Calcification and growth processes in planktonic foraminifera complicate the use of B/Ca and U/Ca as carbonate chemistry proxies

Calcification and growth processes in planktonic foraminifera complicate the use of B/Ca and U/Ca as carbonate chemistry proxies Journal Item How to cite: Salmon, Kate H.; Anand, Pallavi; Sexton, Philip F. and Conte, Maureen (2016). Calcification and growth processes in planktonic foraminifera complicate the use of B/Ca and U/Ca as carbonate chemistry proxies. Earth and Planetary Science Letters, 449 pp. 372–381.


Introduction
B/Ca and U/Ca of planktonic foraminiferal calcite have previously been suggested to reflect the carbonate chemistry of seawater (Russell et al., 2004;Allen et al., 2011 and could potentially be used to reconstruct past ocean carbonate chemistry changes (Yu et al., 2007;Foster, 2008). Documenting past seawater carbonate chemistry changes is important in defining the processes that drive Earth's climate system and carbon cycle, and how these will respond to future anthropogenic climate change. Theoretically, boron exists in seawater as two species, boric acid [B (OH 3

)] and
This leads to an exchange distribution coefficient (K D ) (Yu et al., 2007), (see equilibrium equation (3) However, a variety of evidence also exists arguing against a primary carbonate control on B/Ca ratios including: (1) The lack of correlation of B/Ca and pH-dependent boron isotope composition (Foster, 2008) and measured carbonate system parameters (Babila et al., 2014), (2) Higher B/Ca in upwelling regions despite lower pH conditions (Naik and Naidu, 2014), (3) Species-specific sensitivities of B/Ca to carbonate chemistry , (4) Size-dependent B incorporation into foraminiferal calcite (Yu et al., 2007;Babila et al., 2014;Henehan et al., 2015), (5) Discrepancies between culture and open-ocean studies of the sensitivity of B/Ca to [CO 2− 3 ] (Yu et al., 2007;. Additionally, a number of other environmental variables have been reported to influence B concentrations in foraminiferal calcite including, temperature (Hathorne et al., 2009;Naik and Naidu, 2014), salinity (Allen et al., 2011Henehan et al., 2015), light intensity (Babila et al., 2014), and  ] (Henehan et al., 2015). As with boron, uranium incorporation into planktonic foraminifera is also theoretically controlled by  because it exists in seawater as a series of carbonate complexes and is likely incorporated into calcite as either UO 2 (CO 3 ) 2− 2 and/or UO 2 (CO 3 ) 4− 3 (Yu et al., 2008 and references therein). Therefore, the decreasing abundance of UO 2 (CO 3 ) 2− 2 with greater [CO 2− 3 ] should also be reflected by decreasing foraminiferal U/Ca (Russell et al., 2004). Yet, like boron, a number of other variables in addition to the carbonate system have been suggested to also influence U incorporation, including calcification temperature (Russell et al., 1996;Yu et al., 2008), growth rate (Ni et al., 2007), and species-specific differences (Yu et al., 2008).
Here, we utilise bi-weekly samples from a sediment trap time series from the subtropical North Atlantic (Salmon et al., 2015) to address the unresolved and potentially conflicting issues with these proxies. Sediment trap time series provide a unique opportunity to evaluate controls on the geochemical composition of planktonic foraminifera within their natural habitat where multiple variables influence their calcification simultaneously. This contrasts with laboratory methods, which although have been instrumental in ground-truthing relationships between geochemical proxies and environmental variables, typically isolate only single variables to determine their influence on foraminiferal calcite. The bi-weekly sampling resolution provided by the sediment traps is also advantageous because it captures the lifecycles of most planktonic foraminiferal species, which typically follow the monthly lunar cycle (Jonkers and Kučera, 2015), with evidence of an annual cycle for encrusted Globorotalia truncatulinoides (Hemleben et al., 1985;McKenna and Prell, 2004). We measured B/Ca and U/Ca ratios, test calcification (thickness) and growth parameters (size) of three planktonic foraminifer species from bi-weekly sediment trap samples spanning four years (1998-2000 and 2008-2010), from the Sargasso Sea (Fig. 1a). These data were coupled with in-situ oceanographic data (temperature, salinity, chlorophyll and carbonate system parameters) from the Bermuda Atlantic Time Series (BATS) in the same locality to evaluate the controls on uranium and boron incorporation into planktonic foraminifera (Fig. 1b-c).

