to environmental changes including ocean acidification. Recent advances in microscopy, microprobes, and isotope geochemistry enable the determination of calcifying fluid pH and [CO2− 3 ], but direct quantification of Ar (where Ar = [CO2− 3 ][Ca 2+ ]/Ksp) has proved elusive. Here we test a new technique for deriving Ar based on Raman spectroscopy. First, we analysed abiogenic aragonite crystals precipitated under a range ofAr from 10 to 34, and we found a strong dependence of Raman peak width onAr with no significant effects of other factors including pH, Mg/Ca partitioning, and temperature. Validation of our Raman technique for corals is difficult because there are presently no direct measurements of calcifying fluid Ar available for comparison. However, Raman analysis of the international coral standard JCp-1 produced Ar of 12.3± 0.3, which we demonstrate is consistent with published skeletal Mg/Ca, Sr/Ca, B/Ca, δ11B, and δ44Ca data. Raman measurements are rapid (≤ 1 s), high-resolution (≤ 1 μm), precise (derivedAr± 1 to 2 per spectrum depending on instrument configuration), accurate (±2 if Ar < 20), and require minimal sample preparation, making the technique well suited for testing the sensitivity of coral calcifying fluid Ar to ocean acidification and warming using samples from natural and laboratory settings. To demonstrate this, we also show a high-resolution time series ofAr over multiple years of growth in a Porites skeleton from the Great Barrier Reef, and we evaluate the response of Ar in juvenile Acropora cultured under elevated CO2 and temperature.


Introduction
The calcium carbonate (CaCO 3 ) skeletons built by coral polyps are the building blocks of massive coral reef structures that protect shorelines, bolster tourism, and host some of the greatest concentrations of biodiversity on the planet (Knowlton et al., 2010;Costanza et al., 2014).Critical to the coral calcification process is the extraction of Ca 2+ and CO 2− sions continue unabated, by the end of the 21st century the [CO 2− 3 ] of surface seawater is projected to decline to ∼ 50 % of pre-industrial levels (Hoegh-Guldberg et al., 2014).This rapid change in ocean carbonate chemistry, likely unprecedented for hundreds of millions of years (Hönisch et al., 2012;Zeebe et al., 2016), has sparked concerns for coral growth.Indeed, laboratory experiments repeatedly demonstrate that coral calcification rates decrease in response to lower [CO 2−  3 ] or [Ca 2+ ] (Gattuso et al., 1998;Chan and Connolly, 2013;Comeau et al., 2017), leading to projections that as CO 2 levels continue to rise calcification will decline to unsustainable levels, such that there is net reef erosion (Hoegh-Guldberg et al., 2007;Pandolfi et al., 2011).
Yet signs of resilience do exist.Some coral species are able to maintain normal calcification rates across large natural acidification gradients (Fabricius et al., 2011;Shamberger et al., 2014;Barkley et al., 2015), indicative of adaptation or acclimation (Barkley et al., 2017).One potential mechanism to counteract acidification of surrounding seawater is pH homeostasis at the site of calcification (Georgiou et al., 2015;Barkley et al., 2017).Corals calcify from an isolated fluid located between the living tissue and the existing skeleton (Barnes, 1970;Cohen and McConnaughey, 2003;Venn et al., 2011;Tambutté et al., 2012).Up-regulation of pH within this fluid, potentially achieved via proton pumping and/or symbiont photosynthesis, elevates the saturation state with respect to aragonite ( Ar ), contributing to rapid nucleation and growth of aragonite crystals (Gattuso et al., 1999;Cohen and McConnaughey, 2003;Al-Horani et al., 2003;McCulloch et al., 2012;Kubota et al., 2015).
Characterizing the Ar of the calcifying fluid, and understanding its sensitivities to variations in the reef environment, is therefore essential for accurately forecasting coral calcification responses to 21st century ocean acidification.Observing this fluid has proved difficult though, due to its small size and isolation beneath the living polyp (Clode and Marshall, 2002).Estimates of fluid carbonate chemistry have so far been derived from micro-electrodes, pH-sensitive dyes, boron isotopes, B/Ca, U/Ca, and bulk calcification rates (Al-Horani et al., 2003;Trotter et al., 2011;Venn et al., 2011;DeCarlo et al., 2015;Cai et al., 2016;Holcomb et al., 2016;Raybaud et al., 2017).However, these approaches do not always agree (Ries, 2011;Holcomb et al., 2014), and they have so far focused on calcifying fluid carbonate chemistry without considering the effect of [Ca 2+ ] on Ar .
A potential alternative approach to quantify calcifying fluid Ar is based on the Raman scattering from a laser focused onto the skeleton.When light interacts with a material, a small percentage (typically < 0.0001 %) of the photons are scattered inelastically (referred to as Raman or Stokes scattering), resulting in a change of energy and frequency (Smith and Dent, 2005).The frequency shifts associated with Raman scattering are characteristic of both the internal vibrations of a molecule and the lattice vibrations between molecules in a crystal, which makes Raman spectroscopy a valuable tool for mineral identification (Urmos et al., 1991;Dandeu et al., 2006;Brahmi et al., 2010;Clode et al., 2011;Nehrke et al., 2011;Stock et al., 2012;Foster and Clode, 2016;Stolarski et al., 2016;Roger et al., 2017).Importantly, Raman peaks can also provide information regarding the chemical composition of crystals and the conditions of the fluid from which they formed.For example, in abiogenic CaCO 3 , the shapes and positions of the ν 1 peak at ∼ 1085 cm −1 (which represents symmetric stretching of the carbonate C-O bond) have been correlated with Mg content and/or crystallinity (Bischoff et al., 1985;Wang et al., 2012;Perrin et al., 2016).A highly crystalline CaCO 3 with relatively few defects or impurities will have a narrow ν 1 peak because the C-O bonds throughout the crystals are of the same, or very similar, length (Bischoff et al., 1985).Defects in the crystals and/or trace element impurities cause positional disorder of CO 3 in the lattice.Positional disorder affects the length, and thus the strength and vibrational frequency, of C-O bonds (Bischoff et al., 1985).An increase in disorder leads to an increase in the distribution of C-O bond lengths, causing an increase in ν 1 peak width (Bischoff et al., 1985;Addadi et al., 2003;Lin et al., 2007;Wang et al., 2012;McElderry et al., 2013;Perrin et al., 2016).
In biogenic calcium carbonates, correlations have been reported between ν 1 peak widths and environmental conditions, including temperature and seawater pCO 2 (Kamenos et al., 2013(Kamenos et al., , 2016;;Hennige et al., 2015;Pauly et al., 2015).These changes potentially reflect differences in calcifying fluid carbonate chemistry.Crystals growing from more supersaturated solutions generally have more defects and incorporate more impurities (Watson, 2004), resulting in relatively disordered crystal lattices and wide Raman peaks (Urmos et al., 1991).This hypothesis is generally consistent with empirical observations of biogenic calcium carbonates, in which ν 1 peak widths increase as ambient seawater pCO 2 decreases and/or pH increases (i.e. higher Ar ) (Kamenos et al., 2013;Hennige et al., 2015;Pauly et al., 2015).However, corals exert strong control on the carbonate chemistry of their calcifying fluid by elevating pH and/or Ar at the site of calcification to facilitate more rapid crystal growth (Al-Horani et al., 2003;Venn et al., 2011;McCulloch et al., 2012).Thus, the observed correlations between Raman peak widths and ambient seawater conditions likely do not reflect the true sensitivity of aragonite Raman peaks to seawater chemistry, but rather the sensitivity of calcifying fluid chemistry to external pH and/or Ar .For this reason, information regarding the relationship between carbonate chemistry and Raman peak width in abiogenic experiments is required to quantitatively interpret Raman spectra of biogenic CaCO 3 with respect to the actual calcifying fluid conditions.
Here, we first evaluate the controls on Raman ν 1 peak width using abiogenically precipitated aragonites.These samples allow us to test the sensitivities of ν 1 peak width against geochemical composition (e.g.Mg/Ca) and fluid conditions (including Ar , pH, and temperature) directly, without the confounding influence of a coral polyp.We derive a calibration between ν 1 peak width and Ar , and apply it to estimate Ar of the international coral skeleton standard JCp-1, for which independent lines of geochemical evidence allow us to test the accuracy of our approach.Finally, we demonstrate the applicability of Raman spectroscopy by (1) reconstructing multiple years of Ar variability in a Porites coral collected from the Great Barrier Reef (GBR) (D 'Olivo and McCulloch, 2017), and (2) comparing Ar in cultured juvenile Acropora spicifera exposed to elevated CO 2 and temperature treatments (Foster et al., 2015).

