Geochemical and microstructural characterisation of two species of cool-water bivalves (Fulvia tenuicostata and Soletellina biradiata) from Western Australia

Geochemical and microstructural characterisation of two species of cool-water bivalves (Fulvia tenuicostata and Soletellina biradiata) from Western Australia. Liza M. Roger, Annette D. George, Jeremy Shaw, Robert D. Hart, Malcolm Roberts, Thomas Becker, Bradley J. McDonald and Noreen J. Evans. 5 School of Earth Sciences, The University of Western Australia, Crawley, 6009, Australia Centre for Microscopy, Characterisation and Analysis, The University of Western Australia, Crawley, 6009, Australia Department of Chemistry, Nanochemistry Research Institute, Curtin University, GPO Box U1987, Perth, 6845, Australia Department of Applied Geology, John de Laeter Centre, TIGeR, Curtin University, Bentley, 6102, Australia


Introduction
As calcifiers, molluscs play an important role in the ocean carbonate cycle. The calcium carbonate formed by marine organisms is a complex geochemical source and sink of carbon, which controls total oceanic carbon content and pCO 2 (partial pressure of CO 2 ) and more generally contributes to ocean alkalinity (e.g. Gattuso and Hansson, 2011). The biomineralisation process or, more precisely, the elemental and isotopic composition of biogenic CaCO 3 , is affected by physiological processes (e.g. Freitas et al., 2006;Gillikin et al., 2005b;Lowenstam and Weiner, 1989) and environmental conditions at the time of deposition (e.g. Ferguson et al., 2011;Gazeau et al., 2010;Hahn et al., 2014;Heinemann, 2011;Henkes et al., 2013;Lowenstam and Weiner, 1989;Schöne et al., 2011). Molluscs are globally distributed and demonstrate sequential growth, thus providing high-resolution seasonal and sub-seasonal records of environmental conditions (e.g. Jones and Quitmyer, 1996).
A number of bivalve species are extremely long lived, with lifetimes of many decades or even centuries, e.g. fresh water pearl mussels (e.g. Schöne et al., 2004), geoduck clams (e.g. Strom et al., 2004), ocean quahogs (e.g. Schöne et al., 2005) and giant clams (e.g. Ayling et al., 2006;Bonham, 1965). Many oceanic conditions, such as seawater surface temperature, productivity, circulation and carbon reservoir dynamics, have been reconstructed successfully using bivalve shell records (e.g. Klein et al., 1996;Lazareth et al., 2003;Richardson et al., 2004;Wanamaker et al., 2011). However, an organism's physiological control on the chemical composition of its shell can affect the reliability of the parameters recorded (e.g. Gillikin et al., 2005b;Lazareth et al., 2013Lazareth et al., , 2003. Differences between the measured isotopic and the elemental composition in carbonate materials are typically attributed to a physiological or biological influence commonly called the "vital effect" (Zeebe et al., 2008). Spero et al. (1991) first attempted to describe the vital effect through the quantitative modelling of stable carbon isotopes in foraminifera. The influence of the vital effect is variable between species, which is problematic for the development of a unique geochemical equilibrium model for seawater property reconstructions.
Long-term records of seawater parameters (e.g. temperature, alkalinity and pH) are essential for understanding past climatic and oceanic changes in the current context of global climate change. Data from diverse geographical locations are needed to develop a global understanding of those changes. Records from bivalve shells have contributed considerably to our knowledge, but environmental controls need to be well defined and differentiated from biological and physiological factors (e.g. Jacob et al., 2008).
This study focuses on the microstructure and composition of shells from two species of bivalves endemic to southwestern Australia to gain insight into the physio-chemical processes involved in molluscan shell calcification. It constitutes the first step towards monitoring global climate change through biogenic carbonate in southwestern Australia. In general, limited geochemical data are available for temperate Australian marine bivalves and, to our knowledge, this work is the first multimodal (crystallographic, spectroscopic and geochemical) study focusing on marine bivalve species from Australian waters. The structure and mineral composition of shells from the Western Australian bivalves F. tenuicostata and S. biradiata have not been previously investigated as a potential proxy for future studies using this integrated approach. Given the absence of corals along the southern coast of Western Australia and the south coast of Australia as a whole, bivalve shells could serve as an alternative proxy for generating records of historical ocean chemistry provided that the overarching mechanisms are well defined and understood (e.g. elemental equilibrium and the vital effect). This study contributes to the global understanding of how marine bivalves record environmental change through shell growth and the processes governing marine biogenic carbonate production.
2 Materials and methods

Study area and marine setting
Samples were collected from the muddy sands (600-800 µm grain size) of King George Sound at a depth of 3-5 m on the southern coast of Western Australia near the town of Albany ( Fig. 1). This area experiences seasonal freshwater inputs, a temperate to Mediterranean climate, a mean annual rainfall of 931.5 ± 0.2 mm (between 1877 and 2014), a maximum of ∼ 126.1 ± 0.2 mm of rainfall in August and a minimum of ∼ 30.1 ± 0.2 mm in December (Australia Government Bureau of Meteorology; Brouwers et al., 2013;Klausmeyer and Shaw, 2009). The main currents influencing the area are the Capes Current (coastal current) flowing westward, the Leeuwin Current (shelf current) flowing eastward and the Flinders Current (oceanic current) flowing westward (Cresswell and Golding, 1980). Further south, Australian waters meet the cold and nutrient-rich Antarctic waters. The southwestern coastline of Australia is impacted by strong winter storms influenced by pressure systems over the Great Australian Bight, as well as the Leeuwin Current dynamics. Seawater surface temperatures (SST) vary annually between 16 and 21 • C (August-September to April-May) and bottom temperatures vary between 15 and 20.6 • C (Australian Government Bureau of Meteorology) at 40-50 m of maximum depth.

