The influence of Antarctic subglacial volcanism on the global iron cycle during the Last Glacial Maximum

Marine sediment records suggest that episodes of major atmospheric CO2 drawdown during the last glacial period were linked to iron (Fe) fertilization of subantarctic surface waters. The principal source of this Fe is thought to be dust transported from southern mid-latitude deserts. However, uncertainty exists over contributions to CO2 sequestration from complementary Fe sources, such as the Antarctic ice sheet, due to the difficulty of locating and interrogating suitable archives that have the potential to preserve such information. Here we present petrographic, geochemical and microbial DNA evidence preserved in precisely dated subglacial calcites from close to the East Antarctic Ice-Sheet margin, which together suggest that volcanically-induced drainage of Fe-rich waters during the Last Glacial Maximum could have reached the Southern Ocean. Our results support a significant contribution of Antarctic volcanism to subglacial transport and delivery of nutrients with implications on ocean productivity at peak glacial conditions.

interpretation of negative  13 C values as related to microbial metabolism (see main text). The tips of the Cs crystals show dissolution and corrosion, likely related to the injection into the system of understaturated waters. In contrast, luminescence is dispersed in Dm. This suggests that nucleation of Dm occurred on abundant organic compounds and silicate particles transported into the system by exogenous waters. By abating interfacial energy barrier, particulate and organic molecules favoured formation of myriads of small crystals (microsparite) 2 ; D) Thin section detail of BV9a2c showing clasts of metamorphic sandstone, amphibolite and gneiss embedded in micrite and/or coated by micrite veils (Cg in Supplementary Table 1). Amphibolite and gneiss are typical of the regional Palaeozoic basement 3 , whereas metamorphosed sandstones outcrop above the sampling site, at the crest of the Helliwell Hills, and pertain to the Permo-Triassic Beacon-Supergroup 4 . Angular to sub-rounded amphibolite clasts reflect short lived transport. By contrast, metamorphosed sandstone granules and pebbles are rounded, which suggest transport by aqueous currents from up-glacier. Granules coated by micrite "envelopes" indicate cyanobacteria or endolithic microbes activity 5 . Cyanobacteria can be photosyntheticindependent 6 , and the lack of information on bacterial functions in Antarctica limits our ability to reconstruct the specific species and functions of the ancient microbial community preserved in the calcites. However, ages obtained from micrite coatings (Cg in Supplementary Table 1) suggest that they formed in the LGM, thus, they are likely related to sub-glacial bio-mediated precipitation of calcium carbonate crystals. Their morphologies are typical of lacustrine environments 6 , and allow an up-glacier provenance to be inferred, likely from subglacial lake discharge. Scale bar in B, C, D = 0.5 mm.

Supplementary Figure 2 Scanning Electron Microscopy (SEM) images of spherulites
All images show spherulitic crystals likely consisting of calcium fluoride embedded in the 25.135 ± 0.537 ka to 23.524 ±0.446 ka layer in BV9b. A) Spherulites surrounding clasts in Dm. B) Spherulite preserved within Cs. C-E) Images in back scattered electrons (BSE) mode. The different shades indicate composition changes: calcite is light grey and calcium fluoride is white. Normalized weight % measured by EDX, and processed by Quantax software (Brucker Nano Gmbh) yielded F = 32.7%, Ca = 45.8%. Formation of calcium-fluoride spherulites suggests parent waters rich in F. Changes above threshold concentration of F in Antarctica ice cores have been related to volcanism 7 . Given that Boggs Valley is located within a tectono-magmatic region that has experienced active volcanism throughout the Quaternary 8,9 , it is reasonable to infer that subglacial water bodies generated by volcanic eruptions were discharged down-glacier into Boggs Valley at peak glacial conditions.

Stable isotope signals for Boggs Valley calcites
Stable C and O isotope ratios values distribution according to facies for BV9b; BV99a2, B9a2c and BV11. Negative δ 18 O values for both Dm (red triangles) and Cs (open squares) are typical of other carbonates known to have precipitated from Antarctic glacial meltwaters 10 . The ~2 ‰ difference between δ 18 O average values of Cs and Dm can be related to a different composition of basal ice melts 11 . Negative δ 13 C of both Cs and Dm, as low as -8.4 ‰, suggests influence of microbial metabolism in subglacial environments 12 . Most commonly, calcium and carbonate ions are supplied to growing subglacial calcite surfaces via dissolution of a carbonate bedrock 13 . In the case of Boggs Valley, the bedrock is a non-carbonate and the supply of Ca can be ensured by bio-weathering of silicate minerals, whereas the CO2 must be generated by microbial metabolism 14 . The hypothesis of bio-weathering of silicates in Boggs Valley as the provider of calcium and other trace elements that are incorporated in the calcite lattice, such as Fe (see text for details) is supported by exceptionally high U concentration (Supplementary Table 1). This is uncommon for continental calcites, and was likely to have been released by dissolution of silicates. The negative  13 C values of the calcites suggests the presence of subglacial ecosystems with the characteristics of polar refugia dominated by microbial processes, leading to organic matter oxidation, remineralization of 12 C-enriched organic matter, sulphur oxidation, and possibly sulphate reduction 12 .
Cs Dm