Raman measurements
Raman spectroscopy was used to analyse the abiogenic aragonite precipitates described in DeCarlo et al. (2015) and Holcomb et al. (2016) (Supplement Table S1).Briefly, aragonite was precipitated by addition of Na 2 CO 3 and NaHCO 3 solutions to seawater.Various Na 2 CO 3 / NaHCO 3 ratios and pumping rates produced a range of Ar from 10 to 34 while achieving some independence between Ar and other carbonate system variables (e.g.r 2 between Ar and pH was only 0.34).Most (22 of 28) experiments were conducted at 25.5 • C, but two experiments were conducted each at 20, 33, and 40 • C. Two experiments (f08 and g13) were conducted from seawater with [Ca 2+ ] elevated by addition of dissolved CaCO 3 .
Raman spectra of the abiogenic aragonites were originally reported in DeCarlo et al. (2015) and Holcomb et al. (2016).These initial spectra were collected at Woods Hole Oceanographic Institution (WHOI) with a Horiba LabRam HR800 Raman spectrometer using a 785 nm laser source, 40× objective, 600 mm −1 grating, and CCD detector maintained at −70 • C. Spectral resolution was approximately 1.2 cm −1 .Three grains were analysed per experiment with an integration time of 5 s per spectrum.Subsequently, we repeated analyses of a subset of these aragonites (samples f02, f03, f06, g07, g13, h09) at the Centre for Microscopy, Characterisation, and Analysis (CMCA) at the University of Western Australia.These more refined measurements, which are the basis of this study, were performed with a WITec Al-pha300 RA+ confocal Raman microscope using a 785 nm laser source, 20× objective with numerical aperture of 0.5, 1200 mm −1 grating, and an Andor iDUS 401 CCD detector maintained at −60 • C. The nominal spectral resolution was 1.3 cm −1 .A silicon chip was analysed to facilitate comparison of peak position with other laboratories (the strong Si peak was present at 522.9 cm −1 ).For analysis of the abiogenic aragonites, at least five grains were analysed per experiment with 5-10 spectra per grain and integration times between 0.1 and 5 s (see Supplement for a comparison between instruments and discussion of the general applicabil-ity of our results to measurements in other laboratories).All analyses of coral skeletons described below were conducted with the WITec instrument using 1 s integration times.
The Porites coral skeleton standard JCp-1 (Okai et al., 2002) was analysed to facilitate comparison between Raman results and geochemical estimates of calcifying fluid Ar and [Ca 2+ ].We collected 440 spectra from various grains spread onto a glass slide.Although JCp-1 exists in ground form, we note that analysis of intact coral skeletons is also possible (Wall and Nehrke, 2012;Hennige et al., 2014) and that peak widths apparently do not depend on whether the sample is powdered or intact (Zakaria et al., 2008), as long as the grain size exceeds the laser spot size.
We also collected Raman spectra down-core in a Porites skeleton collected from the Great Barrier Reef (the sample is described in detail in D 'Olivo and McCulloch (2017).Briefly, the core was collected near Havannah Island, an inshore reef in the central GBR where corals bleached during 1998.The skeleton was cut in a slab and cleaned with bleach as described by D 'Olivo and McCulloch (2017), but no additional preparation was required.D 'Olivo and McCulloch (2017) present δ 11 B and trace element/Ca ratios from multiple tracks in the skeleton, and they show anomalies to the skeletal geochemistry corresponding to the 1998 thermal stress event.We used the automated microscope stage and area-mapping features of the WITec instrument to measure Raman spectra in a down-core transect at 50 µm resolution, covering approximately 6 years of growth from 1996 to 2002.The TrueSurface module was used to construct a topography map of the skeleton that was then followed by the instrument to ensure the sample was always in-focus for the Raman measurements.A maximum depth of 1 mm below the cut surface was set to limit the integration of signal from skeleton of different ages.Each time point represents the average of 10 spectra spaced horizontally in 100 µm increments.
Finally, we analysed skeletons of the cultured juvenile Acropora described in Foster et al. (2015Foster et al. ( , 2016)).Briefly, larval planulae were maintained in "Control" (24 • C, 250 µatm CO 2 ), "+T" (27 • C, 250 µatm CO 2 ), "+CO 2 " (24 • C, 900 µatm CO 2 ), and "+CO 2 + T " (27 • C, 900 µatm CO 2 ) treatments.Raman data were previously reported for these corals in Foster and Clode (2016), but those earlier measurements were conducted with a 600 mm −1 grating, which made it difficult to resolve changes in peak width (see Supplement).We therefore collected 25 new Raman spectra from each of three broken skeletal pieces from each of three corals, per treatment.The TrueSurface module was used to ensure all measurements were taken with the optics well focused on the samples.