Materials and methods
We use bi-weekly sediment trap samples selected from the Ocean Flux Programme time series in the Sargasso Sea (31 • 50 ′ N, 64 • 10 ′ W) together with the concurrent oceanographic data from the nearby BATS site (31 • 40 ′ N, 64 • 10 ′ W). We utilise samples collected from two equivalent 2.5-yr intervals (1998-2000 and 2008-2010) at 1500 m water depth to capture seasonal variations in the test parameters and geochemical composition of three species of foraminifera (Globigerinoides ruber (pink), Orbulina universa and Globorotalia truncatulinoides, non-encrusted (nc) and encrusted (c)), each living at different depths in the water column. Test weights and sizes were measured individually in order to calculate test area density (µg/µm 2 ) of each sample (details in section 2.3).
Tests used for geochemical analysis ranged in size (shown in Table 2), but on average were G. ruber (p) = 366 µm, O. universa = 720 µm, G. truncatulinoides (nc) = 492 µm, G. truncatulinoides (c) = 709 µm. G. truncatulinoides reproduce in the surface waters, as evidenced from plankton tows (Hemleben et al., 1985) (Ujiié et al., 2010), as this is the dominant genotype present at our Sargasso Sea study site. Likewise, we analyse the more abundant thinner shelled Sargasso genotype of O. universa, which is morphologically distinct under high magnification, from the thicker-walled Caribbean genotype (Morard et al., 2009). By selecting species from a wide range of water depth habitats (approximately 0-400 m), we are able to compare oceanographic data with species' geochemical compositions across larger environmental gradients; e.g. from 0-400 m at this site, temperature ranges up to ∼10 • C (Fig. 1b) and [CO 2− 3 ] by ∼50-60 µmol/kg ( Fig. 1c), with a minimal change in salinity (∼0.2). After being measured, planktonic foraminifera tests were gently cracked to open and subjected to chemical cleaning involving an extended oxidation step to remove any excess organic matter present in sediment trap material (50% H 2 O 2 in 0.2M NaOH) (Anand et al., 2003), followed by a weak acid leach, prior to final dissolution (Barker et al., 2003). B/Ca and U/Ca analyses were carried out on a Thermo ® Element XR Inductively-Coupled Plasma Mass Spectrometer (HR-ICP-MS), at the Godwin Laboratory at Cambridge University. Long-term precision on standard runs of B/Ca (and Mg/Ca used for calcification temperatures calculations) is <1.0% (2σ ). External precision is <4.0% (2σ ) for B/Ca and U/Ca, and <1.0% (2σ ) for Mg/Ca using Cambridge consistency standards (Misra et al., 2014). B, Mg blank levels were <2% and U was <5% of typical [B], [Mg] and [U] in foraminifera samples. An in-house standard was used to correct for drift over the run.

Calcification temperature and depth calculations
Calcification temperatures were determined from analysing the δ 18 O calcite in the same aliquot of sample used for trace element analyses. Stable isotope analyses were performed on a Finnigan GasBench and Delta plus Advantage stable isotope mass spectrometer at the Open University (long term standard reproducibility is ±0.084❤ for δ 18 O and ±0.061❤ for δ 13 C) and are reported relative to Vienna Pee Dee Belemnite (V-PDB). Temperature and salinity data from different depth habitats were taken from the Bermuda Atlantic Time Series (BATS) to calculate the δ 18 O of calcite in equilibrium with seawater (δ 18 O sw ). The δ 18 O sw at Bermuda was calculated using the 0-50 m δ 18 O sw -salinity relationship for the tropical-subtropical Atlantic available through the NASA seawater database (Arbuszewski et al., 2010;Schmidt et al., 1999).
Calcification temperatures were then calculated using δ 18 O calcite and calculated δ 18 O sw where available. For G. ruber (p) and G. truncatulinoides, we use the rearrangement of the palaeotemperature equation of O'Neil et al. (1969) and Shackleton (1974) and for O. universa, we used the low-light palaeotemperature equation of Bemis et al. (1998). For samples where no stable isotope data  (1998-2000 and 2008-2010). Numbers on the plots are water depths in metres.
were available, we use species-specific Mg/Ca-temperature calibration equations to calculate the calcification temperatures and associated depth habitats (Anand et al., 2003). In order to estimate depth habitats, we then matched these calcification temperatures for individual samples to BATS oceanographic temperatures measured in the previous month to account for foraminifer lifecycle and settling time (see section 2.2 for more details). The depth at which the oceanographic temperature most closely matches the calcification temperature is then used to denote the approximate depth habitat, similar to previous a study (Marshall et al., 2013). Spero et al. (1997) Figure A.2), so no correction was applied here.

