Full annual monitoring of Subantarctic Emiliania huxleyi populations reveals highly calcified morphotypes in high-CO2 winter conditions

Ocean acidification is expected to have detrimental consequences for the most abundant calcifying phytoplankton species Emiliania huxleyi. However, this assumption is mainly based on laboratory manipulations that are unable to reproduce the complexity of natural ecosystems. Here, E. huxleyi coccolith assemblages collected over a year by an autonomous water sampler and sediment traps in the Subantarctic Zone were analysed. The combination of taxonomic and morphometric analyses together with in situ measurements of surface-water properties allowed us to monitor, with unprecedented detail, the seasonal cycle of E. huxleyi at two Subantarctic stations. E. huxleyi subantarctic assemblages were composed of a mixture of, at least, four different morphotypes. Heavier morphotypes exhibited their maximum relative abundances during winter, coinciding with peak annual TCO2 and nutrient concentrations, while lighter morphotypes dominated during summer, coinciding with lowest TCO2 and nutrients levels. The similar seasonality observed in both time-series suggests that it may be a circumpolar feature of the Subantarctic zone. Our results challenge the view that ocean acidification will necessarily lead to a replacement of heavily-calcified coccolithophores by lightly-calcified ones in subpolar ecosystems, and emphasize the need to consider the cumulative effect of multiple stressors on the probable succession of morphotypes.

Sensor measurements from SOTS were used to provide context for the RAS sample collections as follows. PAR measurements were from a spherical sensor (MDS-MKVL2000655, Alec Inc.) mounted in-air on the Pulse-8 surface float. Surface mixed layer salinity was measured using a CTD (Seabird SBE16+ at 36 to 49 m depth) within the RAS package, but because this failed mid-deployment, data was used from a deeper sensor (Seabird SBE37 at 105 m depth -slightly below the mixed layer in summer, which showed excellent correlation and negligible offset with the RAS mounted sensor when both were operational). For the same reason, temperature was taken from a deeper sensor (SBE56 at 45 m depth -within the mixed layer). Chlorophyll-a was excluded from the analysis because the record ended in March 2012. Averages of all the available values for the 7 days prior collection of the sample by the RAS were estimated. Also values higher than three times the standard deviation were considered oultiers and therefore not used for the analysis.
In situ carbonate system measurements at the SOTS site were taken from Shadwick, et al. 3 for the period November 2011 to October 2013. As the coccolith sampling commenced in August 2011, the missing carbonate system data between August to November 2011 was completed with data from year 2012. The assumption that the seasonality of the carbonate system between adjacent years show little variability in the Subantarctic Zone (SAZ) is supported by previous work in the region 4,5 . Lowest TCO2 concentrations were registered in mid-summer (~ 2060 μmol kg −1 ; Figure 4) mainly driven by biological activity while and maximum concentrations (> 2100 μmol kg −1 ) were registered in winter-spring transition when mixed layer pCO2 is in near-atmospheric equilibrium. Both pH and calcite saturation state displayed a nearly opposite pattern with annual maxima in summer (8.10 and 3.97) and minima in early spring (8.03 and 3.03; Fig. 4).

Regional representativeness of the SOTS and SAM sites
Comparison of remote sensing and hydrographic observations suggests that the High Nutrient Low Chlorophyll (HNLC) waters sampled by the SOTS site can be considered representative of a large portion (~90 to 145° E) of the Subantarctic Zone 1,6,7 . Similarly, the sediment trap on the SAM has an estimated particle source area ('statistical funnel') of ~120 km for a particle sinking speed of 100 m d -1 that potentially encompasses a wide area of the northern sector of the SAZ off eastern New Zealand 8,9 . More specific statements regarding the largescale representativeness of the morphometric E. huxleyi observations are not possible, although we note that sparse particulate inorganic carbonate abundance observations from ships and as derived from satellite reflectance show similar patterns over the SAZ south of Australia and New Zealand 10,11 . Importantly, oceanographic observations at the SOTS site and more generally in in the SAZ suggest that the relationships between E. huxleyi morphotypes and environmental conditions are likely to represent local adaptations rather than originate from advective transports to the region. Water parcels with different T-S properties are occasionally advected past the SOTS site (e.g. three periods with relatively warm and salty compositions were observed in the annual record presented by 12 . However, comparison to spatial variations obtained from satellite and ship observations suggests that relatively short trajectories (~100 km) are sufficient to explain these variations, because the Subantarctic Zone waters around SOTS exhibit similar mesoscale variability 13 . Notably, the passage of these T-S anomalies does not exhibit any particular seasonality. Thus, while the passage of discrete water parcels may well contribute to the overall observed constellation of morphotypes, we do not have evidence that it influences their seasonality. More generally, SAZ surface water properties near SOTS do reflect the mixing of warm, salty, low DIC concentration waters supplied from the north (including via eddies released from the extension of the East Australian Current) with cold, fresh higher DIC concentration waters supplied from the south (via both Ekman and eddy tranports), 14,15 . These studies suggest that the balance of this mixing varies seasonally, with greater influence of the northern source in summer. Thus, the broadscale seasonality of advective inputs to the SAZ acts to reinforce the locally driven seasonality of warming and DIC availability, and thus advective inputs of E. huxleyi do not appear to be a viable alternate explanation for the seasonal variations in morphotypes.

