Carbonate polymorphism controlled by microbial iron redox 3 dynamics at a natural CO 2 leakage site ( Crystal Geyser , Utah ) 4

16 Crystal Geyser (Utah, USA) is a CO2-rich low-temperature geyser that is studied as a natural 17 analog for CO2 leakage from carbon capture and storage (CCS) sites. In order to better constrain 18 the biogeochemical processes influencing CaCO3 precipitation at geological CO2 escape sites, 19 we characterized fast-forming iron-rich calcium carbonate pisoids and travertines precipitating 20 from the fluids expelled by the geyser. The pisoids, located within a few meters from the vent, 21 are composed of concentric layers of aragonite and calcite. Calcite layers contain abundant 22

ferrihydrite shrubs in which iron is encasing bacterial forms. The aragonite layers contain less 23 abundant and finely dispersed iron, present either as iron-oxide microspherules or iron adsorbed 24 to organic matter dispersed within the carbonate matrix. We propose that carbonate 25 polymorphism in the pisoids is mostly controlled by local fluctuations of the iron redox state of 26 the fluids from which they form, caused by episodic blooms of iron-oxidizing bacteria. Indeed, 27 the waters expelled by Crystal Geyser contain >200 µM dissolved iron (Fe 2+ ), a known inhibitor 28 of calcite growth. The calcite layers of the pisoids may record episodes of intense microbial iron 29 oxidation, consistent with observations of iron-oxide rich biofilms thriving in the rimstone pools 30 around the geyser and previous metagenomic analyses showing abundant neutrophilic, 31 microaerophilic iron-oxidizing bacteria in vent water. In turn, aragonite layers of the pisoids 32 likely precipitate from Fe 2+ -rich waters, registering periods of less intense iron oxidation. 33

INTRODUCTION 42
Carbon Capture and Storage (CCS) are climate mitigation technologies whereby carbon dioxide 43 (CO2) is captured at large emission sources (e.g., fossil fuel power plants or industrial sites), and 44 injected into deep sedimentary reservoirs for long-term storage. These negative carbon emission 45 strategies are essential tools to meet global climate goals (van Vuuren et al., 2017) but several 46 challenges and uncertainties are still limiting their application (Kelemen et al., 2019). Among 47 them, the potential for CO2 stored in deep aquifers to leak to the surface along natural fractures 48 or through injection wells and boreholes is of particular concern (Alcalde et al., 2018). 49 Naturally occurring as well as man-made CO2 leakage sites have been studied to understand the 50 fate and impact of geological CO2 in surface environments (e.g., Lewicki et al., 2006;Roberts & 51 Stalker, 2017). Crystal Geyser (Utah, USA) is an abandoned petroleum exploration well, drilled 52 in 1935, from which CO2 has been leaking to the surface for decades at a pace of ~12 kt/yr 53 (Gouveia et al., 2005). The well bore reaches a deep CO2-charged aquifer in the Navajo 54 Sandstone (Colorado Plateau), from which CO2 has also been leaking along natural faults for 55 more than 400,000 years (Shipton et al., 2004;Burnside et al., 2013). CO2 at Crystal Geyser is 56 escaping to the surface both as free gas and dissolved in brines expelled during eruptions 57 (Assayag et al., 2009;Kampman et al., 2014). It is estimated that 1 to 10% of the dissolved CO2 58 is precipitated as calcium-carbonates (travertines), forming a mound around the geyser (Shipton 59 et al., 2004;Burnside et al., 2013). 60 Here, we performed a detailed characterization of travertines and pisoids forming at Crystal 61 Geyser, in order to identify chemical and biological processes influencing CaCO3 precipitation, 62 with a particular focus on CaCO3 polymorphism. Various efforts have been made to constrain 63 Profiles of dissolved Fe(II) and total Fe (measured by colorimetry) were obtained across the field 193 site using a portable Hach DR890 instrument in 2014. pH and temperature were measured with a 194 separate Hach multimeter with a pH probe. 195

X-ray diffraction 196
Samples of the pisoids and travertines TB and TL were finely ground with an agate mortar and 197 pestle, and the powders were analyzed using a Bruker D2 Phaser operated at 30 kV and 10 mA. 198 X-ray diffraction (XRD) spectra were obtained in the 2θ range 10-65° using Cu Kα radiation (λ = 199 1.5418Å) and a Lynxeye 1D detector with a step size of 0.02° and collection time of 1 s per step. 200 X-ray absorption spectroscopy 201 X-ray absorption spectroscopy at the Fe K-edge was performed to determine iron speciation in 202 the travertines and pisoids. Samples were ground to a fine powder using an agate mortar and 203 pestle, and loaded into sample folders sealed with kapton tape. X-ray absorption spectroscopy 204 measurements were performed on beamline 4-1 of the Stanford Synchrotron Radiation 205 Lightsource (SSRL). X-ray absorption spectra were collected in fluorescence mode using a Si 206 (220) ɸ = 0 monochromator and a Lytle detector Energy was calibrated by setting the first 207 inflection point of the Fe K-edge XANES spectrum of a reference Fe 0 foil to 7112 eV. Two to 208 four spectra were collected and averaged for each sample. X-ray absorption near edge structure 209 (XANES) spectra were background subtracted and normalized to unit step edge using the 210 SIXPack software package (Webb, 2005). Extended X-ray absorption fine structure (EXAFS) 211 spectra were extracted with SIXPack using a threshold energy of 7125 eV. Previously published 212 spectra of 2-Line ferrihydrite, hematite, goethite (Maillot et al., 2011) and lepidocrocite (Pantke 213 et al., 2012) were used for comparison with the samples. 214

