Chromium Cycling in Redox‐Stratified Basins Challenges δ53Cr Paleoredox Proxy Applications

Abstract Chromium stable isotope composition (δ53Cr) is a promising tracer for redox conditions throughout Earth's history; however, the geochemical controls of δ53Cr have not been assessed in modern redox‐stratified basins. We present new chromium (Cr) concentration and δ53Cr data in dissolved, sinking particulate, and sediment samples from the redox‐stratified Lake Cadagno (Switzerland), a modern Proterozoic ocean analog. These data demonstrate isotope fractionation during incomplete (non‐quantitative) reduction and removal of Cr above the chemocline, driving isotopically light Cr accumulation in euxinic deep waters. Sediment authigenic Cr is isotopically distinct from overlying waters but comparable to average continental crust. New and published data from other redox‐stratified basins show analogous patterns. This challenges assumptions from δ53Cr paleoredox applications that quantitative Cr reduction and removal limits isotope fractionation. Instead, fractionation from non‐quantitative Cr removal leads to sedimentary records offset from overlying waters and not reflecting high δ53Cr from oxidative continental weathering.

: Dissolved oxygen (a) and turbidity (b) in Lake Cadagno during summer 2017 34 Figure S2: Turbulent diffusivity from July and August 2017 35 Figure S3: Chromium, Fe, P, Mn and Mg bi-variant plots 36 Figure S4: Cr-δ 53 Cr plot 37 Figure S5: Full sediment core [Cr] and δ 53 Cr data from Lake Cadagno 38 Figure S6: Distributions of dissolved Al in Lake Cadagno 39 Figure S7: Relationships between Al and other parameters in short core (0-10 cm) sediment 40 leaches 41 Figure Tables  45  Table S1: Water column data 46 Table S2: Sediment trap data  47  Table S3: Sediment standards  48  Table S4: Sediment leaches  49  Table S5: Sediment near-total digests 50 Table S6: Wet and dry sediment mass 51 Table S7: Landsort Deep (Baltic Sea) water column Cr data 52 Table S8: Landsort Deep (Baltic Sea) water column Fe and Mn data 53 Table S9: Sulfide, Fe and Fe/H2S ratios in redox-stratified systems included in Figure  Lake Cadagno is a 21 m deep meromictic alpine lake in Ticino, Switzerland, with a persistently 64 anoxic subsurface separated from oxic surface waters by a strong density gradient. Residence times 65 of water in the sulfidic zone have been estimated previously as 1.5 to 7.5 years (Dahl et al., 2010). 66 The lake is well-studied, with significant prior research on biogeochemical cycling and physical 67 processes in the lake. Of particular importance to this study are the general geochemical and 68 physical characterization (e.g. Del Don et al., 2001), Fe distributions and cycling in the water 69 column (e.g. Berg et al., 2016;Ellwood et al., 2019), lake physical structure (Sommer et al., 2017;70 Sepúlveda Steiner et al., 2019;2021), and the geological history and sediment geochemistry of the 71 lake (e.g. Birch et al., 1996;Wirth et al., 2013;Berg et al., 2022). The lake has also been the focus 72 of numerous prior geochemical studies as an analog system for conditions in the Proterozoic Ocean 73 (e.g. Canfield et al., 2010;Dahl et al., 2010;Xiong et al., 2019;Ellwood et al., 2019). 74 75 Despite the elevated H2S concentrations at depth in Lake Cadagno (up to ~10 2 µmol kg -1 ), 76 dissolved sulfate (~3 mmol kg -1 ) and Fe concentrations are also high (>1 µmol kg -1 , Dahl et al., 77 2010;Ellwood et al., 2019). Surface sediments of Lake Cadagno are consistent with deposition in 78 an anoxic environment, with elevated TOC and TS. Samples were diluted to 1 M HCl and equilibrated at 80° C for at least one hour, then processed 117 through cation exchange chromatography with AG50W-X8 resin (Yamakawa et al., 2009). 118 Samples were analyzed by MC-ICP-MS and found to contain insignificant levels of Cr (< 1 ng, 119 not shown). 120 121 Sediment traps were deployed on 10 July 2017 and recovered on 6 September 2017. Sample cups 122 were rinsed with filtered surface water and then centrifuged, decanted to the extent that was 123 possible without losing material, and stored in acid cleaned PP bottles and centrifuge tubes in the 124 dark in a fridge until analysis. Given the poor settling nature of the sediment, some material may 125 have been lost during trap recovery, centrifugation and decanting, and therefore sediment trap 126 fluxes represent a minimum estimate of particulate fluxes. After evaporating to dryness, samples 127 were microwave digested as with Landsort Deep water samples. Final calculated fluxes are 128 corrected for the Cr and δ 53 Cr of the surface water used to rinse the collection cups (<1 % total Cr 129 in sediment trap samples), with uncertainty estimates derived from standard error propagation. 