Species
The study uses two species of bivalves: Fulvia tenuicostata (Lamarck, 1819) (the "thin-ribbed cockle") and Soletellina biradiata (Wood, 1815) (the "double-rayed sunset clam"; Fig. 2). Both species are native and endemic to southwestern Australia (WoRMS, 2016). F. tenuicostata (Mollusca: Bivalvia: Veneroida: Cardiidae), shown in Fig. 2a and b, is typically 50-55 mm in length (anterior to posterior margin). The shells of this species are cream coloured, becoming yellow towards the margin (the periostracum can show staining from the surrounding mud). The shells are thin and delicate with narrow ribs (Fig. 2a and b). This genus is found in moderately sheltered areas of muddy sand (WoRMS, 2016). The shell structure of F. tenuicostata consists of a prismatic outer layer (OL), a simple crossed-lamellar middle layer (ML) and a complex crossed-lamellar inner layer (IL). S. biradiata (Mollusca: Bivalvia: Veneroida: Psammobiidae) shells, shown in Fig. 2c and d, may measure up to 70 mm. The shells are thin walled and have an elongated shape. With the thick periostracum removed, the shells are cream coloured with tinges of pink and purple and two distinctive pale radial rays extending from the umbo to the growing edge. S. biradiata is also found in moderately sheltered areas with muddy sand substrates. The shell structure of S. biradiata consists of a prismatic OL composed of acicular prismatic crystals directed toward the outer surface and locally curved, a simple crossed-lamellar ML and a complex crossed-lamellar IL. The OL may be absent, in which case the lamellae of the ML extend to the outer surface as noted for other species of this subfamily (Popov, 1986;Schneider and Carter, 2001).

Sample preparation
Shells (five F. tenuicostata and three S. biradiata) were rinsed with fresh water, manually cleaned of any organic residues by scrubbing, then soaked in sodium hypochlorite (12.3 vol%) for approximately 12 h. Shells were thoroughly rinsed with Milli-Q water (EMD Millipore Corporation, Darmstadt, Germany) and then air dried. According to Krause-Nehring et al. (2011), the treatment of biogenic calcium carbonate to remove organic matter may have an impact on the elemental and crystallographic composition in the bivalve Arctica islandica. The effects of sodium hypochlorite (NaOCl) pretreatment (also used in our study) on A. islandica shell powder were insignificant compared to other treatments tested in that study. Indeed, Sr / Ca and Mg / Ca ratios did not change significantly, and very little to no dissolution or phase change was observed. NaOCl treatment has also proved to be the most efficient method to remove organic matter with the least effect on structure and composi- biradiata) as seen in a thin section. Powder was sampled from the ventral margin for X-ray diffraction analysis (the white dashed lines and arrows in a, b, c and d). The area adjacent to the ventral margin was then targeted for confocal Raman microscopy, LA-ICP-MS and EPMA (the red box in e). The lower panel shows a representation of the laser spots along five transects (T1-T5). T1-T3 are parallel to the shell edge, and T4 and T5 are parallel to the external and internal shell margin (respectively). tion in coral cores (Nategaal et al., 2012). Krause-Nehring et al. (2011) performed their study on powdered, pre-cleaned shells (manual removal of the periostracum) and found very little effect. In our study, where intact shells were pretreated with NaOCl, we suggest that the geochemical and crystallographic composition of the bivalves studied here was not significantly altered despite having some hypochlorite solution percolate through.
Shells were sampled along the growing edge using a handheld dental drill for X-ray diffraction analysis (Fig. 2). Fifty milligrams of very fine powder was mixed with ∼ 10 mg of CaF 2 (20 % by weight) and mounted on a low-level background holder according to the settings described in the Supplement S1. The bivalves were then cut along the axis of maximum growth (from umbo to ventral margin), fixed in styrene polymer and mounted on a glass slide with EPO-TEK 301 epoxy glue (Billerica, MA, USA). They were then trimmed and polished to a thickness of 20-40 µm. The thinsectioned bivalves, thus prepared, were used for petrographic observations, confocal Raman spectroscopy and laser ablation mass spectrometry work. All the analyses described below targeted the ventral margin of the shells, which is the youngest material. The growth increments of F. tenuicostata are about twice those of S. biradiata at 1 and 0.5 mm, respectively. Each measurement was made to encapsulate approximately the same growth period.