Raman spectromicroscopy 215
Raman spectra and hyperspectral maps were acquired on thin sections of the pisoids and 216 travertines, as well as on particles from Crystal Geyser's vent waters collected on polycarbonate 217 filters. The analyses were performed using a Horiba LabRAM HR Evolution Raman 218 spectrometer using a 532 nm frequency-doubled Nd:YAG laser and a Si-based CCD detector 219 (1024 x 256 pixels). The laser beam was focused through a 10x or 50x objective lens, yielding a 220 spatial resolution of ~5 µm or ~2 μm respectively. Spectra were collected from 80 to 1800 cm -1 . 221 A spectral resolution full width at half maximum (FWHM) of 4.5-8.4 cm -1 was obtained using a 222 600 lines/mm grating and adjustable confocal pinhole (100-200 μm). Prior to analysis, 223 calibration of the spectrometer was performed using the 520 cm -1 Raman peak of Si. Spectral 224 data were corrected for instrumental artifacts and baseline-subtracted using a polynomial fitting 225 algorithm in LabSpec 6 (Horiba Scientific). Raman maps were used to visualize the distribution 226 of mineral species using least-squares fitting. Spectra were averaged over a small portion of the 227 map containing relatively pure Raman spectra in order to define end-members. These end-228 members were then used to fit the full map dataset by classical least squared constrained by non-229 negativity of the fit coefficients using LabSpec 6 (Horiba Scientific). 230

Electron microprobe 231
Electron probe microanalyses (EPMA) were performed on a polished thin section of a pisoid. 232 The EPMA maps were collected on a CAMECA SX-Five microprobe using a LaB6 source at 233 15kV, 15nA with the beam defocused to 5 µm diameter. The stitched images are a mosaic 234 composed of 3x3 individual maps, collected at 256x256 pixels, with a step size of 10 µm and 235 dwell time of 25 ms per pixel. All elements were collected using the Kα x-ray line except Lα for 236 Sr. A LTAP crystal was used to collect Mg, and Al. A PET crystal was used to collect Ca, S, Si, 237 K, and P. A LPET crystal was used to collect Sr. A LLIF crystal was used to collect Fe. Energy-238 dispersive X-ray spectroscopy was used to collect C and O. 239

Scanning Electron Microscopy 240
Scanning Electron Microscopy (SEM) was performed on different types of pisoid samples: (i) 241 polished pisoid cuts, (ii) pisoid cuts etched with HCl, and (iii) thin sections. SEM analyses were 242 also performed on particles from Crystal Geyser vent waters collected on polycarbonate filters. 243 All samples were coated with gold prior to SEM. The analyses were conducted on a JSM-7401F 244 field emission SEM. Images were acquired in the secondary electron mode with the microscope 245 operating at 5 kV and a working distance of 6 mm, and in the backscattered electron mode at 10 246 kV and a working distance of 10 mm. Elemental analyses and maps were obtained using Energy-247 dispersive X-ray spectroscopy (EDX), performed at 20 kV with a working distance of 8 mm. 248

Focused Ion Beam 249
Specimens from a pisoid were prepared for Transmission Electron Microscopy (TEM) in an FEI 250 Helios NanoLab 600i Ga focused ion beam / scanning electron microscope (FIB/SEM) using an 251 in-situ lift-out method. Electron beam assisted deposition of Pt from a 252 Trimethyl(methylcyclopentadienyl)platinum(IV) source was used as an initial protective layer 253 over the regions of interest. This was followed by ion beam assisted deposition from the same 254 source to a total of ~ 2 μm thickness. An ion beam accelerating voltage of 30 kV was used to 255 mill around the regions of interest and for thinning of the specimens after they had been 256 extracted to a TEM grid by a nanomanipulator. Final thinning of the specimens was performed 257 using a 2kV ion beam accelerating voltage. Locations of the two FIB sections performed in the 258 pisoid are depicted in Figures S2 and S3. 259 Transmission Electron Microscopy 260 TEM analysis of the FIB sections was performed on an FEI Talos F200X instrument using a 200 261 keV accelerating voltage. Analyses included bright field and dark field imaging, high angle 262 annular dark field scanning TEM (HAADF-STEM) imaging, selected area diffraction, and EDX 263 compositional mapping. The EDX data were processed using Bruker Esprit 1.9 software. 264