Carbonate parameter calculations
In-situ seawater carbonate system parameters were calculated using monthly oceanographic data obtained from the BATS database (http :/ /bats .bios .edu /bats _methods .html). Assuming an average 3-4 week lifespan of G. ruber (p) and O. universa (Jonkers and Kučera, 2015) and a 5-day settling period to reach the 1500 m trap (∼300 m/day, Takahashi and Bé, 1984;Marshall et al., 2013), we selected oceanographic data approximately 1 month before the mid-date of the sediment trap opening period to account for the typical lifespan of a foraminifer. Encrusted G. truncatulinoides likely reproduces on an annual cycle (McKenna and Prell, 2004) but numerous non-encrusted specimens are found in the surface waters during the winter months after adult specimens reproduce (Hemleben et al., 1985;Spear et al., 2011), so we also used oceanographic data from ∼1 month before the trap opening for nonencrusted G. truncatulinoides. We then selected an oceanographic temperature that best matched the calcification temperature for that species in that sample. All of the oceanographic data asso-ciated with this temperature (including all of the carbonate data, salinity, and productivity parameters) were then used in accordance with these shell measurement and geochemical data. The temperature, salinity, and two carbonate parameters (dissolved inorganic carbon, total alkalinity data) were then inputted into the CO2Sys_v2.1.xls program (Pelletier et al., 2007) in order to calculate the remaining carbonate system parameters. We applied the carbonic acid dissociation constant of Mehrbach et al. (1973), refit by Dickson and Millero (1987) and the dissociation constant for HSO − 4 (Dickson, 1990).

Calcification and test size data
Previous studies have shown that size-normalised weights (SNWs) or area density (a proxy for shell thickness) of planktonic foraminifera may reflect variations in foraminiferal calcification, mainly related to changes in seawater [CO 2− 3 ] (Barker and Elderfield, 2002;Marshall et al., 2013 and references therein). SNWs are calculated by weighing batches of shells in narrow sieve-size 'windows' (Beer et al., 2010), but this method has largely been superseded by a more precise measurement termed area density (µg/µm 2 ). Area density measurements are calculated using individual, digitally measured test areas normalised to individual test weights (µg/µm 2 ). We use the average area density of a number of individually measured tests (n = 6-30) to represent the area density for each sample (Table B.2). This technique allows for a more accurate estimation of relative changes in shell wall thickness (Marshall et al., 2013).
Although changes in planktonic foraminiferal shell thicknesses represent a complex physiological process and are not directly comparable to the controls on inorganic calcite precipitation rates, previous observations of a positive relationship between SNW or area density and [CO 2− 3 ] (Marshall et al., 2013 and references therein), are consistent with observations of increasing calcite precipitation 'rates' in higher pH seawater (Ruiz-Agudo et al., 2012). However, since foraminifera calcify intermittently during their growth, their calcification rates will inherently vary throughout their lifecycles. Because the lifecycles of both G. ruber (p) and O. universa are externally controlled by the lunar cycle (Jonkers and Kučera, 2015), both species likely calcified over comparable time periods (3-4 weeks). Likewise, non-encrusted G. truncatulinoides do not incorporate any secondary calcite so most likely calcified in the surface waters for a few weeks before the sinking to the trap (Spear et al., 2011). Therefore, the area densities of all three species used in this study (excluding encrusted G. truncatulinoides) represent the intermittent precipitation of calcite over comparable intervals of time. In the context of this study, we use foraminifera test area density to document the average of these intermittent calcification rates over the lifespan of a foraminifer, where greater area densities represent thicker gross test walls and thus faster calcification rates averaged across these intermittent windows of calcification. Area densities thus reflect biologically mediated calcification and are not directly synonymous with inorganic calcite precipitation rates.
Individual foraminifera tests were weighed on a XS Mettler Toledo microbalance and photographed in the same orientation, under a stereomicroscope for size analysis. We calibrated ImageJ analysis software using a microscale image taken at the same magnification as foraminifera tests and adjusted the image threshold to determine 2D silhouette areas of individual tests. Errors on area density measurements were defined as AD ±(1/n) (Marshall et al., 2013). Individual test size data of shells within each sample were used to evaluate biological control of B and U incorporation in planktonic foraminifera (Salmon K., unpublished PhD Thesis, 2015). In line with previous work, we interpret higher area densities as representing thicker shells and hence a greater gross rate of calcification over the lifetime of the foraminifer (Marshall et al., 2013).