X-ray diffraction analysis
X-ray diffraction (XRD) analysis of the powdered shells was performed using a PANalytical Empyrean diffractometer (Eindhoven, The Netherlands) operated at 40 kV and 40 mA (Cu Kα radiation and Ni beta filter) and a PANalytical PIXcel position sensitive detector. The particular settings are described in the Supplement S1. A corundum NIST standard (676a; certificate number 382200) was used as a primary external standard and calcium fluoride (CaF 2 ) was used as a secondary internal standard. The amount of the secondary standard was tested by analysing different CaF 2 : sample ratios. Since no major differences were observed between the various ratios tested, a 20 % weight of CaF 2 was chosen. CaF 2 (lattice parameter a = b = c = 5.4662(2) Å and α = β = γ = 90 • ) is an ideal internal standard because its diffraction peaks do not interfere with those of CaCO 3 .
The mineral phases were identified using the Inorganic Crystal Structure Database (ICSD) and the International Centre for Diffraction Data (ICDD). The Rietveld refinement technique was used to quantify the phases. The Scherrer equation (Cullity and Weymouth, 1957) was used to calculate the coherently scattering domain (CSD) size (i.e. crystallite size) using the dominant reflections of calcite (104) and aragonite (111). Both the Rietveld refinement and the Scherrer calculation were performed using the PANanlytical HighScore Plus software (version 3.0e). The non-crystalline fraction, or amorphous material, in the powder was typically calculated from the overestimation of the secondary standard during the Rietveld refinement (Scarlett and Madsen, 2006).

Confocal Raman microscopy
Confocal Raman microscopy (CRM) uses intense monochromatic light (here, an infrared laser) to irradiate samples. The subsequent molecular vibrations produce a highly specific Raman spectrum that reflects the molecular structure and the chemical identity of the samples. Changes in the intensity and peak position indicate different vibrational modes of a molecule and therefore may reveal subtle differences in the crystalline form, especially between the different polymorphs of CaCO 3 (Nehrke and Nouet, 2011). CaCO 3 peaks measured using CRM are caused by the unit cell symmetry of the crystals and the molecular carbonate ions (CO 2− 3 ). Because of their shared CO 2− 3 -related vibrational modes, the differentiation between aragonite and calcite was accomplished using the lower end of the spectrum, 50 to 1200 cm −1 . In CRM, a spectroscopy system is coupled with a confocal microscope, allowing for the mapping of samples with chemical sensitivity.
CRM was undertaken on thin sections of bivalve shells to confirm XRD-determined mineral phases and to provide key information about the crystallographic structure of CaCO 3 . To mitigate the high fluorescence levels caused by a green excitation laser (wavelength 532 nm), a diode infrared laser module (wavelength 785 nm) with 20 × objective and a 0.4 numerical aperture was used, coupled with the Raman imaging system (WITec, Ulm, Germany; alpha300 RA+). CRM was performed in reflection mode, whereby the scattered light is collected through a 100 µm detection fibre connected to a UHTS 300 spectrometer equipped with a 300 lines/mm grating. The full Raman spectra at each imaging pixel were acquired using a camera with a thermoelectrically cooled, back-illuminated CCD chip. The first-order Raman peak of silicon (520.2 cm −1 ) was used to optimise the alignment and signal intensity before each analysis. Measurements were made as large area scans with the accumulation of three to five spectra to minimise error (±1 cm −1 ) and an integration time of 0.1 to 0.050 s. Two spectral ranges were measured: 0 to 1600 cm −1 and 1600 to 3000 cm −1 ; the second range was explored in an attempt to detect organic functional groups (C−H or C=O) within the crystalline structures. The CRM data were processed and analysed with the WITec Project FOUR software.

Electron probe microanalysis
Quantitative chemical maps were acquired on petrographic thin sections using a JEOL 8530F HyperProbe (JEOL Solutions for Innovation, Peabody, MA, USA) equipped with five tuneable wavelength dispersive spectrometers. The operating conditions for the instrument calibration were comprised of a 40 • take-off angle, a beam energy of 15 kV and a beam current of 20 nA. The beam was defocussed to 2 µm. The elements were acquired using analysing crystals: PET (pentaerythritol) for Ca Kα, S Kα and Sr lα; TAP (thallium acid phthalate) for Na Kα and Mg Kα. The standards used for the instrument calibration were barite for S Kα, calcite for Ca Kα, celestite for Sr lα, periclase for Mg Kα and jadeite for Na Kα. Peak counting times were 20 s and mean atomic number background corrections were used throughout (Donovan and Tingle, 1996). The results are the average of three points and the detection limits ranged from 0.008 wt % for S Kα to 0.010 wt % for Mg Kα to 0.049 wt % for Sr lα. The Armstrong-Love/Scott algorithm was used for data reduction (Armstrong, 1988). Quantitative element maps were obtained using the Probe Image ® software (Probe Figure 3. The X-ray diffraction patterns of specimens of Fulvia tenuicostata (a) and Soletellina biradiata (b). Each XRD pattern is offset from the next (along the y axis) to show the consistency of the diffraction peak positions. The sample ID is indicated to the right of each diffraction pattern. Intensity is relative. Note: G154, G157 and G172 were not analysed using XRD because the sample weight was too low. The main diffraction peaks indicated for each phase identified aragonite, calcite and calcium fluorite CaF 2 .