Scanning Transmission X-ray Microscopy 265
Scanning Transmission X-ray Microscopy (STXM) was performed on FIB sections 1 and 2 on 266 beamline 10ID-1 (SM) of the Canadian Light Source (Saskatoon, Canada) (Kaznatcheev et al., 267 2007). Energy calibration was achieved using the 3p Rydberg peak of gaseous CO2 at 294.96 eV. 268 Images, maps and image stacks were obtained at the C K-edge, the Ca L2,3-edge, and Fe L2,3-269 edges, using a 25 nm zone plate. STXM data was processed using the aXis2000 software 270 (Hitchcock, 2012). Organic carbon maps were obtained by subtracting an image at 280 eV (pre-271 edge) and converted into an optical density (OD) image, from an OD-converted image at 288.5 272 eV (energy of the 1s→π* electronic transition in carbonyl and carboxylic groups). Carbonate 273 maps were obtained by subtracting an OD-converted image at 280 eV (pre-edge) from an OD-274 converted image at 290.2 eV (energy of the 1s→π* electronic transition in carbonate groups). 275 Calcium maps were obtained by subtracting an OD-converted image at 342 eV (pre-edge) from 276 an OD-converted image at 349.2 eV (energy of the Ca L3-edge main peak). Iron maps were 277 obtained by subtracting an OD-converted image at 700 eV (pre-edge) from an OD-converted 278 image at 710.2 eV (energy of main absorption peak in Fe-(oxyhydr)oxides). Pixels with negative 279 values in resulting maps were removed using the Clip Signal tool of aXis2000. XANES spectra 280 where extracted from aligned image stacks as described in Cosmidis & Benzerara (2014). Linear 281 background corrections were applied to the spectra in the 270-282 eV energy range at the C K-282 edge, in the in the 330-345 eV energy range at the Ca L2,3-edges, and in the 690-704 eV energy 283 range at the Fe L2,3-edges. For some image stacks, representative XANES spectra for major 284 components of the samples were extracted, and the relative contribution of these representative 285 spectra at each pixel was mapped using the Stack Fit tool of aXis2000. 286

Lipid biomarkers 287
Lipid biomarkers were extracted from pisoid and travertine samples and analyzed using gas 288 chromatography-mass spectrometry (GC-MS). 20.0 g of Crystal Geyser pisoids and travertines 289 (TL and TB) were powdered using a shatterbox and accurately weighed into 60 mL glass 290 centrifuge tubes. Each sample was spiked with 1 µg of nonadecan-1-ol internal standard. 291 Samples were extracted with organic solvent as follows: 2:1 (v/v) methanol/dichloromethane 292 (×3), followed by 9:1 (v/v) dichloromethane/methanol (×3). For each extraction, the tubes were 293 sonicated for 10 minutes in an ultrasonic bath at room temperature. Extracts were separated from 294 solid residues by centrifugation, and supernatants from each step were combined to give a total 295 free lipid fraction, comprising surface and intercrystalline lipids. This fraction is referred to as 296 the free lipid fraction. 10.0 g of the extracted residues were subsequently diluted in 297 dichloromethane-cleaned water and carefully dissolved in HCl. When there was no evidence of 298 remaining carbonate, lipids were extracted from the aqueous solutions using liquid-liquid 299 extraction (dichloromethane, ×3). Lipids adsorbed to remaining residue were also extracted using 300 sonication/centrifugation as described above, and combined to give a carbonate-bound 301 (intracrystalline) lipid fraction. 302 All extracts were concentrated to minimal volume under a gentle stream of high purity N2. A 303 portion of each extract was then subjected to acid methanolysis (0.5 N methanolic HCl, 60°C 304 ~10 h), followed by silylation (BSTFA (+1% trimethylchlorosilane) in pyridine, 70°C, 2 h). 305 Derivatized samples were analyzed by gas chromatography-mass spectrometry (Agilent 5890 306 GC hyphenated to an Agilent 5975C Mass Selective Detector). The GC was equipped with a 307 Gertsel programmable temperature vaporizer (70°C ramped to 360°C at a rate of 720°C min -1 ) 308 and a J&W 60 m capillary column (0.25 mm inner diameter, 250 µm film thickness). The GC 309 temperature program was: 70°C for 2 min, ramp at 10°C min -1 to 130°C, followed by a ramp to 310 300°C at 4°C min -1 and a final hold time of 20 min. The mass spectrometer was operated in 311 electron impact ionization mode (70 eV), with a mass scan range from m/z 50 to 600. All 312 solvents used were high-purity (OmniSolv) and all aqueous solutions were cleaned with 313 dichloromethane prior to use, and procedural blanks were run to monitor background 314 contamination. The peak areas of analytes were compared with peaks of the internal standard and 315 can be considered semi-quantitative. 316

RESULTS 317
The geochemical environment at Crystal Geyser 318 Table 1 shows concentrations of major elements measured in waters collected from the vent of 319 the geyser on two different days in April 2014, as well as in February 2015. Crystal Geyser 320 waters are rich in chloride, sodium, sulfate, calcium, potassium, and magnesium (in order of 321 decreasing abundance), consistent with a contribution from brines originating from deep 322 evaporite formations (Wilkinson et al., 2009;Kampman et al., 2014). Dissolved iron values 323 measured by ICP-MS were 6.14-10.61 mg.L -1 , falling between the range of values reported by 324 other authors (12.8-15.7   Fine-grained iron minerals were also present, but it is not clear whether they originated from the 334 geyser or if they precipitated on the filters from oxidation of Fe 2+ -rich vent water during sample 335 preparation. 336 Geochemical profiles of dissolved Fe(II), total Fe, pH, and temperature were obtained on-site in 337 April 2014, using measurements performed at different locations along the geyser's flow path, 338 from the pool around the vent to Green River (Fig. 2). The waters expelled by Crystal Geyser 339 have temperatures of ~17-18 °C and pH values of ~6.5. As they flow towards Green River, they 340 get progressively warmer (~24 °C at the river) and slightly more basic (pH ~7.8 at the river). 341 This pH increase is consistent with progressive CO2 degassing. Dissolved Fe(II) (Fe 2+ ) and total 342 Fe (Fetot) concentrations decrease with distance from the vent, showing the progressive oxidation 343 and precipitation of iron from the water. 344