Multiple linear regression
We performed multiple linear regression analyses using the function 'lm' in R (http :/ /www.r-project .org) to test which of the independent environmental (e.g. carbonate chemistry, temperature, salinity, chlorophyll) and ecological parameters (e.g. test size/area density) explain the variance in the dependent variables i.e. B/Ca and U/Ca. The selections of these independent variables are based on previous observations of temperature (Yu et al., 2008), salinity (Allen et al., 2011;Henehan et al., 2015), carbonate chemistry (Yu et al., 2007(Yu et al., , 2013Allen et al., 2011, secondary crust formation (Hathorne et al., 2009) and size fractionation Ni et al., 2007;Friedrich et al., 2012) related to calcification/growth rate (we use area density/test size as a proxy) (Ni et al., 2007;Naik and Naidu, 2014) affecting the incorporation of B and U into the foraminiferal test. We use chlorophyll concentration to test how changes in productivity could affect trace element ratios in samples where these data are available (Table B.2). It is important to note that area densities and test sizes of G. truncatulinoides are positively correlated because tests grow simultaneously larger and thicker, so these shell parameters cannot be decoupled in this species. Our subtropical gyre has negligible dissolved phosphate present in surface waters (Steinberg et al., 2001) meaning we were unable to explicitly test a recently observed B/Ca relationship with [PO 3− 4 ] in planktonic foraminifera (Henehan et al., 2015). For each analysis, non-contributing factors were removed according to their contribution to Akaike Information Criterion (AIC) (Akaike, 1974). AIC provides a statistical framework in which to select a model with the least number of parameters required in order to achieve the best fit. Lower values of the index indicate the model with the fewest parameters that still provides an adequate fit to the data.
The more sensitive the dependent variable is to changes in the independent variable, the greater the value of the slope coefficient in the regression. We have repeated the analyses using a multispecies model, and species-specific models for O. universa and G. truncatulinoides (non-encrusted and encrusted). We were too limited by sample sizes to produce reliable species-specific models for G. ruber (p) and non-encrusted G. truncatulinoides. We have intentionally excluded encrusted G. truncatulinoides from the multispecies model because the presence of secondary crust in this species may bias our results. Confidence in the species-specific O. universa model is reduced compared to the other models owing to the small sample size relative to the number of independent variables.