Calculations and statistics
For all Raman spectra, we utilized the FWHM (full width at half maximum) intensity of the ν 1 peak (Fig. 1).A Gaussianshaped curve was fitted to each spectrum between  White (1974) and Bischoff et al. (1985): T, translation (lattice mode); L, libration (lattice mode); ν 4 , in-plane bend (internal mode); ν 1 , symmetric stretch (internal mode).This spectrum is readily identifiable as CaCO 3 , based on the strong ν 1 peak at approximately 1085 cm −1 , and distinguished as aragonite rather than other CaCO 3 polymorphs (calcite and vaterite) based on the positions of the T, L, and ν 4 peaks (Dandeu et al., 2006).The double peak between 700-710 cm −1 is the feature that most clearly distinguishes aragonite from calcite or vaterite (Urmos et al., 1991).(b) Measurement of ν 1 peak width.The grey points are the measured intensities at the corresponding Raman wavenumber shift.The black lines show the Gaussian curves fit to the data, and the grey boxes indicate the measured peak widths (ν 1 FWHM).Note that measured ν 1 FWHM is converted to true ν 1 FWHM for analysis in subsequent figures.The y axis scale is in arbitrary intensity units.
1100 cm −1 with the following expression: where y is the fit, x is the Raman shift between 1080 and 1100 cm −1 , b is the background intensity, m is the background slope, k is the peak height, p is the peak position (i.e.wavenumber), and s is the standard deviation.Peak intensity varies depending on the smoothness of the surface and the focus of the laser.We removed spectra with ν 1 peak height < 30 intensity units (signal : noise of approximately 5) due to the larger uncertainties associated with curves fit to these spectra.FWHM is calculated following Weisstein (2017): where s is the standard deviation of the Gaussian curve from Eq. ( 1).Measured Raman peak widths are known to be convolutions of the "true" peak widths and instrument noise (Nasdala et al., 2001;Wang et al., 2012;Váczi, 2014).The effect of instrument noise on Raman peak widths is directly related to the spectral resolution, which must be accounted for when comparing measurements conducted on different instruments (see Supplement).We used the Nasdala et al. (2001) formula to account for the effect of spectral resolution and calculate true FWHM: where b t is the true peak width, b m is the measured peak width, and r s is the spectral resolution.
In the analysis of the abiogenic aragonite, we used the linear model function in R (R Core Team, 2016) to test for relationships between ν 1 FWHM and Ar , aragonite precipitation rate, aragonite Mg/Ca, fluid [CO 2− 3 ], fluid pH, Raman peak height (k), and Raman peak position (p).Since Ar was the single variable most strongly correlated with ν 1 FWHM, we also evaluated linear models with two explanatory variables: Ar and the rest of the variables listed above.Analysis of variance (ANOVA) was used to test whether multivariate models were significantly better than the model based on Ar alone.We tested for normality of residuals with Kolmogorov-Smirnov tests and homogeneity of variances with Levene's test.
For the cultured Acropora, we calculated the mean Ar of each skeletal fragment using the abiogenic ν 1 FWHM-Ar calibration, weighting the 25 measurements per fragment by peak height.We then tested whether the data were significantly different from normal distributions with Kolmogorov-Smirnov tests and homogeneity of variances was checked with Levene's test.Finding no significant differences from normal distributions and no significant differences in variance among treatments, we conducted a two-way ANOVA and evaluated the effects with Tukey's honest significant difference test.Statistical significance was defined as p < 0.05.

Mg/Ca partitioning between aragonite and seawater
We also report here the partitioning of Mg/Ca between these abiogenic aragonites and seawater because (1) previous studies have suggested that Mg/Ca is important for interpret-ing Raman peak widths (Bischoff et al., 1985;Urmos et al., 1991), and (2) Mg/Ca is used as a constraint in our comparison of JCp-1 geochemistry and Raman spectroscopy.Partitioning of Mg 2+ between aragonite and seawater can be described by an exchange for Ca 2+ in the aragonite lattice, with a partition coefficient (K D ) expressed as where K D is dimensionless (Kinsman and Holland, 1969;Gaetani and Cohen, 2006).While there is some uncertainty in the mechanism of Mg 2+ incorporation into aragonite (Montagna et al., 2014), the K D nevertheless serves as an empirical measure of the distribution of Mg/Ca between aragonite and seawater under various conditions.DeCarlo et al. ( 2015) reported K D for Sr/Ca from these same samples, and here we follow the same calculations for K Mg/Ca D using the Mg/Ca data from Holcomb et al. (2016), except for the slight modifications to the calculations described below.Briefly, K Mg/Ca D was calculated from the measured Mg/Ca in the bulk precipitate (Holcomb et al., 2016), estimates of Mg/Ca in the initial fluid (seawater), and modelling the evolution of elemental concentrations in the fluid through the course of each experiment.However, since [Mg] in the initial seawater was not reported, we instead calculated initial [Mg] and [Ca] from the established relationships between salinity and concentrations of these elements in seawater (Riley and Tongudai, 1967).Our approach also differs from De-Carlo et al. (2015) in that they combined element/Ca measurements of both the final solution and the bulk solid to calculate K D , whereas we used only the reported aragonite Mg/Ca because final fluid [Mg] was not measured.Two of the experiments (f08 and g13) were conducted with initial elemental concentrations modified from seawater by addition of dissolved CaCO 3 , and they were excluded from K Mg/Ca D calculations due to uncertainty of the fluid Mg/Ca ratio.