Mineralogical description of the Crystal Geyser travertines and pisoids 345
Bulk mineralogical composition 346 Both aragonite and calcite were identified by XRD in the travertine samples TL and TB as well 347 as in the pisoids (Fig. 3A). Fe K-edge EXAFS was used to determine the mineralogical 348 composition of the iron-bearing phases giving the travertines and pisoids their orange color. 349 EXAFS spectra match that of a ferrihydrite reference (Fig. 3A) (with possibly a slightly more 350 disordered structure in the travertine samples as compared with the pisoids). The samples' 351 spectra were fitted with reference spectra of different iron-(oxyhydr)oxides using a linear 352 combination-least squares fitting approach. Two-line ferrihydrite provided the best fit for all 353 samples, and no additional mineral improved the quality of the fits. 354

Mineralogical mapping 355
Raman spectromicroscopy was used to map the distribution of different mineral phases in the 356 samples. In travertine TL, distinct microlaminae dominated by either aragonite or calcite can be 357 found, although both phases are present in all microlaminae (Fig. 4). Quartz grains were also 358 found in travertine TL; it is possible that these quartz grains were undetected by XRD due to 359 their low abundance relative to calcium carbonates. Travertine TB is composed of randomly 360 distributed aragonite and calcite grains, ranging from a few microns to ~100 μm in size (Fig. 5), 361 with distinct areas of the travertine displaying different relative proportions of aragonite and 362 calcite grains (compare Figs. 5D and 5G). The cortices of the pisoids are composed of concentric 363 alternating layers of calcite and aragonite, measuring a few hundred micrometers in thickness 364 ( Fig. 6,S5,S6). Interestingly, the nuclei of all the pisoids analyzed (n = 5) contain aragonite as the 365 main CaCO3 phase, along with quartz grains. Some quartz grains were also found in the cortices 366 of the pisoids. Ferrihydrite was not mapped with Raman, due to its typically weak signal, but  rich regions were visible as red or brown areas on micrographs acquired along the maps. In all 368 pisoids, ferrihydrite is mostly present in well-delimited layers spatially co-located with calcite 369 (whereas aragonite layers are ferrihydrite-free). This co-occurrence between iron minerals and 370 calcite was not observed in the travertine samples, where iron phases are more dispersed and 371 finely intermixed with the carbonates.

Morphological evidence for microbial influences in calcium carbonate and iron precipitation 430
Ferrihydrite-rich layers of the pisoids display "shrub" textures, also called "frutexites" (  (Table S1, 443 Supplementary Text). Total lipid abundances are low overall, reflecting the low abundance of 444 organic carbon in the carbonates by mass. The carbonate-bound lipid fraction of the travertine 445 TB yields 4x lower lipid concentration than the pisoids and travertine TL. Fatty acids are the 446 major lipid components detected in all samples but significant variation between each sample 447 type and free and bound fractions are observed (Fig. 10). The ratio of bound: free fatty acids is 448 4.1:1 for the pisoids, 1.4:1 for travertine TL and 1:4.3 for travertine BL. This confirms the higher 449 abundance of microbial organic matter in the pisoids and travertine TL compared with travertine 450 TB. It is not possible to identify the source of all fatty acids as many, especially n-saturated and 451 monounsaturated, are common to bacteria and eukaryotes (Table S1). However, the presence of 452 iso-and anteiso-branched saturated and 3-hydroxy acids indicates a strong contribution from 453 anaerobic bacteria (Kaneda, 1991;O'Reilly et al., 2017). n-alkanols (notably, the unusual 454 nonacosan-12-ol and hentriacontan-12-ol), β-sitosterol and stigmasterol (both C29 sterols) and 455 odd-carbon-number long chain (>C23)  water is turbulent during eruptions and "bubbling" events. The pisoids are mostly found in the 475 vicinity of Crystal Geyser's vent, with larger pisoids occurring closer to the vent, and smaller 476 ones are found more distally. For these reasons, Barth and Chafetz (2015) proposed that the 477 pisoids may be formed within the plumbing system of the geyser, and ejected during eruptions. 478 The nuclei of the pisoids often contain abundant quartz grains, a major component of the 479 sandstone formation from which Crystal Geyser's waters originate, as well as cementing calcium 480 carbonates. In all pisoids analyzed with Raman, aragonite was the only CaCO3 phase identified 481 in the nuclei (Figs. 6,S5,S6). Aragonite is also the carbonate structure found in all carbonate 482 particles filtered out of the waters expelled from Crystal Geyser's vent (Fig. S4), suggesting that 483 the nuclei of the pisoids were formed under similar geochemical conditions as these particles, i.e. 484 in the subsurface. The nuclei of the pisoids are furthermore relatively free of iron and sulfur (Fig. 485 6C,E), suggesting that they formed under reducing conditions. Indeed, in the presence of oxygen, 486 Fe 2+ dissolved in the water would oxidize and precipitate as Fe(III) phases which would be 487 incorporated in the pisoids during growth (as observed in the cortices). Similarly, in the presence 488 of oxygen, sulfide (which is also present in the water as evidenced by the characteristic sulfide 489 smell at the vent of the geyser) would oxidize as sulfate which is readily incorporated into 490 carbonates. Thus, absence of S and Fe in the nuclei of the pisoids, along with the presence of 491 these elements in their cortices, demonstrates initial formation under reducing conditions 492 followed by further growth in more oxidizing conditions. The fact that at least some pisoid 493 growth occurs above ground is furthermore evidenced by the presence of lipid biomarkers for 494 higher plants and microalgae in the carbonate-bound fraction of pisoids. Overall, results thus 495 support an initial formation of the pisoids in the subsurface in Crystal Geyser's plumbing system, 496 with some further growth after ejection at the surface. 497 The precipitation of CaCO3 minerals composing the pisoids is likely mostly driven by degassing 498 of CO2-rich waters, either in the subsurface while fluids migrate vertically during eruptions and 499 "bubbling", or in pools at the surface. Although a primarily abiotic process, microbial influences 500 on CaCO3 mineralization in the pisoids are reflected in the high proportion of bound fatty acids 501 in these objects (Fig. 10), and the presence of microorganisms encased in the carbonate matrix 502 but the results obtained here on the Crystal Geyser pisoids do not allow to determine whether 508 similar mechanism are at play here. 509