130 Total Cr recovered in the sediment traps ranged from approximately 3000-7000 ng Cr (Table 1). 131 132 Lake Cadagno sediments were sampled using UWITEC Ltd. gravity and piston coring equipment 133 in summer 2019 and summer 2020 (Berg et al., 2022). Piston coring down to a depth of ~940 cm 134 below the sediment water interface retrieved the entire lacustrine sediment record deposited since 135 lake formation. Samples were freeze-dried and hand milled with an agate mortar and pestle. Near-136 total digests were prepared by weighing ~20 mg of sediment into pre-cleaned 14 ml Savillex 137 Teflon beakers and adding 1.5ml of inverse aqua regia. Ten drops of 30% H2O2 were added to 138 oxidize organic matter for 12 hours in the partially covered beaker. Samples were then refluxed at 139 140°C for 24 hours, dried and redissolved in 1 ml 0.5 M HNO3. Authigenic sediment phases 140 (organic matter, amorphous metal oxides and sulfides, see Figure 2, see also  Figure S1: Dissolved oxygen (a) and turbidity (b) in Lake Cadagno during summer 2017. 171 The 10° C and 5° C isotherms are shown as labelled contours. The turbidity peak reflects the 172 microbial community found just below the chemocline. CTD data are available at the following 173 Zenodo A first order estimate of the time necessary to remove all dissolved Cr from the area between the 180 10 m and 14 m sediment trap was made using 181  The measured pCrAuth flux between these traps (Table 1, pCrAuth at 14 m -pCrAuth at 10 m 182 = 4.3×10 3 ng Cr m 2 day -1 ) 183  The water volume between these two traps, based on lake isobaths data ( Table 2 in Del  184 Don et al., 2001, treating these isobaths as stacked 1 m tall disks, and therefore maximizing 185 the potential volume). Volume = 276,300 m 3 186 This dissolved Cr inventory was obtained by multiplying the water volume by the approximate Cr 187 concentration (0.40 nmol kg -1 , and assuming a density of 1 kg l -1 ), and dividing by the 188 accumulation of particulate Cr between these traps, assuming this removal flux was acting only 189 over the area at the bottom of this volume (therefore minimizing the total flux out). The result of 190 this estimate is that all dissolved Cr would be removed from between 10 and 14 m depth in Lake 191 Cadagno within  of 2) with those obtained from tracer release experiments. (e.g., Davis, 1994, Goudsmit et al., 214 1997. This agreement validates the method. For Lake Cadagno, Sepúlveda Steiner et al. (2019)  215 reported excellent agreement between microstructure diffusivities and those obtained from a tracer 216 release experiment (Wuest, 1994). 217 218 The intermittent or constantly varying nature of turbulent diffusivity is illustrated in Figure S2B. 219 This significant natural variability indicates that, while both the 1 st and 3 rd quartile Koc values are 220 equally valid measured states, for a given unit of time, transport at the higher end of the range (e.g. 221 3 rd quartile), results in significantly larger Cr flux than transport at the lower end (e.g. 1 st quartile). 222 Therefore, while the Koc distributions and a median value are also shown in Figure S2B, the 223 maximum likelihood estimation mean value (Baker & Gibson, 1987) is used for quantitative 224 treatment, to better include the impact of elevated transport at the higher end of the observed Koc 225 range. The related statistics should not be interpreted as analytical uncertainties but as a signature 226 of the natural variability of the system (turbulence in stratified fluids being intermittent). Finally, 227 our results, clearly show that the system is dominated by turbulent diffusivity rather than molecular 228 diffusivity in the surface and deep layers. 229 230 The determined turbulent diffusivity is then used to calculate expected diffusive Cr transport from 231 [Cr]-enriched deep waters to the chemocline based on the observed [Cr] gradient. 232 = 233 234 with ΔCr and average Koc taken from 13-14.5 m depth. The calculation is relatively insensitive to 235 the exact range chosen, as is relatively uniform in Lake Cadagno deep water (ranging from 236 0.9×10 -6 to 2.9×10 -6 nmol cm -4 ), with stronger Cr gradients closer to the chemocline ( Figure 1C), 237 and opposing Koc variability with depth ( Figure S2B). Given that these fluxes are estimated at the 238 order of magnitude level, taking into consideration heterogeneity of the system over the months 239 sampled and uncertainty in isolating authigenic sediment trap fractions, such variability is 240 insignificant. 241 242 243 Figure S2: Turbulent diffusivity from July and August 2017.