Results
In the abiogenic aragonites analysed in this study, ν 1 FWHM was strongly correlated with seawater Ar (r 2 = 0.70, p < 0.001; Fig. 2 and Tables 1-2).Aragonite precipitation rate (G) is a function of Ar and temperature (T ) (Burton and Walter, 1987), and thus the experiments conducted at different T allow us to isolate the influence of Ar from G. In contrast to the strong dependence of ν 1 FWHM on Ar , there was no significant correlation between ν 1 FWHM and G (Fig. 3; Table 1).The ν 1 FWHM was significantly correlated with the aragonite Mg/Ca ratio, fluid [CO 2− 3 ], pH, T , and ν 1 peak height (Fig. 3; Table 1).However, these correlations were all weaker than the correlation between ν 1 FWHM and Ar , and in multivariate models combining Ar with either Mg/Ca, [CO 2− 3 ], pH, T , or ν 1 peak height, the only sig- nificant variable was Ar (Table 1).Together, this indicates that Ar is likely the variable controlling ν 1 FWHM, and that the correlations between ν 1 FWHM and other factors may arise as artefacts of the correlations between those variables and Ar .The relationship between Ar and ν 1 FWHM is shown in Table 2 for measured and true FWHM on both the Horiba and WITec instruments (see Supplement for alternate calibration where Ar is fit to ν 1 FWHM).Uncertainty of the calibration propagates to an accuracy of derived Ar of ±∼ 2 for Ar < 20, but increases to ±∼ 6 at Ar = 30 due to both the logarithmic shape of the relationship and the distance from the regression centre.The distributions of residuals were not significantly different from normal distributions (Kolmogorov-Smirnov test, p > 0.8).FWHM data for each experiment are listed in Supplement Table S2.Some uncertainty exists in defining the Ar associated with each Raman spectrum, due to the variability of Ar over the course of each abiogenic experiment.We conducted a Monte Carlo simulation using the data from the Horiba instrument to evaluate the variability in ν 1 FWHM that is expected to arise from the uncertainty in Ar .In each of 10 4 Monte Carlo iterations, we added random error to the mean  ν 1 FWHM residuals (the difference between the predicted and calculated ν 1 FWHM in each iteration) was 0.24 cm −1 , which is 50 % greater than the observed standard deviation of the residuals (0.16 cm −1 ).Since the simulated variability exceeds the observed variability, one or more of the following is implied: (1) most (or all) of the scatter in ν 1 FWHM around the regression line is explained by uncertainty of Ar , (2) the Raman spectra integrate signal from multiple crystals that together approximate the mean Ar during each experiment, and/or (3) variability in Ar is overestimated.Regardless, the data are consistent with a tight, logarithmic dependence of ν 1 FWHM on Ar , with scatter in Ar adding uncertainty to the observed relationship.Analysis of 440 spectra collected from various grains of JCp-1 with the WITec instrument produced a mean measured ν 1 FWHM of 3.51 with 1σ of ±0.09 cm −1 (true ν 1 FWHM of 2.99 ± 0.11 cm −1 ).Applying the calibration equation in Table 2 to calculate Ar for each of the 440 measurements of ν 1 FWHM gave a mean derived Ar of 12.3, with a standard deviation of 2.1, and a standard error of the mean of 0.3.
In the Havannah Island coral, measured ν 1 FWHM changed seasonally by ∼ 0.15 cm −1 .Using our abiogenic calibration (Table 2) to estimate calcifying fluid Ar , these seasonal changes in FWHM translate to approximately 3 Ar units before the 1998 thermal stress event to approximately 1-2 Ar units afterwards.
Measured ν 1 FWHM of the cultured Acropora ranged from mean of 3.45 (±0.017 standard error) cm −1 for the control treatment to 3.366 (±0.009 standard error) cm −1 for the "+CO 2 + T " treatment.Derived Ar (from the abiogenic calibration) was significantly different among treatments.Tukey's test revealed significant decreases in Ar as both T (p < 0.001) and CO 2 (p = 0.045) increased.There was no significant interactive effect of T and CO 2 .K Mg/Ca D determined from the abiogenic aragonites was significantly positively correlated with Ar (Fig. 4; Supplement Table S1; r 2 = 0.82 for experiments conducted at 25 • C): The key finding of the K Mg/Ca D data for this study is the sensitivity to Ar at a single temperature.This dependence of K Mg/Ca D on Ar or precipitation rate was predicted by Gaetani and Cohen (2006) on the basis of the surface entrapment model proposed by Watson (2004), and was also observed in a recent abiogenic aragonite study (AlKhatib and Eisenhauer, 2017) (Fig. 4).Although our K  ].While some apparent correlations exist, statistical models (Table 1) imply that Ar is the variable controlling ν 1 FWHM, and that the patterns observed here may be artefacts of correlations between these factors and Ar .

