Iron mineral dissolution during permafrost thaw releases associated organic carbon

Main Manuscript for Iron mineral dissolution during permafrost thaw releases associated organic carbon Monique S. Patzner, Carsten W. Mueller, Miroslava Malusova, Verena Nikeleit, Thomas Scholten, Carmen Hoeschen, James M. Byrne, Thomas Borch, Andreas Kappler & Casey Bryce* 1 Geomicrobiology, Center for Applied Geosciences, University of Tuebingen, Sigwartstrasse 10, 72076 Tuebingen, Germany. 2 Chair of Soil Science, Technical University Muenchen, Emil-Ramann Strasse 2, 85354 Freising, Germany. 3 Chair of Soil Science and Geomorphology, University of Tuebingen, Ruemelinstrasse 19-23, 72070 Germany. 4 Department of Soil & Crop Sciences and Department of Chemistry, Colorado State University, 307 University Ave, 80523-1170 Fort Collins, US.


Introduction 32
Permafrost soils store almost twice as much carbon as is currently present in the 33 atmosphere (3,4), which is at risk of being released as greenhouse gases (i.e., CO2 or CH4) 34 during thawing. Release of greenhouse gases from permafrost soils provides a feedback which 35 will likely lead to faster warming than predicted from anthropogenic activities alone (5). This is 36 especially concerning considering that high latitude regions are experiencing significantly 37 higher rates of warming than the global average (0.6°C per decade in Northern Hemisphere, 38 1.35°C per decade in the Arctic specifically) (6, 7). However, the extent of greenhouse gas 39 emissions from thawing permafrost remains unpredictable due to knowledge gaps related to 40 controls on the fate of carbon in permafrost soils. Indeed, it has been suggested that the 41 permafrost-carbon feedback is the most important carbon-cycle feedback missing from climate 42 models and one of the greatest uncertainties in future climate predictions (8). 43 The mobility, lability and bioavailability of organic carbon in the environment is determined 44 by a number of interconnected physico-biogeochemical parameters and processes. One such 45 parameter is the presence of reactive iron minerals (defined here as iron minerals that are 46 reductively dissolved by the chemical reductant sodium dithionite, e.g. ferrihydrite or goethite). 47 These minerals are known to sorb and co-precipitate organic carbon (9) and are thought to 48 significantly influence long-term carbon storage in numerous environments (10-13). For 49 example, approximately 20% of total organic carbon in marine sediments is associated with 50 such reactive iron minerals (12). 51 8 Fig. S8). The maximum molar ratio of organic carbon to iron of 1.0, based on the maximal 156 sorption capacity of reactive iron oxides for natural organic matter (23), was exceeded in the 157 palsa transition zone and palsa mineral horizon (9. 27±2.16 OC:Fe). This suggests co-158 precipitation and/or chelation of organic compounds which can generate structures with OC:Fe 159 ratios of 6 to 10, as shown in other studies (23), and are consistent with high sodium 160 pyrophosphate extractable iron values for these palsa layers. 161