Calcium carbonate polymorphism and iron redox dynamics in the pisoids 510
Carbonate polymorphism in CO2-rich spring systems is a complex, multi-parameter problem, 511 and may be influenced by a great number of geochemical and biological factors including water 512 temperature, pH, CO2 content and degassing rate, calcium carbonate saturation state, the 513 presence of sulfate, metals and divalent ions, organics substances, and microbial mats (Chang et 514 al., 2017;Jones, 2017). The pisoids at Crystal Geyser are particularly interesting due to a clear 515 relationship between carbonate polymorphism and iron behavior. Indeed, their cortices are 516 composed of alternating layers of aragonite and calcite, suggestion shifting (bio)geochemical 517 conditions during pisoid growth. While calcite layers contain abundant iron, forming ferrihydrite 518 shrubs, aragonite layers contain only minor amounts of iron, present as ferrihydrite spherulites or 519 iron associated with clays, quartz grains, and organics (Figs. 6,7,8). Different hypotheses to 520 explain this correlation are discussed here. 521

Iron control on CaCO3 polymorphism in Crystal Geyser's pisoids 522
A first hypothesis is that iron behavior exerts a direct control on CaCO3 polymorphism in the 523 pisoids. Numerous experiments have shown that Fe 2+ is an inhibitor of calcite growth (Meyer, 524 1984;Gutjahr et al., 1996;de Leeuw, 2002;Mejri et al., 2015), promoting the precipitation of 525 aragonite over calcite. In Crystal Geyser's plumbing system, where the pisoids are thought to 526 start forming, dissolved iron (Fe 2+ ) is present at relatively high concentrations (values ranging 527 from 3.4 to 15.7 mg.L -1 have been measured by ourselves and others at the vent; Table 1 where Fe 2+ is still present (conditions similar to site 2 Fig. 2), favoring aragonite formation. 532 Ferrihydrite-rich layers in the cortices of the pisoids show that Fe 2+ is episodically oxidized (a 533 process that may be biologically mediatedsee next section), causing Fe(III) precipitation. The 534 resulting local decrease in dissolved Fe 2+ in the pools would remove calcite inhibition and allow 535 the formation of the calcitic layers of the pisoids. In some layers, calcite seems to precipitate 536 before ferrihydrite starts to form (see for instance the outermost calcite layer on top of Fig. 6A). 537 However, it is possible that ferrihydrite incorporation in the growing pisoids occurs with a delay 538 compared to Fe 2+ oxidation in solution. Indeed, observations of rusty materials as well as oily-539 looking films (often attributed to iron-oxidizing bacteria; Dyer, 2003) at the surface of stagnant 540 pools around the geyser (Fig. 1D,E)  Since at least part of the pisoids growth occurs in the plumbing system of the geyser, it is 553 necessary to consider the potential impact of physicochemical fluctuations in the subsurface on 554 carbonate polymorphism. These fluctuations are mostly driven by the eruption cycle of the 555 geyser (Kampman et al., 2014;Han et al., 2017). The temperature and pH of the geyser waters 556 are relatively constant over time and through the eruption cycle, with temperature variations 557 smaller than 3.5 °C (ranging from 15.5 to 18.8 °C) and pH variations smaller than 1.5 units 558 (ranging from 6.2 to 7.6), as measured by several authors (Baer & Rigby, 1978;Shipton et al., 559 2004;Assayag et al., 2009;Heath et al., 2009;Takashima et al., 2011b;Kampman et al., 2014;560 Emerson et al., 2016;Han et al., 2017). Temperature favorizes aragonite precipitation at values 561 greater than 35 °C, and the influence of pH on CaCO3 polymorphism is relatively insignificant 562 compared with other chemical parameters except at pH values higher than 10 (Chang et al., 563 2017;Jones, 2017). It thus seems unlikely that variations in Crystal Geyser's water temperature 564 and pH may be driving CaCO3 polymorphism in the pisoids. Similar to the effect of Fe 2+ , the 565 presence of magnesium (as Mg 2+ ions) in solution inhibits calcite growth and promotes aragonite 566 precipitation. The Mg/Ca ratio appears to be particularly important for controlling CaCO3 567 polymorphism (with higher ratios favoring the precipitation of aragonite over calcite) (Lin & 568 Singer, 2009). Mg/Ca molar ratios in Crystal Geyser's waters measured by ourselves (Table 1)  waters vary by less than 15% through several eruption periods, suggesting that sulfate fluctuation 579 in the subsurface probably do not affect CaCO3 polymorphism in the pisoids. 580 CO2 degassing rate is an important factor likely to vary dramatically over an eruption cycle of 581 the geyser. Aragonite precipitation is thought to be favored over calcite in waters with high CO2 582 degassing rates (Holland et al., 1964;Jones, 2017). This effect is consistent with the presence of 583 aragonite in the nuclei of the pisoids, forming within the plumbing system of the geyser, where 584 CO2 content and degassing rates (due to turbulent mixing during the vertical migration of the 585 fluid) are high. However, a model where CO2 degassing rate is the main driver or CaCO3 586 polymorphism cannot account for calcite and Fe(III) co-precipitation in the pisoids. Indeed, 587 intense CO2 degassing correlates with periods of water-air mixing during eruptions or 588 "bubbling", i.e. turbulent events that are also likely to cause Fe 2+ oxidation and Fe(III) 589 precipitation. Thus, if CO2 degassing were the main driver for CaCO3 polymorphism in the 590 pisoids, ferrihydrite would be mostly co-located with aragonite layers rather than calcite. 