Controls on aragonite ν 1 peak width
We conducted Raman spectroscopy analyses of abiogenic aragonites, precipitated under various carbonate chemistry and temperature treatments as previously described in De-Carlo et al. (2015) and Holcomb et al. (2016).Carbonate chemistry was manipulated in these experiments such that there is some independence between the carbonate system parameters, which allows us to isolate their effects on the Raman spectral peaks.For example, experiments conducted at the same Ar , but different temperatures, allow us to identify that ν 1 FWHM is sensitive to Ar but not to aragonite precipitation rate (G) (Figs. 2 and 3b).Similarly, since Sr/Ca is strongly controlled by temperature in these aragonites (De-Carlo et al., 2015), the lack of a temperature effect on ν 1 FWHM implies that Sr/Ca ratios exert minor or no effects on ν 1 FWHM.Other factors were more difficult to isolate -for instance Ar was correlated with both Mg/Ca solid and fluid [CO 2−  3 ].Yet comparisons among statistical models (Table 1) show that Ar is the only variable that consistently has a significant effect on ν 1 FWHM, regardless of which additional variable is considered in multivariate models.This leads us to conclude that ν 1 FWHM is primarily dependent upon Ar .We also found that the scattering of data around the regression between ν 1 FWHM and Ar can be explained entirely www.biogeosciences.net/14/5253/2017/Biogeosciences, 14, 5253-5269, 2017 by the uncertainty of Ar in each experiment.Further, the absence of any significant correlations between ν 1 FWHM and either ν 1 position or height indicates that the variation of ν 1 FWHM in our study was not the result of instrumental or curve-fitting artefacts.Our ability to isolate the effect of Ar from Mg/Ca solid and fluid [CO 2− 3 ] is best exemplified by the two experiments in which [Ca 2+ ] was elevated in the fluids by addition of dissolved CaCO 3 .Experiments f08 and g13 were conducted with mean fluid Ar of 12 and 10, and [Ca 2+ ] of 12.4 and 10.4 mmol kg −1 , respectively, compared to the mean [Ca 2+ ] range of 6.9-8.7 mmol kg −1 in all other experiments.The addition of Ca 2+ increased Ar , decreased Mg/Ca, and caused precipitation to occur at lower [CO 2− 3 ] than other experiments, thereby establishing some independence among these factors.Over the relatively narrow Ar range of 10-12, which includes f08 and g13 as well as two experiments with lower [Ca 2+ ], there is still a positive correlation between ν 1 FWHM and Ar (Fig. 5a).Conversely, there are no clear patterns between ν 1 FWHM and either Mg/Ca or [CO 2− 3 ] across these four experiments (Fig. 5b-c).While this further demonstrates that Ar is likely the factor directly controlling ν 1 FWHM, more experiments designed to decouple Ar , Mg/Ca and [CO 2− 3 ] will be useful for constraining ν 1 FWHM sensitivities.In addition, whether the stoichiometry of [Ca 2+ ] and [CO 2− 3 ] affects ν 1 FWHM will be an important question to address in future experiments (Nehrke et al., 2007).
The influence of Ar on ν 1 FWHM can be explained by carbonate ion disorder within the aragonite lattice, as originally hypothesized by Bischoff et al. (1985).Increasing Ar causes more defects and/or impurities to be incorporated into the aragonite, increasing the positional disorder of carbonate in the lattice.The absence of a clear Mg/Ca effect on ν 1 FWHM (discussed further below) suggests that our observed changes in ν 1 FWHM are related more to disorder in the lattice (i.e. more amorphous-like aragonites formed at higher Ar ) than to trace element impurities.Disordered carbonate ions have a distribution of C-O bond strengths and vibrational frequencies, which results in wider Raman peaks (Bischoff et al., 1985;Urmos et al., 1991;Perrin et al., 2016).
Raman spectroscopy studies of abiogenic CaCO 3 have so far focused on calcite, vaterite, and amorphous CaCO 3 (ACC) precipitated with wide ranges of Mg/Ca (from 0 to 45 mol %) (Bischoff et al., 1985;Wehrmeister et al., 2009;Wang et al., 2012;Perrin et al., 2016).These studies manipulated the Mg/Ca ratios of the solutions from which the precipitates formed, and they consistently reported strong correlations between ν 1 FWHM and ν 1 position, both of which increased with progressively higher Mg/Ca in the solid phase.The abiogenic precipitates used in our study are different in that (1) they are aragonite, and (2) most of them (26 out of 28) were grown from the same parent solution (seawater) but under different carbonate chemistry treatments (De-Carlo et al., 2015;Holcomb et al., 2016).Most of the Mg/Ca variability in these aragonites arises from the sensitivity of Mg/Ca partitioning to Ar (Fig. 4), not manipulation of Mg/Ca in the initial fluid.Recognizing this methodological distinction is important for understanding why we did not find a correlation between ν 1 FWHM and position.Firstly, aragonite has several orders of magnitude less Mg than high-Mg calcites, potentially dampening the role of Mg in affecting the shape of the Raman peaks in aragonite.Secondly, if the trend between Ar and ν 1 FWHM were driven by Mg content alone, shifts in Raman peaks to higher wavenumbers would be expected, because Mg is lighter than Ca (Bischoff et al., 1985).However, there was no significant correlation between ν 1 position and Ar (p > 0.7), indicating that the mean C-O bond vibrational frequency did not change systematically with Ar .These observations imply that Mg content was not the primary source of lattice defects broadening Raman peaks, an interpretation supported by the absence of a correlation between Mg/Ca and ν 1 FWHM (after accounting for the effect of Ar ; Table 1).The lack of correlation between ν 1 FWHM and position also indicates that ν 1 FWHM was unaffected by crystallite size (Urmos et al., 1991;Zakaria et al., 2008).
That fluid Ar , not Mg/Ca solid , apparently controls ν 1 FWHM in aragonite precipitated from seawater is critical for interpreting the Raman spectra of coral skeletons.Corals transport seawater to the micro-scale site of calcification (Gagnon et al., 2012;Tambutté et al., 2012), and much like in the abiogenic experiments of DeCarlo et al. (2015) and Holcomb et al. (2016), they precipitate aragonite crystals from a seawater-like solution (McConnaughey, 1989).Changes in calcifying fluid Mg/Ca can result from Ca 2+ transport and/or Rayleigh fractionation (Gaetani and Cohen, 2006;Gagnon et al., 2012).Yet Mg/Ca in coral skeletons is generally within the limits of the abiogenic aragonites analysed here (2.5-11 mmol mol −1 ) and, critically, ν 1 FWHM was insensitive to Mg/Ca over this range.While it is still conceivable that Mg/Ca or [CO 2− 3 ] could have some influence on coral ν 1 FWHM, these effects are likely subordinate to the Ar sensitivity based on our abiogenic results.This implies that we can interpret Raman spectra of coral skeletons using the abiogenic sensitivity of ν 1 FWHM to Ar .