Discussion 162
Carbon binding to reactive iron minerals in the palsa area is consistent with previous 163 observations in permafrost regions of the Qinghai-Tibet Plateau (1) where Fe associated 164 carbon represents, on average, 19.5±12.3% of the total soil carbon pool in the upper 30 cm of 165 permafrost soils throughout the year. Our "space for time" approach reveals, for the first time, 166 how we may expect the dynamics of this rusty carbon sink to respond to progressive climate 167 change. This study suggests that, as soon as the conditions in permafrost peatlands become 168 water-logged, the reactive iron minerals are reduced, probably by Fe(III)-reducing bacteria, 169 and the associated carbon is released into the surrounding porewater, potentially leading to 170 greenhouse gas emissions. More work is needed to elucidate the chemical nature of mobilized 171 organic carbon to determine its lability, but our data suggest that direct chelation or co-172 precipitation of Fe-C structures play an essential role in carbon protection. 173 Our findings have far-reaching implications in understanding the carbon cycle after 174 permafrost collapse. Assuming a carbon pool of 191.29 x 10 15 g carbon in the active layer (0-175 30 cm depth) (3) in northern permafrost regions, we suggest that 13.39 x 10 15 to 38.26 x 10 15 176 g carbon could potentially be bound to reactive iron in permafrost soils. The lower estimate 177 assumes, based on our data, an average of 7% of total organic carbon is bound to reactive 178 iron in active layers underlain by intact permafrost. The higher estimate assumes a maximum 179 of 20% of total organic carbon is bound to reactive iron, based on our data and Mu et al (2016)  180 (1). If this iron-bound carbon was mobilized during thaw and is bioavailable, an equivalent of 9 2 to 5% of the amount of carbon which is currently present in the atmosphere could be emitted 182 as greenhouse gases from thawing permafrost sites as a consequence of iron(III) mineral 183 reduction. It is therefore crucial to further determine the amount of carbon bound to reactive 184 iron minerals in numerous permafrost environments, and the lability/bioavailability of this 185 carbon following its release in order to better predict future greenhouse gas emissions from 186 thawing permafrost soils and improve the accuracy of existing climate models. 187

Site description and sample collection 189
Stordalen Mire is a peatland 10 km southeast of Abisko in northern Sweden (68 22ʹ N,19 03ʹ 190 E) (24) 196 angustifolium (19, 20). In this study, the three sub-habitats were ordered following a temporal 197 succession of apparent time from "palsa", to "bog" and "fen" as has been done before (20) 198 following the classification of Johansson et al (2006) (19). The palsa and bog areas are 199 underlain by permafrost with a thickness of 10-20 m (25). The active layer, depending on the 200 surface topography, ranges from 0.5-1 m thickness at maximum thaw (25, 26). These three 201 thaw stages cover ~98% of the mire's non-lake surface (20). A thaw-dependent shift in these 202 habitats was observed from 1970 to 2000 during which palsa regions collapsed and bog and 203 fen areas increased by 17% (27). At the same time, an increase in average annual temperature 204 by 2.5°C between 1913 and 2006 was measured, resulting in an annual mean temperature 205 >0°C during the recent decades (28). The total precipitation also increased during this period 206 of time to an annual average of 306 mm (NORDKLIM, data available at 207 http://www.smhi.se/hfa_coord/nordklim). The expansion of wetlands after permafrost melt is a 208 widespread characteristic of peatlands affected by permafrost thawing (29-32) and the 209 successional shift from palsa to bog and fen areas has been documented in other northern 210 peatlands (31-34). 211 In July 2018, cores were taken in duplicates along a gently collapsing thaw gradient from palsa 212 to bog and fen (Fig. S1). Stordalen mire is a protected area with other ongoing field research, 213 thus the extent of coring is strictly limited due to the risk of accelerating permafrost thaw and/or 214 disturbance to other long-term measurements, especially at sensitive sites like erosion fronts. 215 However, extensive context data (https://polar.se/en/research-in-abisko/research-data/) from 216 the Abisko scientific community is available which ensures representative field sampling of a 217 heterogeneous permafrost area, with cores taken following a transect along the direction of 218 hydrological flow from palsa to bog and fen as described by Olefeldt and Roulet (2012) (35). 219 Given the restrictions in place, it was only possible to collect one core per location with two 220 locations sampled for each thaw stage (see also Fig. S1). A Humax corer of 50 cm length and 221 3-cm-diameter with inner core liners was used. The inner liners were washed three times with 222 80% ethanol, six times with sterile MilliQ water and sealed with sterilized butyl rubber stoppers 223 until coring. Butyl rubber stoppers were boiled three times in deionized water and sterilized at 224 121°C for 20 min in an autoclave. Sharp edges were cut into the end of the coring device to 225 help cut the peat layer. A hammer was used to further sample the active layer. Hammering 226 caused compaction of the cores. Therefore, the recorded depths are not comparable to the 227 initial soil profiles and the data is presented by different layers rather than depth. In the palsa 228 and bog area, cores were taken until the depth of the ground ice. Layers at the bottom of the 229 core which contained predominantly ice were excluded from further analyses. Therefore, the 230 soil investigated in this study represented the seasonally thawed active layer at Stordalen mire, 231 ranging from 30 to 49 cm. The cores were stored vertically at 4°C in the dark. Three cores 232 representing desiccating palsa, bog and fen were processed within 3-4 days (see also Fig. 233 S1). Due to detailed analysis of the first core set (Palsa A, Bog C and Fen E), additional cores 234 (Palsa B, Bog D and Fen F) from each thaw stage were analyzed after storage for 7 months 235 at 4°C in the dark, which is not ideal, but still could be used to determine if preservation of the 236 carbon by reactive iron was stable over longer time periods (Fig. S1). The long-term stored 237 core Palsa B still showed higher abundance of reactive iron minerals than Bog D and Fen F, 238 but less than Palsa A which could be due to natural variability, long-term storage or because 239 it was taken closer to the collapsing edge (Fig. S2). The cores A to F were compared to 240 triplicate cores previously collected in September 2017 at each thaw stage with a Pürckhauer 241 corer and processed directly after sampling, to show that the trends are representative for the 242 whole mire (Fig. S1). The replicate cores showed the same trends of 6 M HCl extractable iron. 243 Readily extractable Fe (defined by 0.5 M HCl extractable iron) showed similar trends to the 244 sodium dithionite citrate or hydroxylamine HCl extraction for all three thaw stages (Fig. S3). 245 The same trend of total organic carbon along the thaw gradient was observed ( Fig S4). 246