591 Physicochemical changes that may be occurring in the rimstone pools where (at least some of) 592 the pisoid growth is occurring should now be considered. Unfortunately, geochemical parameters 593 in pools were not measured in time series. However, an indication of the changes that may affect 594 Crystal Geyser water once at the surface is shown by the geochemical profile in Figure 2. 595 Depending on air temperature, the temperature of the water in the pools may increase with time, 596 but it is not likely to reach the values (well above 35 °C) where it affects CaCO3 polymorphism. 597 Similarly, pH is unlikely to show dramatic changes after a slight increase (by less than 1.5 pH 598 units) due to CO2 degassing. No data on the evolution of Mg, Sr, SO4 2-, or other species likely to 599 affect the structure of precipitating CaCO3 in pools, has been acquired. However, the correlation 600 between CaCO3 polymorphism and iron distribution in the pisoids suggest that what controls 601 shifts from aragonitic to calcitic conditions is probably a redox-active process. Iron is 602 experiencing dramatic changes at the surface due to oxidation (Fig. 2), as also shown by changes 603 in the visual aspect of the pools (Fig. 1D,E). Sulfate is also likely to improve with time in the 604 pools due to oxidation of reduced sulfur species. However, increases in which were not found in the pisoids. However, ferrihydrite in the iron-rich layers of the pisoids 628 forms "honeycomb" microtextures, encasing sub-spherical, rod-shaped or filamentous shapes 629 with sizes ranging from 0.5-2 µm (Fig. 9D-F), interpreted as iron-encrusted microbial cells 630 (Potter-McIntyre et al., 2017). The encrusted cells may correspond to iron-oxidizers of the genus 631 Sideroxydans or other members of the Gallionellales which precipitate extracellular  (oxyhydr)oxides not associated with any stalks or other recognizable extracellular structures 633 (Emerson & Moyer, 1997;Weiss et al., 2007;Fleming et al., 2014). The overall texture of the 634 ferrihydrite layers of the pisoids furthermore corresponds to what has been described as iron 635 shrubs (Chafetz et al., 1998;Chafetz & Guidry, 1999;Takashima et al., 2008;Parenteau & 636 Cady, 2010) or frutexites (Jakubowicz et al., 2014;Guido et al., 2016;Reitner et al., 2017;637 Grădinaru et al., 2020) (Fig. 9A,B), and which are commonly interpreted as microbial in origin. 638 Of particular relevance here, upward-branching iron shrubs described by Takashima et al. (2008) 639 in laminated travertines forming at the Shionoha hot spring (Japan) are composed of ferrihydrites 640 encrusting rod-shaped structures produced by microaerophilic iron-oxidizers of the genus 641 Siderooxidans. Shrub-like dendritic iron-oxide structures associated with Gallionellaceae were 642 also described in carbonates forming from CO2-and iron-rich circumneutral hot springs at 643 Okuoku-hachikurou Onsen (Japan) by Ward et al. (2017). In other hot spring laminated 644 travertines (Ilia Hot Spring, Greece), iron shrubs are associated with iron-oxidizing 645 Zetaproteobacteria (Kanellopoulos et al., 2019). Iron shrubs can also be formed by 646 microorganisms other than microaerophilic iron-oxidizers. For instance, in microbial mats 647 forming in iron-rich hot springs (Chocolate Pots, Yellowstone National Park), iron shrubs are 648 produced by cyanobacteria such as Oscillatoria, Synechococcus, and Cyanothece encrusted with 649 ferrihydrite (Trouwborst et al., 2007;Parenteau & Cady, 2010). 650 The iron-rich layers in Crystal Geyser pisoids may thus record changes in the abundance and 651 activity of neutrophilic microaerophilic iron-oxidizers, episodically precipitating Fe(III). 652 Abundances of these microorganisms can fluctuate in the environment due to a number of 653 physicochemical factors that may include availability of complex organic carbon, iron 654 abundance, and the steepness of the redoxcline (Fleming et al., 2014;Blackwell et al., 2019). At 655 Crystal Geyser, the eruption cycle is likely to be an important factor controlling variations in the 656 abundance and activity of iron-oxidizing bacteria. At the surface, iron-oxidizers may bloom after 657 each eruption of the geyser, introducing reduced iron from the subsurface. In the subsurface, 658 iron-oxidizers may be active during eruptions of bubbling events when turbulent mixing 659 introduces oxygen in water. 660 In iron-poor (aragonitic) layers of the pisoids, iron is present as ferrihydrite spherules (Fig. 7), 661 Fe(III) in clays and coatings of quartz grains, but also associated with organic matter found in the 662 porosity of the carbonate matrix (Fig. 8). Fe(III) has a strong affinity for organic matter, 663 adsorbing on negatively charged functional groups such as carboxylates or phosphorylates 664 (González et al., 2014), and frequently forms organo-ferric colloids in the environment (Ilina et 665 al., 2016;Liao et al., 2017). The origin of the ferrihydrite microspherules is more enigmatic. 666 Since they are included in the aragonite matrix, they are likely to have formed in solution prior to 667 CaCO3 formation. The presence of some space between the spherules and the carbonates (where 668 contaminating organic matter could accumulate during the FIB milling process; Fig. 7D,K,M), 669 suggests some shrinking after their incorporation within the aragonite matrix. Spheroidal 670 ferrihydrite particles were observed to form aggregates around bacteria in iron-rich laminated 671 carbonate spring deposits (Takashima et al., 2011a) but an abiotic origin for the ferrihydrite 672 microspherules in the Crystal Geyser pisoids cannot be discounted. 673