Analysis of JCp-1 and relation to skeletal geochemistry
Using Raman spectroscopy, we derived a mean calcifying fluid Ar for the JCp-1 coral standard of 12.3 ± 0.3.JCp-1 is an internationally calibrated geochemical standard, and has been analysed repeatedly for trace element and isotopic composition.Because the geochemistry of the skeleton reflects the calcifying fluid conditions, we can evaluate the accuracy of our Ar estimates by comparing them to the reported geochemical properties of the JCp-1 carbonate.Specifically, here we calculate calcifying fluid calcium concentration ([Ca 2+ ] cf ) using two different sets of proxies        -the first based on Mg/Ca, Sr/Ca, and δ 44 Ca, and the second based on Raman, δ 11 B (a pH proxy), and B/Ca (see Appendix A for details of calculations).Our approach uses experimentally determined partitioning of these trace elements and isotopes between aragonite and seawater, and a simple model in which [Ca 2+ ] cf is initially elevated relative to seawater (Al-Horani et al., 2003) before precipitation from the isolated calcifying fluid (i.e. a closed system) drives Rayleigh fractionation (Gaetani and Cohen, 2006;Gaetani et al., 2011;Gagnon et al., 2012).We then compare these two independent [Ca 2+ ] cf estimates to test whether our Raman results are consistent with JCp-1 geochemistry.The published Mg/Ca, Sr/Ca, and δ 44 Ca of the JCp-1 coral can be explained with initial Ca 2+ enrichment of 9-11 % and precipitation of 48-59 % of the total Ca 2+ .This corresponds to mean [Ca 2+ ] cf of 8.2 ± 0.7 mmol kg −1 (Fig. 6).We then independently estimated [Ca 2+ ] cf by combining our Raman spectroscopy results with boron systematics.The δ 11 B and B/Ca of JCp-1 together imply a calcifying fluid [CO 2− 3 ] of 987 ± 78 µmol kg −1 .To reconcile the boronderived [CO 2− 3 ] with the Raman-derived Ar of 12.3 ± 0.3 requires [Ca 2+ ] cf of 8.3 ± 0.7 mmol kg −1 .This agrees almost exactly with the mean [Ca 2+ ] cf based on the calculations described above for Mg/Ca, Sr/Ca, and δ 44 Ca systematics (Fig. 6d), giving us confidence in the accuracy of our Raman-based Ar results.One implication of these results is that [Ca 2+ ] cf is less than seawater by 11-25 %, and while this finding should be replicated on additional corals, we note that depleted [Ca 2+ ] cf is consistent with trace element variability in deep-sea and surface-dwelling corals modelled using both "batch" and "flow-through" versions of the Rayleigh calculations (Gaetani et al., 2011;Gagnon et al., 2012).Further, our Raman-derived Ar is generally consistent with estimates made on the basis of coral skeleton crystal aspect ratios (Cohen and Holcomb, 2009).

High-resolution time series of calcifying fluid Ar
Analysis of the Havannah Island coral demonstrates that Raman spectroscopy can detect variability in coral calcifying fluid Ar .The pattern of variability in Raman-derived Ar showed some similarities, but also some differences, compared to the boron-derived Ar presented in D' Olivo and McCulloch (2017).In a sampling track that avoided a 1998 partial-mortality scar ("path D"), boron-derived Ar showed annual oscillations of 2-3 units, but one seasonal oscillation was completely missing during 1998/1999 (Fig. 7).Yet in a track directly adjacent to the scar ("path B"), boron-derived Ar showed overall lower values and less variability, compared to path D. Raman-derived Ar closely tracked variability of path D from 1996 to early 1998, then diverged from the boron-derived Ar between mid-1998 and mid-1999, before returning to similar seasonal variability by 2000.
There are several possible reasons for the differences between Raman-and boron-derived Ar .The first is that the Raman sampling path was located approximately halfway between boron paths B and D, and thus there could be differences in Ar across the skeleton.The boron data indicate lower Ar near the partial-mortality scar, while the Raman data potentially reflect an intermediate level of bleaching stress between the low values of path B and the higher values of path D. There is also potential for the existence of secondary precipitates near the scar, although the Raman spectra clearly indicated that only aragonite was present with no sign of calcite.Second, the Raman measurements have higher spatial resolution than the boron measurements, potentially allowing us to capture variability not observed in the boron sampling paths.Finally, boron-derived estimates of Ar are based on the assumption that [Ca 2+ ] cf remains at, or close to, seawater concentrations, and therefore any substantial changes in [Ca 2+ ] cf will not be recorded in the boron-derived time series.This means that the decoupling www.biogeosciences.net/14/5253/2017/Biogeosciences, 14, 5253-5269, 2017 between Raman-and boron-derived Ar from path D during 1998-1999 could reflect changes in [Ca 2+ ] cf (Fig. 7).D 'Olivo and McCulloch (2017)  ] cf is a consistent response to thermal stress, the ability to detect such changes by using Raman and boron systematics in tandem highlights their combined utility.

Effects of temperature and CO 2 on Ar of juvenile Acropora
Our analyses of juvenile Acropora indicate that calcifying fluid Ar decreased in response to elevated temperature (3 • C above the maximum monthly mean water temperature) and to elevated CO 2 (Fig. 8).In micro-computed tomography (micro-CT) analyses of corals from the same experiment, Foster et al. (2016) showed that elevated CO 2 increased skeletal porosity and decreased total calcification, while elevated temperature had the effect of partially mitigating the response to elevated CO 2 .Thus, the effect of elevated CO 2 on Raman-derived Ar is consistent with the calcification response, but thermal stress decreased Ar without having a negative impact on calcification.One possibility is that the effect of decreasing Ar on aragonite accretion was balanced by elevated temperature, because aragonite crystal growth rates increase with both Ar and temperature (Burton and Walter, 1987).Further, the physiological response to elevated temperature could be complex, potentially involving decreases in Ar balanced by increases in calcifying time or surface area.Finally, Foster et al. (2016) found pitted skeletal surfaces indicative of dissolution, which may have compounded the measured CO 2 effects on calcification and led to their dominance over any apparent effect of temperature in the "+CO 2 + T " treatment.A recent study employing a similar experimental design found that juvenile Acropora B/Ca and U/Ca increased with elevated CO 2 (Wu et al., 2017) based on abiogenic partitioning (DeCarlo et al., 2015;Holcomb et al., 2016).However, while elevated temperature increased δ 11 B (i.e.increased pH), it did not affect B/Ca or U/Ca (Wu et al., 2017), potentially reflecting the lower degree of thermal stress (culture temperatures were within the seasonal range at the collection site) compared to Foster et al. (2015) and/or that [Ca 2+ ] cf played an important role in our Raman data.Nevertheless, our results further highlight that Raman spectroscopy can detect changes in coral calcifying fluid Ar , and that this information is complementary to other analyses, including boron systematics (Sect.4.3) as well as CT-derived porosity and calcification.

