Experimental setup and conditions
This experiment was conducted at the Bermuda Institute of Ocean Sciences (BIOS) in St. George’s, Bermuda. The experimental treatments were two CO2 levels (high and ambient) and two feeding conditions (fed and unfed). The two pCO2 levels were established in static 5.5 gallon aquaria filled with serially filtered (50, 5 um) seawater prior to the addition of metamorphosed larvae. These conditions were achieved and maintained by directly bubbling air (in the ambient condition) or CO2 -enriched air (high CO2 treatment) through micropore bubble "wands" fixed horizontally approximately 5 cm from the base of each aquarium. A pair of Aalborg mass flow controllers maintained the CO2 concentration of the enriched treatment. The resultant average calculated pCO2 for ambient and high CO2 conditions were 421 ± 35 and 1,311 ± 76 uatm (mean ± SD), respectively, with corresponding average Ωar of 3.66 ± 0.15 and 1.63 ± 0.08 (mean ± SD), respectively. Ωar of the high CO2 treatments is within range of average global surface ocean Ωar predicted by global climate models for the end of this century under the IPCC SRES A2 (Steinacher et al. 2009). Corals in fed treatments were isolated (every night for 2 weeks, every other night for the third week) for 3 h in 12.5 cm x 12.5 cm x 3 cm plastic containers filled with seawater from their respective treatment tanks and provided with 24-h-old Artemia nauplii (brine shrimp). Feeding took place at night, shortly after lights were switched off to mimic crepuscular feeding and temporal zooplankton abundance observed in local coral reef environments (Lewis and Price 1975). Unfed corals were not provided nauplii during the 3-week experiment and were not isolated in empty feeding containers. Each CO2 -feeding treatment was conducted in triplicate for a total of twelve aquaria, and all treatments were kept on a 12/12 h light–dark cycle. Fluorescent aquarium lamps maintained maximum light levels of 62 ± 8 umol quanta m-2 s-1 (mean ± SD), which were monitored using a LI-COR probe/meter assemblage. The compensation range for F. fragum spat on Bermuda is not yet known. The investigators used the low end of known compensation ranges for corals (e.g. 3–233 umol quanta m-2 s-1 as reported by Mass et al. 2007) for two reasons. The first was to ensure that corals under elevated CO2 did not bleach (as experienced by Anthony et al. 2009), and the second was to minimize the potential for enhanced photosynthesis to overwhelm or inhibit the feeding-modulated calcification response to elevated CO2. Aquarium temperatures were maintained by in-line chiller/heater systems and monitored every 15 min (Hobo temperature loggers, Onset Corp.). Average temperature for all treatments over the course of the experiment was 27.6 ± 0.1 degrees C (± SD).

Aquarium water was replaced with filtered seawater every week to prevent the build-up of dissolved inorganic nitrogen and other wastes. Prior to removing water from the aquaria, discrete water samples were collected for salinity, alkalinity (Alk), and dissolved inorganic carbon (DIC) from every aquarium. Salinity was measured at BIOS with an Autosal salinometer. The Alk/DIC samples were poisoned with mercuric chloride immediately after collection and analyzed using a Marianda VINDTA-3C analysis system at WHOI. Alkalinity was determined by nonlinear curve fitting of data obtained by open-cell titrations, and DIC concentrations were determined by coulometric analysis. Both measurements were standardized using certified reference materials obtained from Dr. A. Dickson (Scripps IO). The pH (NBS) of each tank was measured every 3–4 d (Orion pH meter and temperature- compensated electrode) to provide a real-time assessment of tank chemistry. Short-term variations in NBS pH were also assessed on a higher-resolution time scale: for one, 24-h period, by measuring pH in each aquarium at 3-h time intervals. The pH within each tank was maintained within ± a few hundredths of a pH unit on both sub-weekly and sub-daily time scales. The carbonate system parameters used to compare treatments (pCO2, [HCO3- ], [CO32-], and Ωar) were calculated from the average temperature and discretely sampled salinity, Alk, and DIC data using the CO2SYS program (Lewis and Wallace 1998; Pelletier et al. 2007) with the constants of Mehrbach et al. (1973) as refit by Dickson and Millero (1987).