Microbial influences on CaCO3 precipitation and polymorphism in Crystal Geyser 674 travertines 675
Evidence for microbial influence on travertine formation 676 Travertine formation at Crystal Geyser is most probably a primarily abiotic process resulting 677 from CO2 degassing. It is unclear what impact microbial activities may have on the intensity of 678 CaCO3 precipitation in such CO2-rich environment. However, microbial influences on CaCO3 679 mineralization have been documented for many travertine systems (Shiraishi et al., 2008;Perri et 680 al., 2012;Okumura et al., 2013a;Kano et al., 2019;Della Porta et al., 2021), producing 681 recognizable sedimentary fabrics and textures (Guo & Riding, 1992;Kano et al., 2019). In 682 Crystal Geyser travertines, such microbial influences are thought to be responsible for 683 lamination, microstromatolitic horizons, and other features such as botryoidal carbonate textures 684 (Takashima et al., 2011b;Barth & Chafetz, 2015). Lipid analyses have shown abundant bound 685 fatty acids in the laminated travertines (TL), while most lipids in the non-laminated travertine 686 (TB) were free (Fig. 10). This difference may indicate more important contributions of 687 microorganisms to CaCO3 precipitation in travertine TL, as compared with travertine TB, in 688 agreement with the absence of lamination in the latter sample. However, the presence of 689 fenestrae and calcified bubbles in travertine TB (Fig. 5A) indicates gas formation concurrent 690 with CaCO3 precipitation, possibly resulting from microbial activity (e.g., O2 production by 691 aerobic phototrophs; Bosak et al., 2010;Della Porta et al., 2021). 692

Origin of CaCO3 polymorphism in the travertines 693
Iron oxidation is not likely to be important factor controlling CaCO3 polymorphism in the 694 travertines, which are formed at a distance from the geyser, where the well-oxygenated waters 695 are Fe 2+ -poor (conditions similar to site 3 in Fig. 2). Ferrihydrite is present at relatively low 696 abundances (compared with the pisoids) and there is no correlation between iron distribution and 697 which was interpreted as differential precipitation kinetics resulting from different levels of 724 enzymatic activities (Clarà Saracho et al., 2020). It can be concluded that different types of 725 microbial processes may be responsible for the small-scale variations in CaCO3 polymorphism in 726 Crystal Geyser travertines, but that, as opposed to the pisoids, iron redox dynamics is not an 727 important factor. Barth & Chafetz (2015) reported the presence of iron-rich tube-like structures 728 morphologically similar to the iron sheaths produced by microaerophilic iron-oxidizing bacteria 729 such as Leptothrix ochracea (Fleming et al., 2011) in Crystal Geyser's travertines. However, the 730 absence of well-defined ferrihydrite layers suggests that the travertine-formation area does not 731 experience intense blooms of iron oxidizers. Moreover, iron oxidizers were not detected in 732 bacterial 16s rDNA phylotype analyses performed on a Crystal Geyser travertine sample by 733 Takashima et al. (2011), showing that they are not dominant members of the bacterial 734 community thriving at the surface of these travertines. 735

CONCLUSIONS 736
Pisoids and travertines formed at Crystal Geyser, a natural analogue for CO2 leakage at CCS 737 sites, were characterized. Microbially driven iron oxidation was shown to exert a strong 738 influence on CaCO3 polymorphism, as recorded in the pisoids. In the travertines, microbial 739 activity may produce small-scale variations in CaCO3 polymorphism, and textural features such 740 as laminations and calcified gas bubbles. A control on CaCO3 polymorphism by iron redox 741 dynamics was shown here for the first time in a natural environment. Microbial iron oxidation 742 may play an important role in controlling polymorphism of the carbonate products of CO2 escape 743 from geological storage, and may also be relevant to subsurface carbonation at CCS sites. 744 Indeed, blooms of iron-oxidizing Betaproteobacteria have been occurring following CO2 745 injections at a geological CO2 storage site (Trias et al., 2017).        Table   Table S1. Free and carbonate-bound lipids in Crystal Geyser pisoids and travertines TL and TB.           Table S1. Free and carbonate-bound lipids in Crystal Geyser pisoids and travertines.