Porewater sampling and analysis 247
The cores were kept in a vertical position during transfer into an anoxic glovebox (100% N2). 248 Three different sections were identified by texture and color changes: (1) an organic horizon 249 on top, (2) a middle transition zone between the organic-rich and mineral-rich layer and (3) a 250 mineral horizon at the bottom (Fig. S1). Rhizon porewater samplers (Rhizosphere research 251 products, Netherlands) with a porous sampling area of 10 cm and 0.15 µm pore size were 252 used to extract porewater from three different depths, resulting in one sample representing 253 each organic horizon, transition zone and mineral horizon. The extracted porewater was 254 analyzed for dissolved Fe (total and Fe(II)), organic carbon (DOC) and fatty acids. The samples 255 were centrifuged for 5 min at 5300 g. For total Fe and Fe(II), the supernatant was acidified in 256 1 M hydrochloric acid (HCl) and quantified spectrophotometrically in triplicate with the ferrozine 257 assay (36). Dissolved OC was quantified in triplicate with a total organic carbon analyzer (High 258 TOC II, Elementar, Elementar Analysensysteme GmbH, Germany). High performance liquid 12 chromatography (HPLC; class VP with refractive index detector [RID] 10A and photo-diode 260 array detector SPD-M10A VP detectors; Shimadzu, Japan) was used to determine the fatty 261 acid concentrations. 262

Core splitting 263
The soil cores were removed from their liners under a N2 atmosphere. Each core was 264 sectioned into an organic horizon of varying thickness (4-10 cm), a transition zone (3-5 cm) 265 and mineral horizon (4-10 cm) (Fig. S1). The transition zone represents the boundary between 266 organic and mineral horizon and was additionally defined due to distinct geochemical 267 conditions in the porewater analysis in the middle of the active layer near the boundary 268 between organic and mineral horizon. Calculated bulk densities as a function of soil organic 269 matter following Bockheim et al (2003) (37) were consistent with other studies conducted at 270 Stordalen mire (38) (Palsa A: organic horizon: 0.03±0.01 g/cm 3 , transition zone: 0.08±0.02 271 g/cm 3 , mineral horizon: 0.84±0.26 g/cm 3 ; Bog C: organic horizon 0.08±0.01 g/cm 3 , transition 272 zone 1.29±0.04 g/cm 3 and mineral horizon 1.74±0.01 g/cm 3 , Fen E: organic horizon 0.21±0.02 273 g/cm 3 , transition zone 1.97±0.2 g/cm 3 and mineral horizon 1.72±0.01 g/cm 3 ). Sub-samples 274 were homogenized and weighed into 10 mL glass vials and kept frozen at -20°C prior to 275 subsequent analysis. 276