Seasonality in B/Ca, U/Ca and calcification temperatures
The seasonal range in calcification temperatures (calculated from a combination of δ 18 O and Mg/Ca), varies from 15-30 • C for all three species over both the sediment trap deployment periods. This is on average within the range of estimated calcification depths for each species ∼16 m for G. ruber (p) and ∼68 m for O. universa (Fig. 2a) (see Table B.2). Generally, most of the nonencrusted G. truncatulinoides appear to calcify in the surface waters during winter corresponding to an average calcification depth of ∼44 m, but a cold-core cyclonic eddy in February-March 2010 caused anomalously low surface water temperatures as indicated by low calcification temperatures in Fig. 2a. Encrusted G. truncatulinoides predominantly reside at depths of ∼300-400 m throughout  Table B.2 for more accurate depth estimates. The errors on calcification temperatures are detailed in Table B.2. The errors on internal standard measurements (σ ) are B/Ca = ±1.84%, U/Ca = ±1.79%, in Cambridge consistency standards containing <20, ∼85 and ∼200 µmol/mol B/Ca and ∼45-65 nmol/mol U/Ca stated in Misra et al. (2014). the winter, and some non-encrusted individuals calcify at 0 m (Fig. 2a).
Generally, the higher B/Ca of G. ruber (pink) coincides with warmer summer temperatures and greater [CO 2− 3 ] compared to O. universa. However, within species, the B/Ca of G. ruber (p) and O. universa do not appear to change with temperature or [CO 2− 3 ] (Fig. 2b). The range of B/Ca in G. truncatulinoides alone is equal to the combined range of G. ruber (p) and O. universa (∼110 µmol/mol), despite this species living in a more limited range of winter temperatures and [CO 2− 3 ] (Fig. 1b-c). Nonencrusted G. truncatulinoides contain greater U/Ca than encrusted individuals but this is not so for B/Ca. Like B/Ca, the U/Ca of G. ruber (p) is greater than U/Ca in O. universa and ranges from ∼5 nmol/mol in O. universa to 15 nmol/mol in G. ruber (p). G. truncatulinoides has the largest range in U/Ca (∼13 nmol/mol) and is also positively offset from G. ruber (p) and O. universa (Fig. 2c). Although calcification temperature appears to explain some variance in B/Ca of all species, this could be an artefact of combining species over a large depth range, because it is not significant in explaining species-specific B/Ca variations in O. universa or G. truncatulinoides individually ( Table 1). The small range in salinity at this site does not appear to control any variance in B/Ca or U/Ca. Likewise, chlorophyll concentrations, used as an indicator of productivity, do not exert any significant control over B/Ca or U/Ca in any of the species tested (although chlorophyll could not be tested in encrusted G. truncatulinoides, below the euphotic zone). None of our tested parameters exert any significant control on B/Ca in encrusted and non-encrusted G. truncatulinoides only, but area density/test thickness and test size explain most of the variation in U/Ca in this species (r 2 = 0.90) ( Table 1).

Controls on trace element incorporation
We aimed to keep our measurements within as narrow test size ranges as possible but G. truncatulinoides has a larger variance in size due to growth and addition of crust (Table 2). We find a correlation between test size and intra-species variations in U/Ca and B/Ca in most species (Table 2). Higher B/Ca values significantly correlate with larger tests but only in G. ruber (p) and the relationship is weak (r 2 = 0.43) compared to higher U/Ca strongly associated with larger tests in all species (r 2 = 0.52-0.81), except encrusted G. truncatulinoides (Table 2).

Carbonate chemistry
U/Ca and B/Ca have been previously suggested as proxies for the ocean carbonate system based on laboratory culturing experiments (Sanyal et al., 1997;Allen et al., 2011. However, our field study in context with others shows that overall, regardless of which carbonate parameter is used to represent carbonate chemistry, there is no significant influence on B incorporation (Fig. 3a). Likewise, along with area density, carbonate chemistry only explains 24% of the variations in the U/Ca of G. ruber (p), O. universa, and non-encrusted G. truncatulinoides (Table 1). Hypothetically, the seasonal range of 30 µmol/kg [CO 2− 3 ] in O. universa at this site should be equivalent to a 0.7 nmol/mol change in U/Ca, according to laboratory culturing work on the same species (Russell et al., 2004). Yet this predicted 0.7 nmol/mol range in O. universa U/Ca is much smaller than the ∼6 nmol/mol change we actually observe (Fig. 2c). Instead, much of the U/Ca variability can be explained through test size fractionation in individual species (Table 1-2).
The 60-75 µmol/mol range in B/Ca in G. ruber (p) and O. universa observed in this study only corresponds to a range in [CO 2− 3 ] of 18-30 µmol/kg in respectively (Fig. 2b). However, according to laboratory culturing calibrations on the same species, this 60-75 µmol/mol range in B/Ca should equate to a range in [CO 2− 3 ] ] makes no difference to the outcome of the multi-species B/Ca regression. Test size was not applied in the multi-species regression because it is species-specific. Samples with missing data for specific variables were not included in any regressions. Coefficients shown in bold refer to variables that were included in a second regression after selection according to the lowest AIC values (see section 2.4 for more details). The r 2 values reflect the outcome of the modelled variables highlighted in bold. Adjusted r 2 accounts for sample size and number of independent variables.