Taxonomic descriptions of Emiliania huxleyi morphotypes
Morphotype identification of E. huxleyi coccoliths was based on taxonomic concepts of Young et al. (2003) and Hagino et al. (2005) with slight modifications concerning the size range of each morphotype adapted for the subantarctic populations analysed here. Coccoliths of E. huxleyi morphotype A are medium-sized (2.5-4 μm), with robust distal shield elements and clearly-visible central-area elements. Given the substantial variability in the degree of calcification of Type A coccoliths, the ratio between slit length (SL) and tube width (TW) 16 was used to classify them into two main categories (Fig. 2). Coccoliths with SL > TW were grouped as "regularly calcified" Type A, while Type A coccoliths with SL<TW or with a closed central area were classified under Type A over-calcified (o/c) (Fig. S1). Morphotype B coccoliths are large (3.5-5 μm), with numerous distal shield elements (≥35) and a central area open or sometimes covered with a thin plate. Type B/C coccoliths display a relatively similar morphology to types B and C but are intermediate in size (2.5-4 µm), usually with 25-33 distal shield elements with central area open or covered by a thin plate. In those few cases in which the length of the distal shield fell within the overlapping size range of B and B/C (i.e. between 3.5 and 4), the number of distal shield elements was used as distinguishing feature between these two morphotypes. Morphotype C coccoliths are small (≤2-3.5µm), often with irregular shape compared to other morphotypes, with a distal shield element number usually ranging between 18-25 and a central area open or covered by a thin plate. In those cases where the length of the distal shield fell within the overlapping size range of morphotypes B/C and C, the number of distal shield elements and shape of the coccolith (i.e. regular vs. irregular) were used as discriminative features.

Coccolith mass and size measurements
The birefringence-based method to provide coccolith mass and size estimates is based on the systematic relationship between the thickness of a given calcite particle and the interference colour produced under polarized light [17][18][19] . In this study, an apical rhabdolith of the genus Acanthoica collected by a sediment trap at the SOTS site was used for calibration. The microscope light settings and camera parameters were kept constant throughout the imaging session. A calibration image of the same rhabdolith was taken at the beginning of each imaging session to account for possible bulb ageing. Additionally, the same E. huxleyi coccolith (termed "calibration coccolith") was imaged after every calibration in order to assess the consistency of the coccolith mass and length measurements between sessions. Images were then processed using C-Calcita software 17 . The output files for all calcite particles with a diameter ranging between 1 and 8 µm were visually examined and E. huxleyi coccoliths selected. The excellent coccolith preservation in the water and sediment trap samples allowed us to differentiate E. huxleyi coccoliths from species of genus Gephyrocapsa (i.e. the other members of Noëlaerhabdaceae family present in the SAZ with relatively similar coccolith shape and size range) on the basis of the presence or absence of a conjunct bridge. The standard deviation of the mass and length of the "calibration coccolith" was used as a measure of calibration error across sessions. The calibration error was ±5 and ±2 % for coccolith mass and length, respectively. Because our birefringence-based method uses grey scale images to estimate coccolith thickness, it can be applicable only to coccoliths thinner than 1.55 mm 20 . This does not represent a limitation in our analysis because the thickness of all E. huxleyi morphotypes are below this threshold.
The wide range of coccolith mass estimates proposed in the literature for E. huxleyi ranges between ca.1.4 and 7.0 pg 21,22 . The source of these variability is mainly due to two factors: differences in coccolith mass between morphotypes and methodological biases associated to each of the most commonly used techniques for coccolith mass estimation (i.e. morphometrics, regression and birefringence). Because morphotype B/C is the most abundant morphotype in our samples and is geographically restricted to the Southern Ocean 23,24 , we limit the comparison of our results to studies conducted only in the Southern Ocean using the same methodology. It should be noted that this is a conservative approach, because a previous study in the AZ waters south of Tasmania 21 using our birefringence-based approach showed good agreement with previous estimates in the Southern Ocean obtained with morphometric and birefringence approaches 22,25 .

