The site is briefly described here. For further details, we refer to Agostini et al. (2018). Mikama Bay is located on the south-southwest corner of Shikine Island, offshore of the Izu Peninsula, Japan (34.320865∘ N, 139.204868∘ E), and has several venting locations in the north of the bay (Fig. 1). The gas emitted from the seep contains 98 % CO2, and the bay has a spatially and temporally constant total alkalinity averaging at 2265±10 µmol kg-1. Samples were collected from three preestablished pCO2 sites (Agostini et al., 2015), a “control pCO2” site in an adjacent bay outside the influence of the CO2 seep (pCO2 309±46 µatm), a “mid-pCO2” site (pCO2 ca. 460±40 µatm), and a “high-pCO2” site (pCO2 769±225) (Fig. 1). pCO2 estimates are based on the carbonate chemistry parameters (pHNBS, temperature, salinity, and total alkalinity) of water in the bay and calculated using the program CO2sys (Agostini et al., 2018; Harvey et al., 2018). Temperature (annual range ca. 14 to 28 ∘C) and salinity (ca. 34 ‰ ) are relatively homogenous throughout the bay and do not differ between the elevated pCO2 sites and control pCO2 sites (Agostini et al. 2018). Currents and wind were measured in October 2014 and April 2015 (Agostini et al., 2015). October 2014 measurements showed moderate turbulent winds (ranging from 0.6 to 11.5 m s-1, with an average of 4.5 m s-1) associated with current velocities (ranging from 0 to 1.6 knots, average 0.4 knots) at 5 m in the surface water, whereas April 2015 measurements showed moderate north-northeast winds (1.5–8.6 m s-1, average 5.1 m s-1) associated with low current velocities (0–0.2 knots, average 0.04 knots). Monthly surveys in the bay over the past 5 years show that these sites have similar annual mean values for temperature, salinity, and currents. Weather station data show that the severity of seasonal extreme weather event (e.g., typhoons) varies on an annual basis (Japan Meteorological Agency, https://www.jma.go.jp/en/typh/, last access: 1 April 2019).

Samples were collected in June and September of 2016 (indicated in Fig. 1). All samples were collected in at least triplicate for each site (control pCO2, mid-pCO2 and high-pCO2 site). Additional control sites (ca. 1.8 and 2.4 km away from the CO2 seep) around the island were taken to ensure the fidelity of the control site closest to the seep. June sampling included surface waters for dissolved inorganic carbon (DIC) measurements, surface sediments, and benthic diatoms attached to surface sediment through extracellular polymeric substance production. In September, macroalgae and plankton tows were collected, in addition to surfaced water DIC and surface sediments, taken in triplicate at each site on four separate days.

Water for the δ13C of DIC analysis was collected by overfilling glass vials with sea surface water and adding mercury chloride (0.5 %) before closing with a septa cap and sealing with electrical tape. Surface sediments were collected by divers using geochemical sample bags. Macroalgae and benthic diatoms were scraped off submerged rocks at each respective site. . A 25 µm mesh plankton net (“small plankton net”, Rigo, Saitama, Japan) was towed 50 m three times per site and filtered using a portable hand aspirator on the boat over a single 0.7 µm GF/F (glass fiber filter; combusted prior to sampling for 4 h at 450 ∘C). All samples were immediately frozen; once back in the lab, these were freeze-dried and kept in a refrigerator.

Each seawater sample was measured for the δ13C of DIC in duplicate on a Thermo Scientific Gas Bench II coupled to a Thermo Scientific Delta V mass spectrometer. Prior to running samples, 12 mL vials were prepared with 100 µL of 85 % H3PO4 and flushed with He. A total of 500 µL of seawater was added and left to react for at least 1 h prior to measuring the headspace. Standards were run at the start, end, and every six runs of a sequence. Standards were prepared with 0.3 mg of Na2CO3 and 0.4 mg of Ca2CO3 (all calibrated against NBS-19) flushed with He, injected with 500 µL of 85 % H3PO4, and reacted for 1 h. The headspace was measured, and average values and standard deviation errors reported are based on 6 measurements for June (3 at the high-pCO2 site and 3 at the control) and 36 measurements for September (3 each at the high-pCO2 site, mid-pCO2 site, and control, collected on four separate days).