244
Mean water column stability (red) and temperature (blue) are shown in (a)  The burial flux of authigenic Cr in Lake Cadagno surface sediments is estimated based on 252 previously published sediment accumulation rates (4-6 mm yr -1 , Birch et al., 1996), measured 253 sediment densities and porosity (Table S6) and the authigenic Cr content of surface sediments (~8 254 ppm, Table S4) following: The resulting flux (49-74 ng cm -2 yr -1 , or 1.3-2.0 × 10 3 nmol m -2 d -1 ) is approximately 25% of the 257 estimated sinking particulate and upward diffusive fluxes near the chemocline and approximately 258 50% of the sinking particulate flux from the deepest sediment trap. This supports the release of a 259 significant fraction of Cr from particulates into deep waters during particle sinking and at the 260 sediment surface. This is also consistent with (

S.5 Discussion of authigenic corrections
In sediments and sinking particles in natural aqueous systems, Cr will be present in authigenic as well as detrital phases. However, it is only the authigenic phases that might provide insight into in situ Cr cycling and yield potential paleoproxy information. The two main and often combined strategies to isolate the authigenic signal from the composite sample matrix are chemical leaches designed to target various more labile phases, and corrections based on immobile silicate-hosted elements (i.e. Al and Ti) (Equation S1, where UCC indicates average Upper Continental Crust, e.g. Rudnick & Gao, 2014).
However, both of these approaches are prone to error. First, while leaches are designed to target specific authigenic phases, these leaches are operationally defined and attack multiple authigenic and detrital phases with varying strength (e.g. Rauret et al., 1999;Frank et al., 2019). Second, detrital corrections rely on a well-constrained ratio of Cr to the chosen normalizing detrital element, which, however, is known to vary in crustal material (e.g. Cole et al., 2017). Furthermore, such corrections assume that detrital sources strongly dominate the extracted signal of the chosen normalizing element, with negligible contributions from authigenic phases. Therefore, there is no perfect isolation of authigenic signals, and normalization approaches must be considered carefully based on the environmental conditions and available data of each scenario.
We present near-total sediment digests, sediment leaches targeting organic matter and sulfides (30% w/w H2O2 at pH = 2 with HNO3, Rauret et al., 1999), and aggressive oxidizing digestions of sediment trap material (refluxing with a mixture of 90% v/v concentrated HNO3 + 10% v/v 30% w/w H2O2 followed by microwave digests in 7 M HNO3). Because near-total sediment digests were performed to characterize the bulk sediment Cr signal, no corrections were made to these data. Previous research in black shales has shown that the lability of Al in detrital phases is more comparable to Cr than that of Ti (Frank et al., 2019). Therefore, we base corrections on Al and, regarding whether to apply Al-based detrital corrections to our sediment leaches and sediment trap digestions, we consider our leaching/digestion protocol and ancillary metal data.
Sediment trap samples (denoted as pCr in the text and figures) included poorly settling material. Therefore they were stored with some overlying supernatant present, which was eventually evaporated on a hot plate (due to elevated Cr filter blanks, e.g. Scheiderich et al., 2015, we avoided collecting particles on filters followed by filter digestion). Consequently, it was not clear that native speciation was preserved and gentle digestions were thus not applied. Instead, an approach with HNO3 and H2O2 was chosen for aggressive oxidation of the organic-rich samples and dissolution of FeMn oxides, while minimizing leaching of aluminosilicates. This more aggressive approach likely liberated some detrital Cr, and a correction is applied using average Cr/Al for upper continental crust (denoted as pCrAuth in the text and figures, Rudnick & Gao, 2014), and the δ 53 Cr of igneous material (δ 53 Cr = 0.12 ± 0.10, 2SD, Schoenberg et al., 2008). However, we note that dissolved [Al] in Lake Cadagno is clearly non-conservative, and instead reflects the formation of non-detrital Al phases ( Figure S6, see also Ellwood et al., 2019). Therefore, we caution that this correction likely overestimates the detrital contributions and therefore leads to an underestimate of the true authigenic Cr flux. Similarly, the correction likely overcorrects δ 53 Cr to excessively high authigenic values. The true authigenic fluxes and isotope compositions likely lie between the corrected and the uncorrected ones. However, given that all corrected δ 53 Cr values are within uncertainty of uncorrected values, this does not impact our interpretations. Both uncorrected and corrected data are shown in Table 1 and Figures 1 & 2.