Variable
All that is over an order of magnitude greater (∼300->600 µmol/kg) (Allen et al., 2011. Indeed, the full range of all species' B/Ca recorded in this study (∼110 µmol/mol) should be equivalent to a pH range of ∼0.6 according to culture calibration  but we only see a ∼0.1 unit change in pH over the depth habitats of all species. Comparison of our results with other coretop and sediment-trap based studies shows that our sediment-trap data yield the largest range in B/Ca with one of the smallest ranges in [CO 2− 3 ] (Fig. 3a). The lack of a distinct relationship between B/Ca and [CO 2− 3 ] in this study, and other studies and species clearly indicates the existence of additional competing controls on boron incorporation in the natural environment.
One primary control on B/Ca could be [B(OH) − 4 /HCO − 3 ] seawater (equation (3)), which is a function of temperature and pH   (Babila et al., 2014), equivalent to a 3-5 µmol/mol change in B/Ca according to the calibration from , just a fraction of the total ∼110 µmol/mol B/Ca range observed here (Fig. 3b).

Calcification temperature
In general, there is too much scatter to conclude there is any significant relationship between temperature and B/Ca (Fig. 3c). The lack of intra-species correlations of B/Ca and temperature, suggests that significant correlations highlighted in Table 1 could be an artefact of comparing multiple species. Yu et al. (2008) suggested that uranium incorporation into planktonic foraminifer calcite was strongly influenced by calcification temperature, based on core-top sediments. However, our sediment trap data reveal no significant relationship between uranium and temperature in any of the species that we examined (Table 1).