Software, Inc., Eugene, OR, USA) for X-ray intensity acquisition. The beam current was 80 nA with a 40 ms per pixel dwell time and a 4 × 4 µm pixel dimension. Image processing and quantification was performed off-line with the CalcImage ® software (Probe Software, Inc.) and output to Surfer ® (Golden Software, LLC, Golden, CO, USA).

Laser ablation ICP-MS
The analysis of carbonate material was undertaken using a Resonetics RESOlution M-50A-LR (Australian Scientific Instruments, Fyshwick, Australia) incorporating a Compex 102 excimer laser (Coherent Inc., Santa Clara, CA, USA) coupled with an Agilent (Santa Clara, CA, USA) 7700s quadrupole ICP-MS on shell material that had already been thin sectioned. Following two cleaning pulses and a 20 s period of background analysis, samples were spot ablated for 30 s at a 7 Hz repetition rate using a 75 µm beam and a laser energy of 5 J cm −2 . Oxide polyatomic interferences were minimised by tuning flow rates for a ThO / Th ratio of < 0.5 %. The sample cell was flushed with ultra-high-purity He (350 mL min −1 ) and N 2 (3.8 mL min −1 ), and high-purity Ar was employed as the plasma carrier gas. The international glass standard NIST 612 and the coral standard MACS-3 were used as the primary reference materials to calculate the elemental concentrations (using stoichiometric 43 Ca as the internal standard element) and to correct for instrument drift on all elements. Standard blocks were run after every 20th unknown sample. The mass spectra were reduced using iolite (Armstrong, 1988). Data were collected on a total of 16 elements measured as 7 Li, 24 Mg, 28 Si, 31 P, 34 S, 44 Ca, 52 Cr, 55 Mn, 56 Fe, 63 Cu, 66 Zn, 75 As, 88 Sr, 111 Cd, 138 Ba and 208 Pb. The Si concentration was used to determine if or when the laser hit the glass slide. The measurements taken when the laser hit the glass were discarded.
Trace element concentration profiles were measured in four shells of F. tenuicostata and two of S. biradiata. Five transects were measured on each shell: three parallel transects of seven to eight spots (T1-T3) oblique to the shell surface, one transect along the outer portion of the outer shell layer (T4; 10 spots) and one along the inner portion of the outer shell surface (T5; 10 spots; Fig. 2). Transects T1-T3 were measured to assess the elemental variation towards the inner shell surface and the potential thickness of the last growth layer deposited. T4 and T5 were measured to compare trace element concentrations in the innermost layer of the shell to the outermost layer. 3 Results

Shell mineral composition
The XRD patterns indicate that the shells are composed of calcite and aragonite based on matching diffraction peaks (Fig. 3). The third identifiable diffraction pattern represents the internal standard, CaF 2 (20 % weight of the total sample). The Rietveld refinement was applied in order to obtain an accurate phase quantification for each sample; however, the values produced by the software corresponding to the weight ratio of CaF 2 added were lower than the actual amount. Consequently, the results are not considered useful. Quantification errors in XRD are commonly caused by residual moisture, large organic content, amorphous crystallites or, in this case, errors due to small sample sizes and insufficient particle size reduction (Scarlett and Madsen, 2006). In the present case, moisture, organic material and amorphous crystals could all cause an overestimation (Scarlett and Madsen, 2006), but the underestimation was likely caused by the small sample size (∼ 50 mg). Accurate quantification on small sample sizes can be obtained by using synchrotron technology. Although the phases cannot be quantified, the lattice parameters and crystallite sizes (i.e. the coherently scattering domain or CSD) produced post-refinement and after the Scherrer calculation can be used because of the addition of the pre-calibrated CaF 2 standard. All of the samples analysed were indexed to the orthorhombic symmetry system for aragonite and the hexagonal system for calcite. The lattice parameters (a, b, c) generated by the Rietveld refinement differ from the geological reference materials for both calcite and aragonite (Table 1). All samples of F. tenuicostata manifest expansion in aragonite and calcite along the a axis (Fig. 4). All samples of S. biradiata, except one (G170), show the same phenomenon. In contrast, G170 shows contraction along all axes. The b axis in both phases and both species also generally shows a positive distortion (stretching) with a few exceptions (F. tenuicostata, G159; S. biradiata, G170 and G173; all three in aragonite only). The calcite c axis shows shrink- Figure 5. The typical Raman spectra from S. biradiata and F. tenuicostata (a). The blue boxes correspond to the filters applied to the data: filter 1 ( peak area) centred at 209 cm −1 (width = 36.6 cm −1 ), filter 2 ( peak area) centred at 282 cm −1 (width = 66.3 cm −1 ), filter 3 (centre of mass, weighted width) centred at 1085 cm −1 (width = 36.6 cm −1 ) and filter 4 ( peak area) centred at 1450 cm −1 (width = 600 cm −1 ). The blue dashed box, filter 5, encapsulates filter 1 and a filter centred at 152 cm −1 ( peak area, width = 36.6 cm −1 ) to represent their calculated peak intensity ratio. The blue and red spectra (b) show the peak intensity difference seen at 152 and 209 cm −1 . Each filter measures the peak area, except filter 3, which measures the full width at half maximum and produces a qualitative map. Filter 1 targeted the aragonitic phase, and filter 2 targeted calcite. Filter 3, by measuring variation in FWHM, represents crystallinity and crystallite order. Filter 4, by targeting the tail of the spectrum, shows fluorescence, since the peaks related to organics tend to show best in the tail of the spectrum. And finally, the peak intensity ratios (filter 5) show changes in the crystallite orientation (aragonite crystallites).