Freeze-dried sediments, benthic diatoms, and macroalgae were homogenized using mortar and pestle and extracted using a Dionex 250 accelerated solvent extractor at 100 ∘C, 7.6×106 Pa using dichloromethane (DCM) : MeOH (9:1 v/v). GF/Fs containing plankton net material were cut into 1 mm × 1 mm squares and extracted using ultrasonication (5×) with 2 mL dichloromethane (DCM) : MeOH (9:1 v/v). All total lipid extracts (TLEs) were hydrolyzed by refluxing the TLE with 1 N of KOH in MeOH for 1 h and neutralized to pH 5 using 2 N of HCl in MeOH. Bi-distilled water (2 mL) and DCM (2 mL) was added (5×) to the hydrolyzed centrifuge tubes and the organic matter in the DCM layers were pooled and dried over Na2SO4. The resulting hydrolyzed TLEs were eluted over an alumina packed column and separated into apolar (hexane : DCM, 9:1 v/v), ketone (DCM), and polar (DCM : MeOH, 1:1 v/v) fractions. Polar fractions were silylated using pyridine : N,O-Bis(trimethylsilyl)trifluoroacetamide (BSTFA) (1:1 v/v) and heated at 60 ∘C for 1 h prior to analyses on the gas chromatography-flame ionization detector (GC-FID), gas chromatography-mass spectrometry (GC-MS), and gas chromatography isotope-ratio mass spectrometry (GC-IRMS).

Silylated polar fractions were analyzed by GC-FID for quantification. Based on the quantities, fractions were diluted with ethyl acetate and ca. 1 µg of polar fraction was injected on-column for GC-MS to identify compounds and for GC-IRMS to measure the isotopic composition of specific compounds. Each instrument is equipped with the same CP-Sil 5 column (25 m × 0.32 mm; df 0.12 µm) and He is used as carrier gas. GC oven was programmed from 70 to 130 ∘C at 20 ∘C min-1 and then to 320 ∘C at 4 ∘C min-1, which was held for 10 min. All three instruments use the same in-house mixture of n-alkanes and fatty acids to check chromatography performance at the start of each day (GC-standard). For compound-specific stable carbon isotope analysis using GC-IRMS, additional standards with known isotopic values (-32.7 ‰ and -27.0 ‰) of per deuterated (99.1 %) n-alkanes (C20 and C24, respectively), were co-injected with the GC-standard. Samples were also co-injected with the same GC-IRMS standards to monitor instrument performance. Every day, the Isolink II combustion reactor of the GC-IRMS was oxidized for at least 10 min, backflushed with He for 10 min, and purged for 5 min; a shorter version of this sequence is conducted in post-sample seed oxidation, which includes 2 min oxidation, 2 min He backflush, and a 2 min purge conditioning line. Longer oxidations were run weekly. Each derivatized compound was corrected for the δ13C of the BSTFA used in silylation (-32.2 ‰ ).

All three biomarkers, phytol, loliolide, and cholesterol, show a negative shift in δ13C values with increasing pCO2 in each matrix and each season (Fig. 3), agreeing with the theory that higher pCO2 conditions result in lower δ13C values in biomass (Farquhar et al., 1982). However, despite all having algal sources, the absolute isotope values vary for (1) each compound, (2) each matrix, and (3) both seasons, which we will now discuss.

First, the absolute values of δ13C values vary among the three compounds. This may be expected given the different biosynthetic pathways leading to formation of each compound (Schouten et al., 1998), as well as the different contributors to each compound. Loliolide, considered a diatom biomarker in paleoreconstructions (e.g., Castañeda et al., 2009), is a diagenetic product of fucoxanthin (Repeta, 1989; Klok et al., 1984), a xanthophyll that contributes to approximately 10 % of all carotenoids found in nature (Liaaen-Jensen, 1978). Phytol, considered a photoautotroph biomarker in paleoreconstructions (Hayes et al., 1990), is the side-chain of the vital and omnipresent pigment chlorophyll a that directly transfers sunlight energy into the photosynthetic pathway in nearly all photosynthetic organisms. Sterols, considered a general eukaryotic biomarker in paleoreconstructions, are the eukaryotic tetracyclic triterpenoid lipids used for critical regulatory roles of cellular functions, e.g., maintaining membrane fluidity (Nes et al., 1993). Although sterols are virtually restricted to eukaryotes, some exceptions have been found in bacteria (Wei et al., 2016). Here we only examine cholesterol, which is universally absent in prokaryotes and composes of up to 20 %–40 % of eukaryotic plasma membranes (Mouritsen and Zuckermann, 2004). Phytol and cholesterol may also have terrestrial sources, given that they are derived from all photoautotrophs and all eukaryotes, respectively. However, these samples were taken off the coast of a small island in the open ocean, and the absence of characteristic terrestrial biomarkers indicates that terrestrial contributions can be considered to be minimal. The close resemblance of the isotopic composition among all three compounds, including the primarily diatom-limited compound loliolide, suggests that these compounds share relatively similar source organisms. Cholesterol shows a lessened isotopic shift compared to the other two compounds from the ambient to elevated pCO2 sites. Although we cannot fully exclude that this is due to terrestrial input, it is more likely due to the mobile eukaryotic zooplankton in the water column, which also contribute to the cholesterol signal.