Supplementary Figure 4
Micro X-ray Absorption Near Edge Structure (XANES) spectra of BV9b a) Fe XANES spectra. The spectra were normalized for atomic absorption, based on the average absorption coefficient of the spectral region from 7200 to 7250 eV. A reference pure standard Fe foil (Kα edge at 7112 eV) was used to provide accurate calibration of the monochromator. Measured spectra on BV9b sample are those labelled from (a) to (e), the others are reference spectra. In XANES spectra (a) to (e) the strongest peak shows a progressive shift toward lower energies and a shoulder around 7126 eV. When compared with the references, this suggests that (a) spectrum corresponds to a species with Fe (3+) in 6-coordinated sites, as for ferrihydrite (best fitting spectra for point (a)), goethite and hematite (peak at 7132 eV). For (b) to (e) spectra, there is additional contribution of Fe (2+) in 6-coordinated sites as shown by the sharpest peak position of Fe 2+ species such as hedenbergite ((Fe, Mg) CaSi2O6) and melanterite (FeSO4·7H2O) (peaks at 7126 and 7130 eV). b) Sulphur XANES spectra. The spectra were normalized on average absorption coefficient of the spectral region from 2510 to 2540 eV. A reference pure Ca(SO4)·2H2O powder was used to provide an accurate calibration of the monochromator (maximum absorption at 2482.5 eV). All samples display a strong peak at 2482.5 eV characteristic of oxidised sulphate phases (ref. Ca(SO4)·2H2O). Samples (d) to (f) show a low intensity peak at 2473.3 attributable to organic compounds, possibly amino acids (ref. cysteine). The pre-edge peak at ~2478 eV is due to photoreduction.  15 . The GHF during the initial days of the subglacial eruption reached values six to seven orders of magnitude above the typical background rates 15 . Since the lateral extent of the elevated heat fluxes could not be constrained, a range of horizontal areas based on prescribed radii (r) from an arbitrary central point was used. An ice thickness (H) of 750 m calculated for the LGM (see main text), was used to limit meltwater production in the vertical dimension. By assuming radial reduction in ice melting away from the central volcanic source, total meltwater (M) was calculated as a conical volume (V) by using Equation S1: The partial derivative in Equation S2 describes the vertical temperature gradient in the ice, which was set to the conditions described in the text. The other terms are the ice density ρi and the latent heat of freezing Li. The figure shows that volcanism elevates subglacial heat fluxes and dramatically increases the discharge if the melted ice area is in the range 10 2 to 10 6 km 2 . The horizontal black line in the figure denotes the approximate meltwater flux rate above which a significant impact on Antarctic Bottom Water formation would occur, based on climate modelling experiments 16 . The vertical dashed line (thk limit) simply means that any additional or ongoing heat will not melt any additional ice, because the ice that could be melted would already have done so. The model implies the potential for rapid transfer of hydrothermally influenced waters from the plateau along the axis of the Boggs Valley and Rennick ice streams when the hydraulic barrier was breached.

Supplementary Figure 6
Hand specimens used for ancient DNA extraction Sample BV8a and BV9b mostly consist of Cs and thin Dm layers and have similar stratigraphy. Sample BV11 is composed solely of Dm embedding angular to sub-angular sediment and particulate. Arrows show the top of the crusts.

BV8a
BV9b BV11 Supplementary "Depth (mm)" is the total extension from sample outer surface to its base. Fabrics codes are: Cs = clear, columnar calcite sparite; Cc = isopachous calcite cements; Cg = micrite coating grains (Coated grains); Dm = dirty microsparite. The U concentration is also reported where analysed. Activity ratios were determined using a Nu Plasma MC-ICP-MS at the University of Melbourne 17,18 . Ages are expressed in ka before year 2000 AD and were calculated by using the decay constants reported in ref. 19 , corrected for initial 230 Th following the procedure proposed by ref. 20 and by assuming an initial [ 230 Th/ 232 Th] of 1.5 ±1.5. "[ 234 U/ 238 U]i" is the reconstructed 234 U/ 238 U activity ratio at time of sample formation given its calculated age. Uncertainties are 95% external confidence intervals. The analyses reported in italics red have a [ 230 Th/ 232 Th] lower than 30, suggesting detrital contamination, and were excluded from the age distribution plot in Fig.4b.