ing in both species. The distortion along the c axis in aragonite is variable (two out of four F. tenuicostata show contraction and two out of three for S. biradiata). The largest distortion was found in calcite, particularly for a and b (+0.12 to +0.02 %); c only shows small distortions of −0.03 to −0.004 %. Aragonite is distorted along the a axis between +0.03 and +0.017 %; the b and c axes manifest smaller distortions of +0.014 and −0.008 % at a maximum, respectively. The main difference between the two species considered here lies in the amount of distortion, whereby F. tenuicostata manifests a smaller distortion compared to S. biradiata (Fig. 4). The maximum distortion in aragonite is 0.02 %. It is 0.066 % in calcite for F. tenuicostata (both a/a) and 0.03 and ∼ 0.13 %, respectively, in S. biradiata (also a/a). Furthermore, although minimal, F. tenuicostata shows generalised stretching (0.004 % maximum) along the c axis in aragonite, whereas S. biradiata shows shrinking (−0.008 % maximum).
The CSD sizes estimated using the Scherrer equation (with K = 0.9) for the four dominant aragonite diffraction peaks corresponding to reflections (111), (102), (201) and (122) reveal aragonite crystals ranging from ∼ 34 to ∼ 144 nm long ( Table 2). The majority of the CSD values fall in the nanocrystal category (< 100 nm), but some, especially those for the (122) plane, belong to the larger ultra-fine category (100-500 nm). The CSD sizes estimated for calcite used one dominant diffraction peak corresponding to reflection (104) to reveal smaller crystallites than in aragonite (∼ 23 to ∼ 29 nm long).
Different phases in the samples were identified using a combination of lattice modes and internal modes. Cal- Table 2. The aragonite and calcite crystallite size (coherently scattering domain, or CSD) in nm, estimated using the Scherrer equation (Cullity and Weymouth, 1957) with K = 0.9 on four main diffraction peaks of aragonite and the peak (104)  cite and aragonite can be identified using the librational modes L c (282 cm −1 ) and L a (209 cm −1 ), respectively, and the in-plane band v 4 (713 cm −1 for calcite and 702, 706 and 717 cm −1 for aragonite); the shared peaks (translational mode, T , ∼ 152 cm −1 and symmetry stretch, v 1 , 1085 cm −1 ) represent the CO 2− 3 ion motion and the C−O bond stretching common to both CaCO 3 polymorphs (Cuif et al., 2012, Fig. 5). Although very small and broad compared to aragonite-specific peaks, only the L c peak identified Figure 6. Images extracted from the filters applied to the Raman spectra (see Fig. 5) from samples G171 (S. biradiata) (ae) and G155 (F. tenuicostata) (f-j). Scale bar: 7000 µm (a-e) and 1000 µm (f-j). Colour scale: the bright colours correspond to high intensities (high CCD counts) and the dark colours correspond to low intensities (low CCD counts). The sub-horizontal and curved lines, most visible in (a-c) and (f-h), are structural features of the shells. Scale bar: 500 µm.
calcite. The most pronounced aragonite peak is L a . The scan range of the spectrometer was increased to 3000 cm −1 in order to detect the potential presence of organic compounds (e.g. Nehrke and Nouet, 2011); however, no features characteristic of these groups were found. As a consequence, the spectra presented here only extend to 2000 cm −1 .
Four filters were applied to each Raman spectrum (Fig. 5) to highlight different phases or reveal particular features of the shell: (1) peak area of aragonite (centred at 209 cm −1 ); (2) peak area of calcite (centred at 282 cm −1 ); (3) centre of mass and weighted width (full width at half maximum) of v 1 (centred at 1085 cm −1 ); and (4) peak area of the tail of the spectrum targeting fluorescence (centred at 1450 cm −1 ). Peak intensities at 152 cm −1 (translational mode T ) and 209 cm −1 were also ratioed (PIR) to distinguish the crystal orientation variations (Figs. 5 and 6).
Aragonite and calcite were both found in all samples analysed using CRM. The small intensity of the calcite peak indicates that it is a minor phase. Two distinct types of spectra are visible for aragonite (Fig. 5), both with identical peak positions but different peak intensity ratios (PIR) (Figs. 5, 6e and j) between the translational mode T and the other peaks characteristic of aragonite (filter 5, Figs. 5 and 6).
The full width at half maximum (FWHM) of the symmetric stretch v 1 (CaCO 2− 3 ) reveals sequential variations or, more precisely, alternating high and low peak widths, especially in sample G154 ( Fig. 6c and h). The other samples show more discrete changes in FWHM. Filter 4 (the filter showing fluorescence) reveals clear banding throughout the samples ( Fig. 6e and j).