Within the same biomarker and same season, some differences among matrices were observed. This difference may be due to the mobility of the matrix, as well as the algal assemblages. The plankton tow that captured free-floating surface water algae from that specific growth season is more readily transported by wind than the surface sediment, which likely reflects the culmination of multiple growth seasons throughout the water column. This is seen, for example, in the δ13C of cholesterol collected in September from the same control site where surface sediments are -22.2 ‰ and plankton tows are -25.1 ‰, where the latter has possibly been transported from sites with elevated pCO2 levels. Similar differences among matrices are also observed in phytol and loliolide. The hypothesis of transportation affecting the isotopic signal in certain matrices is supported by the results from the macroalgae. The macroalgae, in contrast to the algae collected by plankton tows, were unaffected by transportation due to being fixed to the nearby rocks at each site. Thus, the isotopic composition of compounds of the macroalgae was similar to that of the surface sediments accumulated over a long period, e.g., -22.7 ‰ for the δ13C of cholesterol at the September control site.

Finally, there is a difference in the δ13C values for biomarkers between seasons. The June-collected surface sediments and algae yielded a larger difference in δ13C values along the CO2 gradient than the September-collected surface sediments and algae. This seasonal difference may be due to extreme weather conditions experienced between the two sampling campaigns. Although typhoons are common in this region, in the weeks preceding the fieldwork in September, Shikine Island experienced an unusually high quantity of storms. The storms were also of unusual strength for this region of the Pacific, including typhoons Mindulle and Kompasu, the severe tropical storms Omais and Chanthu, and the long-lived, erratic Lionrock typhoon. This atypical abundance and severity of storms observably ripped corals out of the rocks around Shikine Island and thus likely resuspended and transported some sediment around the bay. This would explain the reduced δ13C difference between the control and high-pCO2 site in the surface sediments collected in September, as well as the readily transportable algae collected by the plankton tow, and would explain why the rock-affixed macroalgae, also collected in September, maintained a strong δ13C change across the transect.

To further validate the impact of pCO2, we calculated the isotopic fractionation of algal biomass based on the δ13C of the three biomarkers. Here we focus on surface sediments as they are a close analogue to the geological sediment records. Although the macroalgae and benthic diatoms also show strong isotopic fractionation, they represent a limited number of species and a single growth season. Furthermore, we calculated the εp from the June-collected surface sediments, which appear to be the least affected by typhoon activity and represent fractionation over multiple seasons.

To calculate εp in the June-collected surface sediments, we correct the δ13C of the organic matter (δp) for the δ13C of the inorganic carbon source for the producers of these compounds (δd) in Eq. (1): 1εp=1000⋅δd+1000/δp+1000-1. δp is calculated by correcting the δ13C for each individual biomarker for the offset with photosynthetic biomass caused by isotopic fractionation during biosynthesis. The isotopic offset between phytol and biomass is 3.5±1.3 ‰, based on the average of 23 species compiled in Witkowski et al. (2018), and the isotopic offset between sterols and biomass is 4.5±3.0 ‰ based on the average of eight algal species (Schouten et al., 1998). The isotopic offset for loliolide from biomass, however, has not been determined. Because isoprenoids are formed from the same biosynthetic pathway, here we average the offset of the other two isoprenoids here (4.0 ‰) to estimate a value for the difference between loliolide and biomass.

δd is calculated by correcting the measured δ13C of DIC for temperature (Mook, 1974) and pH (Madigan et al., 1989), which considers the relative contribution of different inorganic carbon species to the measured DIC. Based on the equations of Mook et al. (1974), we correct for the temperature-dependent carbon isotopic fractionation of dissolved CO2 with respect to HCO3- using the annual mean sea surface temperature for Shikine Island of 20.4 ∘C (Agostini et al., 2018). Based on the equations of Madigan et al. (1989), we corrected for the δ13C of HCO3- and δ13C of CO2[aq] mass balance calculation that accounts for the relative abundance of these inorganic carbon species based on pH (Lewis and Wallace, 1998) at the high-pCO2 site (7.81 pHT) and mid-pCO2 site (7.99 pHT), relative to the ambient control (8.14 pHT). The corrected δd values yield -10.1 ‰ at the control site, -10.0 ‰ at the mid-pCO2 site, and -9.5 ‰ at the high-pCO2 site (Table S2).