Calcification rate and test size
We find that B/Ca displays a strong positive correlation with area density, pointing to a calcification control on boron incor-poration in G. ruber (p), O. universa and non-encrusted G. truncatulinoides (Fig. 4a). Although area density reflects the complex physiological process of biological calcification, and is not a direct measure of the rate of inorganic calcite precipitation, the inter-species correlation between area density and B/Ca is consistent with recent inorganic precipitation experiments which observed a dependency of boron incorporation on calcite precipitation rate (Ruiz-Agudo et al., 2012;Gabitov et al., 2014). If we assume that higher area densities represent faster calcification rates, as suggested by previous investigations (Spero et al., 1997;Marshall et al., 2013 -references therein), our results suggest that calcite growth rates, even when biologically mediated, could affect boron partitioning. This mechanism may explain why thicker tests with higher area densities contain more boron.
G. ruber (p) is the only species where B/Ca has a (weak) correlation with test size (r 2 = 0.43) (Fig. 4b), whereas U/Ca has a strong positive correlation with test size in all species (r 2 = 0.52-0.81) ( Table 2) (Fig. 4d). This is consistent with previous observations of greater B/Ca in larger tests of G. ruber (white and pink) (Ni et al., 2007;Babila et al., 2014;Naik and Naidu, 2014;Henehan et al., 2015) and U/Ca in larger G. ruber (white and pink) and G. sacculifer tests (Ni et al., 2007). Most studies have attributed this test size fractionation to faster foraminiferal growth (where growth is represented by increases in chamber formation, not crystallographic precipitation) and calcification rates in larger individuals (Ni et al., 2007;Babila et al., 2014;Naik and Naidu, 2014). A faster calcification rate may be less effective at discriminating against incorporation of trace elements causing the organism to incorporate higher concentrations into its calcite lattice (Rickaby et al., 2002;Russell et al., 2004;Ni et al., 2007;Schmidt et al., 2008). However, if this were the case in all of our species, we would also expect to see a stronger positive relationship between U/Ca and area density (Fig. 4c), and also a positive relationship between B/Ca and test size in O. universa and non-encrusted G. truncatulinoides (Fig. 4b), but we do not. Whilst assuming larger tests grow faster and hence have faster calcification rates appears to be a deduction inconsistent with the majority of our data, it may be true for at least some species, such as G. ruber (white and pink). served a 15-20 µmol/kg offset in B/Ca between the 200-300 and 300-400 µm size fractions of G. ruber (white). Although they attribute this offset between size fractions to increased light intensity governing greater boron incorporation in larger tests due to a greater density of symbionts, we suggest that it could equally arise from increased growth/calcification rates in larger tests. We find that area densities in larger G. ruber (p) tests from this study are on average higher, when compared to area densities from smaller tests; 311-363 µm (equivalent sieve size = 250-300 µm, Figure A.1) = area density of 1.23 × 10 −4 µg/µm, compared to 371-431 µm (equivalent sieve size = 300-355 µm, Figure A.1) = area density of 1.34 × 10 −4 µg/µm. We find this offset in area densities between larger and smaller tests of G. ruber (p) is equivalent to a ∼22 µmol/kg offset in B/Ca, almost the same discrepancy observed between sieve-size fractions by Babila et al. (2014). Yet Babila et al. (2014) discounted a calcification rate control on B/Ca because they do not see a similar offset between size fractions for Sr/Ca and Mg/Ca. However, a calcification rate control would not necessarily have to be reflected in both Mg/Ca and Sr/Ca to be plausible in B/Ca. Ni et al. (2007) concluded there was no calcification control on Mg/Ca in G. sacculifer and G. ruber even though there was in B/Ca, Li/Ca and U/Ca. Additionally, we discount a light intensity control on B incorporation in this study based on three observations. First, we find no correlation between test size or B/Ca and δ 13 C in symbiont-bearing G. ruber (p) or O. universa (which would be enhanced in larger tests with more symbiont activity due to greater carbon fixation, Spero and Parker, 1985) (see Figure A.3a-b). Second, we also see no effect of test size on boron incorporation in symbiont-bearing O. universa (Fig. 4b). Third, non-encrusted symbiont-barren G. truncatulinoides possess a greater B/Ca concentration than O. universa, and comparable B/Ca to G. ruber (p) (Fig. 4a), which should not be the case if its incorporation were primarily controlled by symbionts enhancing microenvironment pH ( Figure A.3a-b).
Recently, an observed correlation between [PO 3− 4 ] and B/Ca ratios in Globigerinoides ruber led to the suggestion that [PO 3− 4 ] may be a control on foraminiferal B/Ca (Henehan et al., 2015). These authors proposed that this may arise from chemical interactions involving P at the mineral-water interface, either through paired substitution, increased disorder in the crystal lattice, or perhaps by stabilisation of an amorphous calcium carbonate precursor phase. However,  ] at our subtropical gyre site is negligible in the surface waters and only reaches 0.26 µmol/kg at ∼400 m (Steinberg et al., 2001), which according to the approximate B/Ca-[PO 3− 4 ] relationship described by Henehan et al. (2015), should be equivalent to ∼30 µmol/mol change in B/Ca, just a quarter of our observed 110 µmol/mol range. In addition, we observe higher foraminiferal B/Ca at our site (G. ruber (p) and nonencrusted G. truncatulinoides) than Henehan et al. (2015) (Henehan et al., 2015), we suggest that the positive correlation between B/Ca in G. ruber (w) and  ] could be caused by higher growth and calcification rates, which coincide with areas of higher productivity. This is consistent with other studies that have observed greater B/Ca in larger (and hence faster growing) G. ruber (w) tests during upwelling periods (Naik and Naidu, 2014) and also in cultured foraminifera fed every day, compared to lower B/Ca in open-ocean foraminifera from a comparable pH (Henehan et al., 2015). We do not observe a significant relationship between chlorophyll and foraminifera B/Ca variations, suggesting that productivity is not the dominant control on foraminiferal calcification and hence boron incorporation at this site.
In contrast to B/Ca, we observe greater U/Ca in non-encrusted G. truncatulinoides compared to the symbiont-bearing species, G. ruber (p) and O. universa, which is consistent with a higher microen- vironment pH in symbiont-bearing species (Fig. 4d). However, if U/Ca were predominantly controlled by the microenvironment pH, we would expect U/Ca to decrease with increasing test size, due to a higher pH in larger tests of symbiont-bearing species ). Yet our observations indicate the reverse, displaying a strong species-specific increase in U/Ca with larger test sizes in all species (excluding encrusted G. truncatulinoides) but for B/Ca, only in G. ruber (p) (Fig. 4d). Additionally, the symbiont-bearing species tend to experience a proportionally greater increase in U/Ca with larger test size, than non-encrusted G. truncatulinoides do (Fig. 4d). This indicates a stronger species-specific growth control on U compared to B incorporation. A calcification control is unlikely to explain U/Ca variability in symbiont-bearing species because it does not share a significant relationship with area density (Table 1 -O. universa only, Fig. 4c). Furthermore, unlike B/Ca, U/Ca is inversely related to [CO 2− 3 ] (Russell et al., 2004); because tests calcify thicker shells (potentially at faster rates) in higher [CO 2− 3 ] conditions (Marshall et al., 2013), it would be counterintuitive for an increase in U/Ca to be attributed to faster calcification rates as suggested by Ni et al. (2007). Our results suggest that the incorporation of B is distinctly different from U, because the size-dependent, species-specific relationships are more apparent in foraminiferal U/Ca, compared to B/Ca (Fig. 4d). Based on larger shells representing faster growth (Schmidt et al., 2008), ratedependent discrimination can act against the biological pumping of larger ions by the cell (Rickaby et al., 2002;Ni et al., 2007), which may explain greater U incorporation in larger foraminifera tests. However, it is still unclear why this mechanism would selectively favour U over B incorporation in larger tests. Future studies should therefore concentrate on determining which cellular pro-cesses contribute to variations in foraminifera test size and area density and how these relate to different incorporation strategies for B and U.
Overall, our results suggest more research is needed to quantify the impacts of calcification and biological fractionation on the incorporation of B and U before they can be used effectively as carbonate system proxies.