Elemental composition
The EPMA results show that the total elemental compositions are consistently between 98 and 100 wt % (Figs. 7a and 8a). Calcium is generally consistent across the samples with an average concentration of ∼ 39.5 wt % (Figs. 8b, 9b). S and Na comprise up to 0.16 and 0.66 wt %, respectively, but importantly show variations in concentration with opposing trends; i.e. high Na corresponds to low S and vice versa (Fig. 6). The concentrations of S and Na correspond to growth lines with overall Na content higher towards the external surface of the shell and S towards the internal surface. Low S and high Na correspond to higher fluorescence levels detected using CRM (1450 cm −1 filter; Fig. 6d and i). These results are consistent for both S. biradiata (Fig. 7) and F. tenuicostata (Fig. 8). Ca is typically uniformly distributed throughout each shell; nevertheless, the patterns revealed by CRM correspond to the patterns on the Ca distribution maps. The patterns revealed by filters 4 and 5 (crystallinity and crystal orientation, respectively), when applied to the Raman spectrum, correspond to those visible on the Ca wt % map. Mg and Sr are below the instrumental detection limit (Table 3) and are not presented here.
The trace element concentration profiles measured using LA-ICP-MS showed that the NIST160 and MACS standard reproducibility is in the range of acceptable values (between 0.5 and 4 % on most elements; Supplement S2). Very little reproducibility was observed between T1, T2 and T3 (Figs. 9 and 10) for all six shells tested; however, compositionally distinct layers were identified, as were differences between species (Figs. 11 and S3). In F. tenuicostata, the ratios of Sr / Ca, S / Ca and Pb / Ca along transects T1-T3 generally increase from the external sur- face inwards (maximum +1 mmol mol −1 , +3 mmol mol −1 and +0.05 µmol mol −1 , respectively; Supplement S3). Only Sr / Ca in G158 decreases by 2 mmol mol −1 inwards (Supplement S3). Mg / Ca, P / Ca and Ba / Ca (Figs. 11 and S3) follow the opposite trend, with a maximum decrease of 1.5 mmol mol −1 , 0.06 mmol mol −1 and 3 µmol mol −1 , respectively. The elemental ratios in S. biradiata all decrease along the T1-T3 transects from the outer to the inner layer (Mg / Ca −0.15 mmol mol −1 , P / Ca −0.04 mmol mol −1 , Li / Ca −4 µmol mol −1 , Ba / Ca −2 µmol mol −1 ), except Pb / Ca (+0.06 µmol mol −1 ) and S / Ca, which is relatively stable at 1.3 mmol mol −1 . Sr / Ca increases inwards (+1.5 mmol mol −1 ) but decreases to its original level at the edge of the internal layer (∼ 2 mmol mol −1 ). Li, Ca, Mn and Ba concentrations in F. tenuicostata are not linearly correlated to the other trace elements measured, but a statistically significant (paired t test) linear correlation was found between = 0.55, p < 2.2 × 10 −16 ). T4-T5 measurements confirm the compositional differences between the external and internal layers (Figs. 12 and S3). Overall, the external layer of the S. biradiata shells studied demonstrates higher ratios of Li / Ca, Mg / Ca, P / Ca, Sr / Ca and Ba / Ca compared to the internal layer. Only S / Ca and Pb / Ca do not follow this pattern, where S / Ca is the same in both layers and Pb / Ca is higher in the internal layer (Figs. 12 and S3). In comparison, F. tenuicostata shells show higher ratios of Mg / Ca, Sr / Ca, S / Ca, Ba / Ca and Pb / Ca in their internal layer compared to their external layer. P / Ca decreases inwards and Li / Ca is essentially consistent.
S and P are indicators of organic content, and their presence in the bivalves reflects a variation in macromolecule content throughout the shells. F. tenuicostata has a higher S-containing macromolecule content in the organic matrix of its internal layer, whereas a higher P / Ca ratio in the outer layer suggests that the organic matrix is richer in Pcontaining macromolecules in that section of the shell. The shells of S. biradiata are visibly more complex. They also show high S / Ca and low P / Ca towards the innermost layer, but for most of the shell, variable S / Ca and P / Ca ratios suggest that the external layers are characterised by alternating P-and S-containing macromolecules in organic components (Figs. 7-8 and 10-11).