Conclusions and outlook
We tested whether Raman spectroscopy records the saturation state of the fluid from which aragonite precipitates.In abiogenic aragonites with known fluid conditions, we found a clear dependence of Raman peak width on Ar , an observation that is not confounded by other factors, including temperature, Mg/Ca, pH, and [CO 2− 3 ].In the JCp-1 coral standard, for which calcifying fluid Ar is not known directly, Raman-derived Ar is consistent with multiple, independent lines of geochemical evidence based on Mg/Ca, Sr/Ca, B/Ca, δ 11 B, and δ 44 Ca.These results support the use of Raman spectroscopy as a proxy for coral calcifying fluid Ar .While it remains theoretically possible for other fac- tors such as trace element variability to have some influence, our Raman results from abiogenic aragonites and the JCp-1 coral can be explained by Ar alone.Nevertheless, future investigations of controls on Raman spectra of corals skeleton will be useful, for instance in checking for positive correlations between ν 1 FWHM and wavenumber that would potentially indicate an effect of Mg/Ca (Bischoff et al., 1985;Perrin et al., 2016;Borromeo et al., 2017).We found no such relationships in either our Porites or Acropora analyses (p > 0.05, r 2 < 0.01 in both cases).
Our approach could potentially be applied to other marine calcifiers that build aragonitic shells and skeletons, such as sclerosponges and some molluscs.However, the relationship between Raman peak width and is likely to be mineralspecific.For example, calcite often contains magnesium with concentrations several orders of magnitude greater than in aragonite, which is likely the reason for the wider peaks in Raman spectra of calcite (Urmos et al., 1991).In contrast, we have shown that Raman spectroscopy is well suited for analysis of coral calcifying fluid Ar .Our initial applications to corals revealed annual cycles of Raman-derived Ar in a field-collected Porites and Ar responses to elevated temperature and CO 2 in cultured Acropora, highlighting that Raman spectroscopy is indeed capable of detecting changes in coral calcifying fluid Ar .Additionally, the high spatial resolution (< 1 µm) and rapid sample processing (≤ 1 s per spectrum) of Raman are currently unrivalled.Further, the same Raman spectra used to estimate Ar can confirm that the sample is aragonite, effectively eliminating the possibility of contamination from other mineral phases.Finally, our analysis of JCp-1 suggests that the Raman ν 1 FWHM may be an accurate proxy of fluid Ar , making it a complementary approach to B/Ca and δ 11 B because combining information from the two approaches enables us to calculate the full carbonate system via [CO 2− 3 ], pH, and possibly now [Ca 2+ ] cf .The Ar of 12.3 that we derived for JCp-1 corresponds to an instantaneous inorganic (Burton and Walter, 1987) aragonite precipitation rate of 0.2 g cm −2 yr −1 .This result is somewhat surprising because the average annual calcification rate of Porites living at 25 • C is typically 1.3 g cm −2 yr −1 (Lough, 2008), a factor of 6.5 faster than expected on the basis of the inorganic calcifying fluid Ar systematics.Since coral calyces have complex 3-D shape such that their skeletal surface area is actually much greater than the planar area of the colony surface (Barnes, 1970), this implies that calcifying surface area is an important factor in determining coral calcification rates (D' Olivo and McCulloch, 2017) and it complicates the use of bulk calcification rates for estimating Ar (Raybaud et al., 2017).Multiple studies using micro-CT have shown increases in porosity and surface-area : volume ratios in corals cultured under elevated CO 2 levels (Tambutté et al., 2015;Foster et al., 2016).This may represent a strategy for controlling calcifying surface area (i.e. the area over which crystals are precipitating from the calcifying fluid at a given time) to maintain bulk calcification rates and partially offset the impacts of ocean acidification.
In summary, recent studies and technological advances in boron isotope measurements (McCulloch et al., 2014), CT (DeCarlo and Cohen, 2016;Foster et al., 2016) and Raman spectroscopy (this study) are now making it increasingly feasible to quantify coral calcification responses with a multipronged approach.Future investigations combining Raman spectroscopy with B/Ca, δ 11 B, and porosity measurements on corals grown under elevated CO 2 will be valuable for understanding mechanisms of resilience to ocean acidification.These include pH homeostasis, maintenance of high calcifying fluid Ar by [Ca 2+ ] or [CO 2− 3 ] regulation, and increases in calcifying surface area as potential strategies for coral calcification to persist in a high-CO 2 world.Evaluating their relative importance in the face of both increasing reef-water temperature and declining [CO 2− 3 ] will be critical to informing our predictions of where and when resilience may be found.

Appendix A: Details of JCp-1 geochemistry calculations
The JCp-1 coral grew at an average temperature of 25 • C (Okai et al., 2002), and the mean salinity in the area is 34.54 (Levitus, 2010).At 25 • C, the K Sr/Ca D is 1.123 (DeCarlo et al., 2015), and at 25 • C and Ar of 12, K Mg/Ca D is ∼ 6.5 (±0.5)× 10 −4 (Fig. 4; in agreement with the mean value of AlKhatib and Eisenhauer (2017) at < 20 Ar and within error of our regression).We calculated a range of possible skeletal Sr/Ca and Mg/Ca ratios based on various combinations of Ca 2+ pumping (enrichment of initial [Ca 2+ ] cf relative to seawater between 0 and 30 %) and Rayleigh fractionation (0-100 % of total Ca 2+ precipitated), and we compared the results to the known Sr/Ca (8.838 ± 0.042 mmol mol −1 ) and Mg/Ca (4.199 ± 0.065 mmol mol −1 ) ratios of JCp-1 (Fig. 6) (Hathorne et al., 2013).For δ 44 Ca, we used the abiogenic aragonite fractionation factor (Gussone et al., 2003(Gussone et al., , 2005)), seawater δ 44 Ca = 1.88 ‰ (Hippler et al., 2003), Ca added via Ca-ATPase (i.e.[Ca 2+ ] cf enrichment) of 0.68 ‰ (Inoue et al., 2015), and an average coral skeleton δ 44 Ca of 0.73 ± 0.1 (Inoue et al., 2015;Chen et al., 2016;Gothmann et al., 2016).Our calculation scheme utilized the Rayleigh equation where i/j is Mg/Ca, Sr/Ca, or 44 Ca / 40 Ca, f is the fraction of Ca remaining; D is K A similar result is achieved when considering the calcifying fluid as semi-closed and using the steady-state "flowthrough" model of Gagnon et al. (2012).We used the same K Mg/Ca D and K Sr/Ca D listed above and the equations of Gagnon et al. (2012) to solve for their "γ " term (defined as the ratio of calcium pumping to precipitation) that best fit the JCp-1 data (0.35).Next, we found the unique value of " P kzρ " (3.1 mmol Ca kg −1 ; where P is precipitation flux, k is seawater exchange rate, z is the ratio of calcifying volume to surface area, and ρ is seawater density) that was consistent with this γ and the JCp-1 Mg/Ca and Sr/Ca.Finally, using these terms matched to JCp-1 geochemistry, we follow the equation given in Gagnon et al. (2012): The [Ca 2+ ] cf calculated by this method is 8.1 mmol kg −1 , within error of the 8.2 ± 0.7 mmol kg −1 derived from our "batch" model calculations above (see Fig. 6).
In the first set of calculations with Mg/Ca, Sr/Ca, and δ 44 Ca, we assume only that Ca 2+ is added to the fluid, whereas in the second set of calculations with Raman spectroscopy and boron systematics we assume carbonate chemistry is modified from seawater.However, since we do not directly connect the initial Ca 2+ enrichment of the fluid with the carbonate system calculations, we do not need to make assumptions regarding the stoichiometry of any potential proton-Ca 2+ exchange.