εp values consistently yield much higher values at the elevated pCO2 sites than the ambient control sites for all three biomarkers, which share similar trends and absolute values (Fig. 4; Table S3). εp derived from loliolide averages 7.2±1.6 ‰ at the control, 9.2±1.6 ‰ at the mid-pCO2 site, and 15.9±1.6 ‰ at the high-pCO2 site, εp derived from phytol at 8.6±0.4 ‰, 8.6±0.9 ‰, and 14.9±1.0 ‰, respectively, and εp derived from cholesterol at 7.6±3.0 ‰, 9.2±3.1 ‰, to 13.7±3.1 ‰, respectively, where errors represent the standard deviation of the triplicate samples taken at each site. These results show that CO2 has a profound impact on εp as it is the only variable with a large gradient in the bay. Given that maximum fractionation for algae species is ca. 25 ‰ to 28 ‰ in laboratory cultures (Goericke and Fry, 1994), the CO2 seep values suggests strong fractionation, but does not approach maximum fractionation (εf) at the high-CO2 site. This may be due to presence of carbon concentrating mechanism in phytoplankton, which utilize 13C-enriched bicarbonate, or possibly due to the presence of rubisco types with different εf values than previously assumed (Thomas et al., 2018).

We estimate pCO2 from the εp values, a relationship first derived for higher plants (Farquhar et al., 1982, 1989) and later adapted for algae (Jasper et al., 1994; Rau et al., 1996) in Eq. (2): 2pCO2=b/εf-εp/K0, where εf reflects the maximum rubisco-based isotopic fractionation, b reflects species carbon demand per supply such as growth rate and cell-size (Jasper et al., 1994), and K0 reflects a constant to convert CO2[aq] to pCO2 based on temperature and salinity (Weiss, 1974). εf for algal species range from 25 ‰ to 28 ‰ in laboratory cultures (Goericke and Fry, 1994); we use an average 26.5 ‰ with an uncertainty of 1.5 ‰ uniformly distributed for these general algal biomarkers (Witkowski et al., 2018). The b value is difficult to estimate as it is a catchall for factors other than pCO2 that affect fractionation and is particularly difficult to estimate for general algal biomarkers because they are derived from a multitude of species. Previous studies using phytol's diagenetic product phytane as a pCO2 proxy (Bice et al., 2006; Sinninghe Damsté et al., 2008; van Bentum et al., 2012) have used a mean value of 170 ‰ kg µM-1, similar to the mean of alkenone producers. This is supported by a compilation of the δ13C values of modern surface sediment organic matter mean average of 168±43 ‰ kg µM-1 (Witkowski et al., 2018) and a single study on phytol in the equatorial Pacific Ocean (Bidigare et al., 1997). We apply this average, rounded to 170±50 ‰ kg µM-1, to all three general algal biomarkers.

The resulting reconstructed pCO2 estimations show the expected values in the control sites and much higher values in the elevated CO2 sites among all three biomarkers (Fig. 5; Table S3). Loliolide shows the biggest shift, from 239+50/-49 µatm at the control, 266+57/-54 µatm at mid-pCO2 site, and 437+113/96 µatm at the high-pCO2 site. Phytol has a similar but slightly smaller shift in pCO2 estimates to loliolide, with estimations of 264+55/-54, 291+56/-53, and 444+98/-87 µatm at the control, mid-pCO2 site, and high-pCO2 sites, respectively. Cholesterol shifts in a similar manner to the other two biomarkers with 244+67/-54, 266+77/-61, and 358+136/-90 µatm, respectively. These reconstructed values closely match each other and trend in the same direction as the actual values.

The reconstructed pCO2 values derived from the δ13C of general algal biomarkers closely match the actual measured pCO2 values of the control (Fig. 5), i.e., 309±46 µatm (Agostini et al., 2018; Harvey et al., 2018), when considering the uncertainty in the reconstructed estimations. However, the proxies underestimate the absolute values measured at the elevated pCO2 sites (Fig. 5; Table S3), i.e., 460±40 µatm at the mid-pCO2 site and 769±225 µatm at the high-pCO2 site (Agostini et al., 2018; Harvey et al., 2018). There are several possible explanations as to why there is an underestimation. As discussed above, carbonate concentration mechanisms may be operating in a large number of phytoplankton, such that they become relatively enriched in 13C and thus lead to lower reconstructed pCO2 values (Stoll et al., 2019; Badger et al., 2019). There is also a large uncertainty in the b value applied, which may be much lower than the value assumed here. However, if this is the case, then pCO2 values reconstructed for past times may be much higher, leading to considerable discrepancies with other pCO2 proxies (cf. Witkowski et al., 2018). A simple explanation for this underestimation may be some site limitations. The high variability of pCO2 at these sites could have impacted the reconstructed values, as these algae could have been exposed to much different, and perhaps lower, levels than those observed during the times that pCO2 values were measured. Furthermore, there is a strong possibility of allochthonous marine input of sediment at the mid-pCO2 and high-pCO2 sites, i.e., input from sediment outside of the bay area. This allochthonous input seems likely given the intense weather conditions that occur annually in this small bay in which lateral transport of sediment could bring algal material grown in ambient pCO2 conditions into the bay and dampen the overall pCO2 signal picked up in the biomarkers. Future research conducted at another CO2 seep setting with different weather and current conditions could illuminate this.