Mineralogy and microstructure
Although S. biradiata and F. tenuicostata belong to subfamilies typically described as aragonitic (Schneider and Carter, 2001), the XRD analysis and CRM results reveal a dominant aragonitic composition and a minor calcite fraction. The Rietveld refinement did not allow for the quantification Figure 8. The EPMA elemental map of sample G155, F. tenuicostata, for totals (a), Ca concentration (b), S concentration (c) and Na concentration (d). Mg and Sr were below the instrumental detection limit and are therefore not presented. The dashed line indicates scan overlap, since the sample was scanned twice to optimise precision and quality. Scale bar: 500 µm. The arrow indicates the direction of growth. Figure 9. The location of the laser spots along the five transects. T1-T3 transect towards the inner shell surface and parallel to each other. T4 transects along the outer portion of the outer shell layer and T5 along the inner portion of the outer shell layer. The inner and outer portions of the outer shell layer are indicated by "Int." and "Ext.", respectively. The shell layers are indicated by OL for the outer layer, ML for the middle layer and IL for the internal layer. Note the absence of the outer shell layer and the inner shell layer in these sections of S. biradiata and F. tenuicostata, respectively. T4 and T5 were sampled from different shell layers as labelled. of the different phases; however, CRM confirmed the presence of calcite by measuring small intensities of the calcite librational mode L c compared to that of aragonite L a . The Raman spectra of S. biradiata have a slightly more intense calcite peak at 282 cm −1 in comparison to F. tenuicostata, and the XRD pattern shows that the main calcite peak in F. tenuicostata is generally sharper and offset towards 30 • 2θ compared to that of S. biradiata. According to Schroeder et al. (1969), this peak shift is indicative of the presence of Mg in carbonate, which suggests that some Mg 2+ ions have beeb substituted for Ca 2+ in the CaCO 3 crystal lattice in F. tenuicostata. Considering the preferential incorporation of Mg 2+ into calcite as opposed to aragonite (Mucci and Morse, 1982), we conclude that although both species appear to have similar amounts of calcite incorporated within their aragonitic shell, in F. tenuicostata, Ca 2+ ions have been partially substituted by Mg 2+ in the calcite lattice. This is not the case in S. biradiata. Sample G170 (S. biradiata), which shows shrinking in its aragonite phase according to the XRD results, has a negative peak shift (CRM) indicative of Sr 2+ substitution in aragonite. The LA-ICP-MS analysis of G170 was not undertaken because the thin section was too cracked towards the ventral margin; therefore, the higher Sr / Ca in G170 is assumed from the distortion measurements and CRM. Aragonite is present in two different orientations in most of the F. tenuicostata samples. The combination of these two orientations (e.g. Fig. 6) clearly corresponds to the crossedlamellar microstructure characteristic of many bivalve shells. The microstructure of S. biradiata shells consists of a prismatic outer shell layer, a crossed-lamellar middle layer and a complex crossed-lamellar inner layer (not present in the portion of the shell in Fig. 6). Both species precipitate their shells differently from a structural point of view. The precipitation of microstructural units in a crossed-lamellar pattern is complex. The structural complexity of F. tenuicostata, with its crossed-lamellar pattern associated with marked ribs, suggests a higher level of biological control on CaCO 3 precipitation, which may influence the overall elemental composi- Figure 11. The LA-ICP-MS results for the T4 and T5 transects. The blue lines represent the results for S. biradiata (G171) and the red lines represent the results for F. tenuicostata (G155). tion of the shell. For example, Paquette and Reeder (1995) suggested that crystal surface structure has an effect on trace element incorporation. As such, the different microstructures present in the two species studied here may be expected to yield different elemental compositions.
The measured lattice parameters reveal lattice distortion in both F. tenuicostata and S. biradiata compared to mineral CaCO 3 . Pokroy et al. (2006) also reported lattice distortions (Perna canaliculus, Acanthocardia tuberculata and Strombus decorus). After lattice parameter permutation, stretching was found along the a and b axes in both species, and shrinking was found along the c axis in S. biradiata but not in F. tenuicostata. The distortion was not as pronounced in S. biradiatata and F. tenuicostata compared to the species studied by Pokroy et al. (2006), which can be explained by the use of annealing to relax the CaCO 3 structure. The maximum anisotropic distortion in aragonite is a/a 1.1×10 −3 , b/b 2.5 × 10 −4 and c/c 3 × 10 −4 for F. tenuicostata, and a/a 1.7×10 −3 , b/b 7×10 −4 and c/c −6×10 −4 for S. biradiata. According to Pokroy et al. (2006), the organic molecules could be the source of these structural distortions. However, given that the F. tenuicostata lattice is more stretched than that of S. biradiata, the substitution of Ca 2+ ions by Mg 2+ ions described previously could also be a source of distortions. Ion substitution, which has been measured here using XRD, could also be a cause of this distortion because crystal structures accommodate trace elements through dilation or contraction of the ionic site (structural relaxation), which is a factor in the flexibility and stability of the surrounding structure (Finch and Allison, 2007;Loste et al., 2003).

Chemical composition
The LA-ICP-MS results show that Mg is positively correlated to both S and P, as is Sr to P in F. tenuicostata. These Mg correlations are not evident for S. biradiata, but Sr is positively correlated to S, Ba and Pb (Fig. 12), as is Ba to S in this latter species.