Annual estimates
In order to facilitate comparisons with other settings, annual flux estimates were calculated. For the SOTS sediment trap records, the unobserved interval occurred in winter when fluxes were low and, therefore, annual estimates were obtained by using an average flux value of the first and last cups (both collected during the winter) to represent mean daily fluxes during the unobserved period. For the SAM sediment trap, the gaps over the collection interval were quasi-evenly distributed throughout the time series. In this case, the gaps in the record were filled by linear interpolation of the closest cups and the winter gap was filled using the same approach as in the SOTS time-series. Annual E. huxleyi coccolith mass and length estimates were estimated following the same approach. With regard to the annual contribution of E. huxleyi morphotypes, two different annual estimates were calculated: the average relative contribution of each E. huxleyi morphotype over the collection interval and the annualized flux-weighted relative contribution of the morphotypes collected by the sediment traps. This information is illustrated in Figure S2.

Seasonal variations in coccolith mass and length
Monthly variability E. huxleyi coccolith mass and length measured at the sediment traps at the SOTS and SAM sites was evaluated using a Generalised Additive Model (GAM), to account for non-linear relationships, fitted with Gaussian distribution errors (WOOD). In particular, the relationships of coccolith mass and length with month were modelled using cyclic cubic regression splines, whose ends match, to take into account that after month 12 comes month 1. Stepwise model selection was carried out by AIC.
Month and site explained 74% of the variability in coccolith mass, and both variables were highly significant. Overall the seasonal variability in coccolith mass and length was similar for both sites, with maximum coccolith mass and length observed during September and lowest during February-March (Fig. S3).

Comparison analysis of E. huxleyi coccolith mass across depths and sites in the Southern Ocean
The variability in E. huxleyi coccolith mass across sites and depths of the SOTS, SAM (SAZ) and 61°S 26 (Antarctic Zone; AZ) sites was analysed using a linear model, assuming a Gaussian distribution of the dependent variable. Then model assumptions were verified visually. A full model was produced including all interactions (Mass = Site + Depth + Site:Depth). An automatic stepwise model selection procedure showed that there was no significant effect of depth, neither any of the interactions, thus the final model only included site (Fig. S4). The similar average annual coccolith weight observed at the three traps depths in the SOTS site together with the good preservation of the coccoliths suggested by SEM observations, in both SOTS and SAM sites, indicate that negligible coccolith dissolution occurs at meso-and bathypelagic depths in the study region or within the sediment trap cups.    : Changes in average E. huxleyi coccolith mass across sites and depths (mean and respective bootstrap 95% confidence interval). Coccolith mass changed significantly between the location of the sampling stations, with the different sites accounting for 30% of the variability in the masses (F[2,97]=20.84, p-value < 0.001, R 2 =30%). Site 61°S showed the smallest masses (2.10 ± 0.2 pg 95% CI), followed by site SAM (2.46 ± 0.18pg 95% CI), with SOTS site showing heaviest coccoliths (2.83 ± 0.12 pg 95% CI). Note that average coccolith mass values over the collection intervals (i.e. not annualized flux-weighted coccolith mass values) are presented in this figure.

Supplement figures.
Supplement Tables. Table S1. a. Sampling dates and morphotype relative abundance of E. huxleyi coccolith assemblages collected in the surface layer at the SOTS site. b. Sampling intervals, fluxes and morphotype relative abundance and morphometric measurements of E. huxleyi coccolith assemblages intercepted by the sediment traps at the SOTS and SAM sites.