Crust and size control in G. truncatulinoides
Mature individuals of G. truncatulinoides form a secondary crust as they descend deeper in the water column, the composition of which can be chemically distinct from the primary calcite for some trace elements (Spear et al., 2011). Yet, we find similar range of B/Ca in both non-encrusted and encrusted G. truncatulinoides, whilst U is depleted in encrusted individuals (Fig. 4a, c). If U incorporation was controlled by [CO 2− 3 ], we would expect encrusted individuals to contain higher U/Ca but this is not the case. Our results indicate that the addition of secondary crust does not significantly affect the bulk test B/Ca, but does affect U concentration, further supporting different controls on B and U incorporation. For instance, the B/Ca of encrusted individuals could conceivably reflect the original area density from the non-encrusted stage, with the addition of the secondary crust tripling the area density without increasing the boron concentration (Fig. 4a). Unlike their non-encrusted equivalents, encrusted G. truncatulinoides have a large range of B/Ca; the 100 µmol/mol range of B/Ca in encrusted individuals almost encompasses the entire 110 µmol/mol range of B/Ca in non-encrusted, O. universa and G. ruber (pink) combined (Fig. 4a). Other encrusted globorotaliid species, such as G. inflata and G. scitula have also shown large intratest B/Ca variability (Hathorne et al., 2009;Allen et al., 2011), even when grown under identical conditions (Allen et al., 2011), suggesting this heterogeneity could reflect biological changes in the microenvironment, rather than an external environmental control. The depletion in U in encrusted G. truncatulinoides could be explained by dilution of [U] when the secondary crust is added. Further research using laser ablation is needed to determine if there are lower concentrations of U in the secondary, compared to the primary calcite.

Conclusions
Here we aimed to resolve conflicting interpretations of the controls on B/Ca and U/Ca in three species of planktonic foraminifera G. ruber (p), O. universa, G. truncatulinoides (non-encrusted and encrusted), and evaluate their use as proxies for ocean carbonate chemistry.
We find that species-specific B/Ca is not related to in-situ carbonate chemistry variations during the calcification of planktonic foraminifer tests in this study or collectively when other studies (such as those from core-top sediments) are taken into account. Instead, we suggest that boron incorporation is likely to be controlled by calcification rate owing to a strong positive correlation with test area density (thickness). We find little dependence of B/Ca on test size, except in G. ruber (pink).
We find significant increases of U/Ca with test size in all species (r 2 = 0.52-0.81), indicating some other biological control (perhaps related to growth) on incorporation of U compared to B. Our study is the first to show that both area density (proxy for test thickness) and test size may affect the incorporation of trace elements differently whereas previously, only changes in test size have been thought to affect trace element composition of planktonic foraminifera.
We recommend that future work should utilise planktonic foraminifera from narrow test thickness and size windows in order to isolate the dominant environmental controls on B and U incorporation in planktonic foraminifera. Our findings caution against the use of fossil planktonic foraminifera B/Ca and U/Ca as reliable proxies for the carbonate system until we have an improved mechanistic understanding of how calcification and growth rates regulate boron and uranium incorporation into the foraminifer test.