Coral collection, spawning, and larval settlement
In July 2010, approximately 1 week prior to anticipated peak larval release date (Goodbody-Gringley and de Putron 2009), the investigators collected 30 mature colonies of the brooding coral, F. fragum, from the Bailey’s Bay patch reefs off the northwest Bermudan coast at approximately three to seven meters water depth. Adult colonies were maintained in outdoor flow-through seawater aquaria at BIOS under ambient light and temperature conditions. Parent colonies were kept isolated in glass jars during planula release, which occurred over the course of 6 nights. The live zooxanthellate planulae were collected from all parents and pooled together. Ceramic tiles, approximately 9 square cm, were left out on the reef for 2 months prior to the start of the experiment and further conditioned for larval settlement by scattering bits of freshly collected crustose coralline algae on the tiles. Immediately after collection, actively swimming larvae were transferred to small plastic tubs each containing ceramic tiles and filled with seawater preset to targeted CO2 levels. The tubs had mesh lids, allowing for water exchange, while they are submerged in the treatment aquaria. After 48 h, larvae had settled and metamorphosed into primary polyps (at this stage, larvae are "spat"). Spat on tiles were quickly counted, and tiles were pseudo-randomly distributed among the experimental aquaria so that each aquarium had approximately the same number of juvenile corals. Calcification was visible approximately 3 d after settlement. At the end of 3 weeks (± 1 d), 20–50 primary polyps (including their primary corallite) per treatment were removed from the tiles and frozen at- 80 degrees C for analysis of total lipid. Tiles were then removed from treatments and submerged in a 10% bleach solution for 1 h, which removed the polyp tissue from all of the remaining juvenile corals and exposed the calcified skeleton or primary corallite.

Quantification of baby coral skeletal development, size, and weight
Each bleached skeleton was digitally photographed, removed from the tile, and weighed using a Metro-Toledo micro-balance. Images of the baby corals (i.e. spat) were examined for skeletal development and size using Spot Imaging software. Length of the primary septa (present in all samples) was used to estimate corallite diameter (i.e., size). The septa are lateral CaCO3 plates that corals accrete in cycles. In our experiment, most spat accreted both primary and secondary septa; the tertiary septa were the last septal cycle accreted by any of the juvenile corals. Rate of skeletal development was defined as percent spat exhibiting tertiary septa, and a two-way ANOVA was used to test for differences in the mean proportion of spat with tertiary septa between the treatments. Feeding treatment and CO2 level were fixed effects. Data were arc sin square root transformed to homogenize variances prior to analyses. To test for differences in mean spat weight and diameter among treatments, a two-way, nested multivariate analysis of variance (MANOVA) was performed on natural log transformed weight data and square root transformed diameter data. Feeding treatment and CO2 levels were fixed main effects, while tank effect was the random factor nested within feeding and CO2 levels. Eight univariate F tests were conducted to test each of the dependent variables. A Bonferonni corrected alpha value of 0.0062 was used to declare significance of F statistics. It should be noted that the MANOVA only considers corals that have data for both diameter and weight. If part of a corallite is lost during weighing or was attached to coralline algae, both coral size and weight were excluded from the MANOVA analyses. Likewise, if the skeleton was irregularly shaped (i.e., primary septa did not lie in a straight line), the data for those corals were not included. In order to account for any bias that may have resulted from corallite exclusion in the MANOVA, ANOVAs for the dependent variables, weight, and diameter were conducted. These tests considered all data for a given dependent variable to compare with the MANOVA’s univariate results.

Quantification of baby coral total lipid and symbiont density
Ten individual spat from each aquarium were pooled per tissue lipid sample for quantification of total lipid by gravimetric analysis. Pooling was necessary due to the small size of the spat at 3 weeks. Extraction methods follow that of Folch et al. (1957) and Cantin et al. (2007). Five individual spat from each aquarium were pooled per sample for quantification of symbiont density. Spat were homogenized, centrifuged and the resultant pellet was re-suspended in 250 l L filtered seawater. Symbionts from multiple (6–9) aliquot sub-samples of the slurry were counted on a known volume hemocytometer grid. Both total tissue lipid and symbiont counts were normalized to the circular area described by the average primary septa length (diameter) for a respective tank and then divided by the number of corals pooled in the sample (i.e., 10 or 5). Both area-normalized lipid content and symbiont density were compared among levels of CO2 and feeding conditions using two-way ANOVAs with tank as a random factor nested within the CO2 and feeding combinations. Total lipid concentration was transformed to - 1/x in order to homogenize the variances. All statistical analyses were conducted on SYSTAT.