The strong correlations found in the present study seem to indicate that Mg incorporation is influenced by both phosphorus-and sulfur-containing macromolecules, i.e. both soluble and insoluble fractions of the organic shell matrix in F. tenuicostata but not in S. biradiata. The incorporation of Sr appears to be influenced by the insoluble fraction of the organic matrix in S. biradiata and the soluble fraction in F. tenuicostata. The correlation found between Ba and S also suggests that the insoluble organic matrix influences the incorporation of Ba in S. biradiata. Shirai et al. (2014) suggested that the incorporation of Sr is highly biologically mediated based on the strong correlation between Sr / Ca, S / Ca and the shell microstructure and that the elemental composition varies at a scale comparable to that of crystallites, i.e. the micrometre scale. The relatively coarse resolution used for LA-ICP-MS (75 µm) does not allow us to verify the spatial scale (< 10 µm) at which the elemental composition of F. tenuicostata and S. biradiata varies, but we can differentiate between the types of organic macromolecules influencing the shell geochemical composition, i.e. S-dominated or P-dominated. Our findings agree with those of Shirai et al. (2014) in that changes in the microstructure and the S / Ca ratio reflect changes in the organic composition at the mineralisation front. Our findings further show that P / Ca also reflects compositional changes at the mineralisation front. Consequently, we can say that not only do the organic matrices at the calcification front impact the chemical composition of the shell, but its composition also impacts the incorporation of trace elements in the shell, regardless of whether it is S-or P-dominated. This result is consistent with a number of previous studies showing that S-containing and P-containing macromolecules of the shell matrix, insoluble and soluble, respectively, influence nucleation, growth and crystal structure (e.g. Borbas et al., 1991;Crenshaw, 1982;Reddy, 1977;Wheeler and Sikes, 1984). In the context of geochemical proxies, it becomes evident that the combined effects of the shell organic matrix composition and the environmental parameters (e.g. SST) may cause variation in the Mg / Ca and Sr / Ca ratios, two of the most common proxies used for reconstructing paleo-SST.
Additional complexity is also evident. P and S concentrations are consistent (compare the transects of T1, T2 and T3) in F. tenuicostata but highly variable in S. biradiata. In the latter, low Mg corresponds to high P and S, and high Sr to high S and low P. This suggests that the incorporation of these elements within the shell of F. tenuicostata is predominantly influenced by seawater, particularly Mg. In S. biradiata, it is predominantly influenced by the organic matrix composition, as demonstrated by the strong correlations in the elements related to the organic matrix. The fact that S. biradiata is visibly richer in the organic matrix and owes its strength to its organic fraction, whereas F. tenuicostata shells have a thicker and more complex structure (e.g. ridges), might be one explanation for this finding. Another remarkable point of comparison that might help in understanding the differences between the two species of interest lies in their taxonomy. Indeed, both species belong to the Heterodonta subclass and the Cardiida order, but they belong to different superfamilies: Cardioidea and Tellinoidea for F. tenuicostata and S. biradiata, respectively (Ponder and Lindberg, 2008). Different organic content and shell organic matrix composition might be explained by evolved differences in the mineralisation processes (Cuif et al., 2011). However, without quantifying organic matrix components (total, soluble and insoluble) and identifying its composition, this remains speculative.
These results not only reinforce the knowledge that the shell organic matrix of bivalves is involved in biomineralisation, but also that its level of influence depends on its composition (phosphorus-containing or sulfur-containing macromolecules) and the specific element being incorporated into the CaCO 3 . Furthermore, these results show that this phenomenon is species specific. The incorporation of the main elements (e.g. Mg, Sr and Ba) used as proxies from marine properties (e.g. SST) is highly mediated by the shell organic matrix in general.

Conclusions
The combination of geochemical, spectroscopic and crystallographic analyses on specimens of F. tenuicostata and S. biradiata from King George Sound (south Western Australia) has revealed compositional and microstructural differences. Although the specimens were sampled from the same location and experienced the same marine conditions during their growth, they display different characteristics. Both species have been described as purely aragonitic, but low levels of calcite and Mg-calcite are present in S. biradiata and F. tenuicostata, respectively. The LA-ICP-MS measurements revealed correlations between levels of Mg, S and P (F. tenuicostata) and Sr, S and Ba (S. biradiata). Although analysing the composition of the shell organic matrix was beyond the scope of this study, the organic matrices of both bivalve species differ in composition and consequently in the way trace elements are incorporated into the CaCO 3 . The incorporation of Sr and Ba is affected by the S-containing matrix of S. biradiata, and the incorporation of Mg and Sr in F. tenuicostata is more influenced by S-and P-containing organic macromolecules. The substitution of Ca 2+ by Mg 2+ in calcite in F. tenuicostata appears to be influenced by the composition of the organic matrix. The combination of Sand P-containing organics could be promoting ionic substitution in the shells of F. tenuicostata.
The differences between these two species are clear and raise questions about genetic determinism, considering they belong to different taxonomic superfamilies. This has consequences for environmental proxy applications. The present study shows that the elemental composition of marine bivalve shells is species specific and influenced by multiple factors, such as crystallographic structures, organic macromolecule composition and environmental setting. These factors complicate the use of bivalve shells as a proxy for reconstructing the past and present properties of seawater. Mg / Ca and Sr / Ca ratios are amongst the most common proxies used for SST reconstructions; however, this study shows that the elemental changes are not solely caused by SST variation. The high definition multimodal and multidisciplinary characterisation of both the organic composition and the microstructure are central to understanding calcification and elemental incorporation. Only then can geochemical proxies be used to reconstruct true time series and the "vital effect" accurately described.
Data availability. The data in this paper form part of the PhD research of Liza M. Roger, which has not yet been submitted for examination. The thesis and all data generated as part of the thesis work will be available in the UWA thesis repository after examination.
The Supplement related to this article is available online at doi:10.5194/bg-14-1721-2017-supplement.
supported by the School of Earth Sciences and the Centre for Microscopy, Characterisation and Analysis of the University of Western Australia.
Edited by: H. Kitazato Reviewed by: A. D. Wanamaker Jr. and two anonymous referees