Supplementary Text: Detailed interpretation of lipid analyses
Fatty acids, ranged from 14 to 30 carbon chain lengths, were a major lipid component in all samples. Monounsaturated and methyl-branched (typically iso and anteiso chain positions) were present as well as n-saturated fatty acids. The highest abundances of fatty acids were observed in the bound lipid pools of the pisoids and travertine TL. Some of these fatty acids (especially nsaturated and monounsaturated) are common to both bacteria and eukaryote domains, so there is some uncertainty in source assignment. Methyl-branched fatty acids are primarily sourced from bacteria (Kaneda, 1991). 3-hydroxy acids were also identified in relatively minor amounts in all samples apart from the free lipid pool of travertine TL. Carbon chain lengths ranged from 12 to 18, and methyl-branched isomers were a major component, particularly iso-and anteisoheptadecanoic acid. These are derived from gram negative bacteria bacteria, as components of lipopolysaccharides, and are often associated with anaerobic heterotrophic bacteria (Rietschel, 1976;Wang et al., 2016).
Monoalkyl glycerol monoethers (MGM) were present in minor amounts in the bound lipid fraction of travertine TL and in the free lipid fraction of travertine TB. These are bacterial lipids, and based on current evidence are particularly common in aquatic extremophiles and heterotophic mesophiles engaged in sulfur cycling (particularly sulfate reduction) (Rütters et al., 2001;Hernandez-Sanchez et al., 2014). MGM are rarely reported in non-aquatic settings, and a large contribution from methyl-branched members is indicative of sedimentary (probably suboxic/anoxic) bacteria rather than aerobic aquatic bacteria. The presence of MGM in the travertines and their absence in the pisoids suggest that the microbial communities present during precipitation of pisoids and travertine minerals are different.
n-alkan-1-ols between 12 and up to 24 carbon chain lengths are majorly associated with photosynthetic microalgae, while n-alkan-1-ols between about 24 and 32 carbon chain lengths are majorly associated with higher plants (Pancost & Boot, 2004;Volkman, 2006). Long chain n-alkan-12-ols are rarely reported and appear to be quite restricted to cuticular wax lipids from Tamarix-type species (Basas-Jaumandreu et al., 2014), which are abundant in the region of Crystal Geyser. As such these likely reflect aeolian-sourced inputs. Minor amounts of alkan-12ols were identified in all free and bound lipid fractions, although the abundance in the bound lipid fractions was generally over an order of magnitude lower. Interestingly the ratio of hentriacontan-12-ol to nonacosan-12-ol was quite varied (0.4 to 6.7, Table S1). Assuming, similar degradation rates, this indicates distinct differences in the input of different vascular plant species and/or pathways between samples.
Sterols were abundant in all free lipid fractions, and are diagnostic for higher plants and microalgae (Volkman et al., 1998;Volkman, 2006). These likely reflect aeolian input from regional vascular plants and photosynthetic microalgal biomass. Cholesterol is the sole sterol in animals and heterotrophic microeukaryotes, as well as a relatively minor sterol in certain photosynthetic microalgae. The presence of cholesterol amounts similar to C29 sterols likely reflects input from heterotrophic microeukaryotes such as protists. Cholesterol was a major sterol in pisoids samples and likely reflect aquatic protists. High concentrations of stanols relative to stenols is generally indicative of reducing environments and significantly anaerobic bacterial hydrogenation reactions (Wakeham, 1989). Cholestanol/cholesterol ratios for the pisoids are <0.1, 0.5 for TB FL fraction, and > 1.0 for TL BL and TB FL. Thus, while the absence of sterols in bound lipid fractions may be related to low original concentrations, it may also reflect microbial activity under reducing conditions during mineral precipitation.
Hydrocarbons were identified in the free lipid fractions of all samples, and were dominated by long chain odd carbon number n-alkanes, particularly heptacosane, nonacosane and hentricontane. This is strong evidence of vascular plant wax lipid input (Eglinton & Hamilton, 1967;Bush & McInerney, 2013). One methyl-branched alkane was found in the free lipid pool of travertine TB. This is tentatively identified as 2,2-dimethyl-hexadecane and indicates the presence of cyanobacteria (Gomes et al., 2020). Cyanobacterial hydrocarbons were not identified in any other sample, suggesting they play a minor role at Crystal Geyser, apart from a possibly role in mineral precipitation in travertine TB.
Archaeol (di-O-phytanylglycerol) is a membrane lipid found in certain archaea (Kate, 1993). It was found in relatively high abundances in the pisoid free lipid fraction. This most likely reflects aquatic, possibly extremophilic, archaea living in the water close to close to the geyser. The fact that it was not detected in the bound lipid fraction of the pisoids indicates that these archaea are associated with the aqueous phase and as detritus on mineral surfaces but are not closely associated with mineral precipitation in the Crystal Geyser pisoids.