For our sediment samples, oxidative leaches were applied to sediment subsamples (25-35 mg) in 2 mL 30% H2O2 in low molarity acid (0.01 M HNO3, pH = 2), designed to target authigenic phases. This procedure was chosen because it targets the main authigenic phases present in our samples: organic matter (≥ 14% TOC in most of our sediment samples), sulphides (1-3% TS, Figure 2 & Table S4) (e.g. Rauret et al., 1999). In addition, our sediments may contain potential fast-sinking amorphous metal oxides that survive reduction before reaching the sediment surface (Berg et al., 2022). Acidic H2O2 efficiently dissolves Mn oxides (e.g. Neaman et al., 2004). While H2O2 is known to react catalytically with rather than dissolve Fe oxides (e.g. Kwan & Voelker, 2002), and is only poorly effective at dissolving crystalline Fe oxides (Neaman et al., 2004), the acidic environment of our leach is capable of dissolving low levels of amorphous Fe oxides such as ferrihydrite (Shi et al., 2011). Given a sediment surface Fe(III) of approximately 100 µmol Fe g -1 dry sediment (Berg et al., 2022), and about 25-35 mg sediment in a 2 mL leach, this would yield dissolved Fe(III) from amorphous oxides up to ~1 mM, much lower than the dissolvable Fe(III) in this leach solution based on previous studies (e.g Shi et al., 2011). Therefore, the leach used should also effectively target authigenic Cr in the low levels of metal oxides that may reach the sediment surface at Lake Cadagno.
The weakly acidic nature of this leach, which is one to two orders of magnitude lower molarity than previous acidic leaches applied to similar sediments (e.g. Reinhard et al., 2014;Frank et al., 2019), is unlikely to strongly attack silicates. Indeed our leachate Al data do not appear to reflect extraction of detrital phases. Rather, leachate Al shows positive correlations with other elements known to have major authigenic phases in sediments (TS, Mn, Fe) and no correlation with leachate Si hosted, for instance, in clays ( Figure S7). Furthermore, we find the lowest leachate Al in the sample with by far the largest detrital composition (20.5 cm, >90% detrital, in comparison with <60% to a maximum of ~70% detrital in other samples). Therefore, we do not apply any detrital corrections to our sediment leachate data. However, for the sake of comparison, Table S4 shows the calculated values assuming all of the leached Al were detrital. Again, for the reasons outlined above, we argue this is not correct and we stress that we believe uncorrected values to more accurately reflect authigenic Cr than these recalculated values. Nevertheless, the qualitative results remain the same regardless of which data are used ( Figure S8). Namely, these are:  [Cr]Auth increases with depth  δ 53 CrAuth decreases with depth  δ 53 CrAuth is distinct from the water column, with a variable offset depending on sediment depth and water column depth (i.e. euxinic zone average or Cr removal zone).
Indeed, applying Al-based detrital corrections magnifies the downcore features. The interpretations are thus supported by 'corrected' and uncorrected data, despite clear artefacts with the 'corrected' data.
For future studies, it is important to consider these results in the context of detrital corrections using normalizing elements. Starting from the potential endmembers of: 1. A 100% digestion of sediments (whereby elements like Al will overwhelmingly show detrital control), and 2. A hypothetical perfect leach, releasing only authigenic fractions (whereby Al will entirely reflect authigenic phases) one should carefully evaluate where the chosen leach and depositional environment may fall. This should include: 1. Estimates of the degree to which the normalizing element behaves nonconservatively (and therefore how much non-detrital signal may be expected) 2. The relative magnitude of non-conservative behavior of the normalizing element and the element of interest (i.e. potential AlAuth/CrAuth). In the case of Lake Cadagno, dissolved Al shows variability on the order of 100 µg kg -1 , while Cr is on the order of 30 ng kg -1 , resulting in a high potential AlAuth/CrAuth. 3. How well the leach is believed to isolate authigenic features, and therefore the potential degree of detrital contamination. While there is no perfect solution for authigenic corrections, one must ensure that applying such a correction does not introduce more artefacts than it may correct. As leaching protocols are refined and become progressively more gentle and better extract authigenic phases while limiting detrital phase extraction, it is increasingly likely that authigenic corrections following standard elemental ratio approaches (e.g. Equation S1) may do more harm than good.   Trends are the same for both the original data and data with detrital 'corrections' based on Al. For δ 53 Cr all data are indistinguishable within uncertainty below the 2.5 cm sample. As discussed in section S.5, the Al-based 'corrections' create clear artefacts due to non-trivial authigenic Al phases, and these data therefore do not represent true authigenic Cr distributions. Rather, the original data are believed to more accurately reflect authigenic Cr.    In addition to the ferruginous Hall Lake, data from the ferruginous Lake Matano are also available (Crowe et al., 2008) but have not been included here. Lake Matano site shows broadly similar behavior (incomplete Cr removal at the chemocline, and increasing Cr below the chemocline relative to chemocline values); however, the ultramafic setting results in Cr concentrations (>100 nM) considerably higher than other modern lakes, inland seas and oceans (~0.5-6 nM).

References
For references cited in the supplemental material, see main text.