Selective extractions 277
The soil layers were subjected to several chemical extractions to quantify the different iron 278 phases. The soils were kept frozen prior to analysis, then dried at 20°C under anoxic conditions 279 until no further weight loss was observed (less than 24 hours). 0.3 g dry soil was weighed into 280 a 10 mL glass vial with 6.25 mL extractant and N2 headspace. Prior to use, all glassware was 281 washed with 1 M HCl for 10 min, flushed three times with deionized water and once with MilliQ 282 water. Afterwards glassware was sterilized at 180°C in the oven for 4.5 hours. All samples 283 were centrifuged at room temperature for 10 min at 5300 g. After centrifugation the supernatant 284 13 was decanted into another 10 mL glass vial. Each extraction was performed in duplicates for 285 each layer. Throughout the extraction, samples were kept in the dark under anoxic conditions 286 (N2 atmosphere). The extracts were analyzed for Fe and DOC as described above. 287 Additionally, the samples were acidified in 1% (v/v) HNO3 and analyzed in duplicates by MP-288 AES/ICP-MS to get the total Fe, S, P and Al concentrations (Fig. S5). The illustrated iron values 289 throughout the whole study represent the iron values obtained by the ferrozine assay (for 290 differences in iron concentrations through the different analysis see all the same iron trends with depth and along the thaw gradient (Fig. S6). For additional 296 extractant-specific experimental parameters see below. 297

M HCl 298
To quantify the total extractable Fe of the soil layers, dried samples were subjected to a 70°C 299 6 M HCl extraction for 24 h (39, 40). 300

Sodium pyrophosphate 301
The sodium pyrophosphate extraction was performed following Coward et al (2017) (10) at pH 302 10 to determine the colloidal or OM-chelated iron. 303

Hydroxylamine-HCl 304
To extract the short ranged ordered (SRO) Fe oxides, an acidic hydroxylamine-HCl (pH <2) 305 extraction was carried out under the same conditions as the sodium pyrophosphate extraction 306 (10). 307

Dithionite-citrate 308
Extractions were conducted using a solution of 0.27 M trisodium citrate, 0.11 M sodium 309 bicarbonate and 0.1 M sodium dithionite (total ionic strength: 1.85 M), as previously described 310 (12). This extraction was used to also quantify the reactive iron minerals but in particular the 311 organic molecules binding to it (released during iron mineral dissolution). Instead of heating to 312 80°C as described by Lalonde et al (2012) (12), the dithionite-citrate extraction was performed 313 under the same conditions as the sodium pyrophosphate and hydroxylamine-HCl extraction 314 (on a rolling shaker at room temperature for 16h) for better comparison between the different 315 extractions. The citrate addition as a metal ion complexing agent was necessary to avoid 316 under-estimation of iron and organic carbon as a result of complexation or mineral precipitation 317 during extraction. Without citrate addition, we obtained 64±3% less iron and 57±28% less 318 carbon after sodium dithionite reductive dissolution. As described in Lalonde et al (2012) (12), 319 we also used a 1.85 M sodium chloride/0.11 M sodium bicarbonate extraction as a control 320 experiment under the same conditions (same solid:solution ratio, temperature, time, ionic 321 strength) to distinguish between organic carbon (OC) which is readily desorbed and organic 322 carbon which is released by the reduction of iron(III) minerals. To determine the DOC 323 background concentrations caused by the trisodium citrate, blanks (trisodium citrate sodium 324 bicarbonate solution) were analyzed during each measurement. The background 325 concentration was later subtracted from the total DOC value, as well as the DOC concentration 326 of the control experiment (sodium chloride sodium bicarbonate solution), resulting in the OC 327 which is released by the reduction of reactive iron (see also Fig. S7). 328