Figure 2 .
Figure 2. Sensitivity of true ν 1 FWHM to Ar .The x axis represents the mean, and the horizontal error bars represent 1 standard deviation, of Ar while aragonite precipitated in each experiment.The crosses indicate individual spectra, the filled points indicate means of each experiment, and the vertical error bars represent the standard error of replicate spectra.Colours indicate experiments conducted at various temperatures, and black circles indicate the two experiments conducted with elevated [Ca 2+ ].Grey shading represents the standard error of the curve fit to the individual spectra, and the dashed grey lines represent the standard error of prediction.
to those ofAlKhatib and Eisenhauer (2017), their K Mg/Ca D are systematically lower.This offset is potentially explained by the different media used in the experiments: filtered seawater inDeCarlo et al. (2015) andHolcomb et al. (2016) as opposed to ammonium carbonate solutions inAlKhatib and Eisenhauer (2017).

Figure 3 .
Figure 3. Scatter plots of Raman true peak width (ν 1 FWHM) and (a) aragonite Mg/Ca, (b) aragonite precipitation rate, (c) experimental fluid [CO 2− 3 ], (d) experimental fluid pH, (e) Raman peak intensity, and (f) Raman peak position.Colours indicate experiments conducted at various temperatures, and black circles indicate the two experiments conducted with elevated [Ca 2+].While some apparent correlations exist, statistical models (Table1) imply that Ar is the variable controlling ν 1 FWHM, and that the patterns observed here may be artefacts of correlations between these factors and Ar .

Figure 4 .
Figure 4. Sensitivity of K Mg/Ca D to Ar at 25 • C. Horizontal error bars represent 1 standard deviation of Ar while aragonite precipitated in each experiment.Grey shading represents the standard error of the curve, and the dashed grey lines represent the standard error of prediction.The grey squares show the AlKhatib and Eisenhauer (2017) data replotted against Ar (see their Fig. 4 for a plot against precipitation rate).

Figure 5 .
Figure 5. Isolation of Ar , Mg/Ca and [CO 2− 3 ] controls on true ν 1 FWHM based on manipulation of fluid [Ca 2+ ].The same data are plotted here as in Figs. 2 and 3, but only over the Ar range of 10-12 because it includes two experiments initiated with seawater [Ca 2+ ] (grey) and two experiments with elevated [Ca 2+ ] (black).With fluid [Ca 2+ ] manipulation, ν 1 FWHM is still sensitive to Ar (a), but not to Mg/Ca (b) or [CO 2− 3 ] (c).

Figure 6 .
Figure 6.Evaluation of Raman-derived Ar based on geochemistry of the JCp-1 coral.In each panel, the x axis f is fraction of total [Ca 2+ ] cf remaining, and the y axis is initial [Ca 2+ ] cf enrichment relative to seawater.Colours indicate calculated (a) skeletal Sr/Ca, (b) skeletal Mg/Ca, (c) skeletal δ 44 Ca, and (d) [Ca 2+ ] cf .In (a-c) the grey lines bound the reported values of JCp-1 (±1σ ), and the black polygon shows the parameter space consistent with Sr/Ca, Mg/Ca, and δ 44 Ca.Panel (d) shows [Ca 2+ ] cf and the grey lines show the constraints from panels (a-c).The dashed black line indicates the [Ca 2+ ] cf independently estimated based on combining our Raman data with published boron data (see text for details of the calculations).The two approaches are within error (i.e. the dashed black line intersects the solid black polygon) for Raman-derived Ar between 10.8 and 13.7, consistent with the observed Ar of 12.3 ± 0.3.

Figure 7 .
Figure 7. Time series of boron-and Raman-derived Ar in a coral from Havannah Island.(a) X-ray image of central slab cut from the core.The black line indicates the Raman sampling path.Boron paths B and D were located approximately 1 cm into and out of the page, respectively (see Fig. 1 in D'Olivo and McCulloch (2017) for details).(b) The red and blue lines show boron-derived Ar assuming a constant [Ca 2+ ] cf .Raman-derived Ar is plotted in black with grey error bars indicating 1σ .The means of the two data sets (mean of Raman Ar was 9.5) have been removed to facilitate comparison and because of uncertainty of absolute [Ca 2+ ] cf for the boron calculations.

Figure 8 .
Figure 8. Raman-derived Ar of Acropora cultured under different temperature and CO 2 treatments.The grey bars indicate treatment means and error bars represent ± standard error of the mean.Significant effects were observed for both temperature and CO 2 .
44−40 ; and it is assumed that Mg and Sr concentrations in the initial fluid prior to precipitation are equal to those of seawater.

Table 1 .
Linear model statistics for fits to ν 1 FWHM.: none of the multivariate models were significantly different from the Ar model.a 1.77 degrees of freedom for univariate models, and 2.76 degrees of freedom for multivariate models; except for ν 1 position, which had 1.59 degrees of freedom.b Standard deviation (1σ ) of the ν 1 FWHM residuals from the model fit. Notes

Table 2 .
Regression equations between mean ν 1 FWHM and Ar .
Notes: equations describe ν 1 FWHM = slope×ln( Ar ) + intercept, where the regressions were performed on the mean values of each experiment.Parentheses indicate 1 standard error.