TOC analysis 329
To quantify the total organic carbon (TOC), soil samples from each layer were dried at 60°C 330 until the weight remained constant. The dry soils were then ground and acidified with 16% HCl 331 to remove the inorganic carbon. After washing with deionized water and subsequent drying, 332 the TOC content was analyzed by an Elementar vario El (Elementar Analysensysteme GmbH, 333 Germany). The TOC content goes in line with previously reported values (41). 334

EXAFS/XANES analysis 335
Samples were dried under an N2 atmosphere and stored anoxically in a glove box prior to 336 analysis. Sample were then sealed in plastic multi-sample holders with Kapton polyimide tape 337 and kept anoxic until they were transferred to a sample mount at the beamline. The sample 338 holder was in a cryostat during analysis to limit beam damage and to prevent oxidation of 339 from chi values of 2 to 12 with an x-weight of 3. Non-negative fits were performed, and 358 components were chosen based on prior knowledge of the sample mineralogy. 359

Correlative SEM and nanoSIMS 360
The two end-members, palsa and fen, were analyzed using SEM and nanoSIMS (see also 361 Figs. S8 and S9) using only the free particles of the fine fraction of the transition zone and the 362 mineral horizon. As described by Kopittke et al (2018) (45) and Keiluweit et al (2012) (46), 363 subsamples of each layer (1 mg) were dispersed in 10 mL of anoxic deionized water and gently 364 shaken to obtain the free organo-mineral particles from the fine fraction of the soil. 100 µl of 365 the suspension was placed onto a silica wafer and dried under an N2 atmosphere. The samples 366 were sputter-coated with 12 nm platinum (Pt) using a Bal-Tec SCD005 sputter coater. 367 To characterize the organo-mineral particles of the fine fraction by size and crystallinity and 368 identify representative particles, a field emission scanning electron microscope (FE-SEM; Jeol 369 JSM-6500F), equipped with secondary electron detector, was used prior to nanoSIMS 370 analysis. The acceleration voltage was set to 5 kV, with a working distance of 10 mm. 371 The nanoSIMS analysis were performed at the Cameca nanoSIMS 50L of the Chair of Soil 372 Science (TU München, Germany). Prior to the measurements, the samples were additionally 373 coated with Au/Pd layer (~30 nm) to avoid charging during the analysis. The Cs + primary ion 374 beam was used with a primary ion impact energy of 16 keV. Prior to the final measurement, 375 any potential contaminants and the Au/Pd coating layer were sputtered away at 50 x 50 µm 376 with a high primary beam current (pre-sputtering). To enhance the secondary ion yields, Cs + 377 ions were implanted into the sample during this pre-sputtering process. The primary beam 378 (~1.2 pA) was focused at a lateral resolution ~100 nm and scanned over the sample with 12 C -379 , 16 O -, 12 C 14 N -, 31  Finally, the nanoSIMS images were analyzed using the Open MIMS Image plugin available 387 within ImageJ (available free-of-charge at, https:// imagej.nih.gov/ij/). All presented images 388 were corrected for the electron multiplier dead time (44 ns), as well as drift corrected, and the 389 planes accumulated. A median filter was applied on all images. 390

600
19° 2'41.30"E) also showed the three layers and were water saturated. The cores represent the active 601 layer in July 2018. Yellow: Core A, C and E were immediately split and processed (3-4 days). White:

602
Core C, E and G were stored for 7 months. The Scale bar = 3 cm. (E) Example of the subdivision into

603
(1) organic horizon, (2) transition zone and (3) Table S1. Absolute and % values of iron and carbon in Palsa B, Bog D and Fen F. In most of the layers, 665 the maximum molar ratio of organic carbon to iron exceeds 1.0, the maximal sorption capacity of reactive 666 iron oxides for natural organic matter (23). Co-precipitation and/or chelation of organic compounds can 667 generate structures with OC:Fe ratios up to 6 to 10 (23). Errors indicate the range of duplicate analyses 668 of each layer in each thaw stage.