Earth and Planetary Science

Iron isotope compositions (expressed as δ 56 Fe) in sedimentary pyrite have been widely used as tracers of redox and chemical evolution of the ocean through geological time. Previous studies mostly built on the mechanical extraction of sulﬁdes from bulk rock samples, and focused on visible macroscopic pyrites, which may introduce a sampling bias. In situ analyses of micropyrite grains can provide new insights into the processes of pyrite formation and their time evolution. Here, we compile ca. 2000 in situ iron isotope compositions of Archean to Paleoproterozoic sedimentary pyrite, from previous literature as well as new data. Contrasting with bulk analyses, micropyrite displays a large and constant range of δ 56 Fe values, from -4 to +4 (cid:2) , through time. Micropyrite δ 56 Fe values are not signiﬁcantly inﬂuenced by metamorphic grade. A bimodal distribution of positive versus negative δ 56 Fe values can be attributed to two different processes of pyrite formation, Fe (oxyhydr)oxide sulﬁdation, versus kinetic and possibly microbially mediated pyrite precipitation. These processes are tightly related to rock lithology and thus to sedimentary conditions, and have existed since 3.8 Ga. © 2023


Introduction
Iron (Fe) was a dominant chemical element of the early ocean, as illustrated by the deposition of abundant Fe-rich sedimentary rocks such as banded iron formations (BIFs) through the Archean (Klein, 2005;Bekker et al., 2010;Konhauser et al., 2017). Fe played a key role in the development of early life (Lyons et al., 2015), and likely supported microbial activity, in particular anoxygenic photosynthesis and dissimilatory iron reduction (DIR). Some microorganisms can use Fe(II) as an electron donor for anoxygenic photosynthetic carbon fixation, others use Fe(III) as an electron acceptor in anaerobic respiration (Johnson and Beard, 2006). Phylogenetic studies highlight the great degree of conservation of genes responsible for most metabolic reactions involving Fe, suggesting that they evolved close to the root in the tree of life (Lonergan et al., 1996). Moreover, iron metabolic activities can induce Fe isotope fractionation, which can be recorded in Fe mineral phases through the geological record . DIR signatures were evidenced in sedimentary sulfides as old as 3.28 Ga (Marin-Carbonne et al., 2020) and more broadly at 2.7 Ga (Archer and Vance, 2006;Craddock and Dauphas, 2011), while fingerprints of anoxygenic photosynthesis have also been reported in the Archean sedimentary record (Czaja et al., 2013;Rego et al., 2021). Significant Fe isotope fractionations occur during oxidation of Fe(II) to Fe(III) (Welch et al., 2003;Balci et al., 2006) and reduction of iron (oxyhydr)oxides (Crosby et al., 2005(Crosby et al., , 2007Wiederhold et al., 2006;Frierdich et al., 2019). Therefore, Fe isotope composition has been extensively used as a proxy for reconstructing redox and chemical evolution of the ocean, together with co-evolution of early living organisms (see review in Anbar and Rouxel, 2007;Johnson et al., 2008;Dauphas et al., 2017).
In a pioneer work, Rouxel et al. (2005) measured bulk Fe isotope composition of pyrite grains in black shales, from late Archean to Paleoproterozoic, and discovered a negative isotope excursion (i.e., enrichment in light Fe isotopes) just before the Great Oxygenation Event (GOE, 2.4 Ga;Bekker et al., 2004). The negative δ 56 Fe values recorded in Neoarchean pyrite were interpreted as inherited from extensive seawater Fe(II) oxidation and Fe (oxyhydr)oxides precipitation, subsequently reduced to iron sulfides https://doi.org/10.1016/j.epsl.2023.118070 0012-821X/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons .org /licenses /by /4 .0/). (Rouxel et al., 2005). The evolution to positive δ 56 Fe values during Paleoproterozoic was proposed to reflect the onset of oxidative weathering in the wake of the GOE and increased sulfate delivery to the ocean. Alternatively, the observed negative δ 56 Fe values could reflect the onset of DIR-bacteria, which use ferric iron as an electron acceptor and release light Fe(II) to sediment porewater, from which light pyrite can precipitate . Kinetic effects during pyrite precipitation from aqueous FeS precursors could also drive δ 56 Fe variability, with Fe isotope fractionations as low as -3 produced during abiotic pyrite precipitation (Guilbaud et al., 2011). The expression of kinetic versus equilibrium effects during pyrite precipitation is controlled by the Fe/S ratio and the availability of organic material (Mansor and Fantle, 2019). The formation of pyrite follows a complex crystallization pathway with several steps and various soluble and insoluble precursors (Rickard et al., 2017) that can obscure Fe isotope composition record. Recently, Heard and Dauphas (2020) compiled bulk δ 56 Fe values in 3.3 to 1.8 Ga-old pyrite, and highlighted a secular evolution. From Archean through the GOE, δ 56 Fe values are mostly negative and progressively increase, before reaching positive values after the GOE. This trend, combined with the evolution of iron formation δ 56 Fe and S isotope compositions in sedimentary sulfide from early Archean to late Paleoproterozoic, was proposed to reflect Fe and S availability control over δ 56 Fe, and to point to a kinetically driven pyrite precipitation indirectly linked to the redox change of the ocean. Most of these previous studies have focused on millimeter-scale pyrite selected for bulk δ 56 Fe measurement, as large grains are more easily mechanically extracted from their matrix. However, this approach may introduce a sampling bias since only large visible pyrite grains are selected for analysis. Moreover, very few data in pyrite older than 3 Ga are available from bulk measurement, leaving most of the Archean period uncovered. In situ analyses of pyrite at the micrometer scale allow to study a larger variability of samples, even in rocks with limited amount of pyrite grains. Microscale δ 56 Fe values previously measured in Precambrian pyrite show a wide range, including positive values throughout the Archean (e.g., Yoshiya et al., 2012;Agangi et al., 2015;Yoshiya et al., 2015b;Marin-Carbonne et al., 2014, potentially controlled by local environmental variations (e.g. Decraene et al., 2021b). In situ δ 56 Fe database might bring new perspective on the Fe isotope variations in Archean and Paleoproterozoic pyrite. In the present contribution, we report the first compilation of in situ Fe isotope compositions in pyrite, throughout Archean and Paleoproterozoic. Our compilation includes data from previous studies (Table S1) where x is 6, 7 or 8. Fe isotope composition is usually only expressed as δ 56 Fe, as most isotope variations are mass-dependent (e.g., Dauphas et al., 2017). The variability of isotope compositions in natural reservoirs is inherited from isotopic exchanges during chemical and/or physical reactions. This includes various processes such as redox reactions, mineral precipitation, adsorption onto mineral or biological surfaces, organic complexation and/or biological uptake. Isotope fractionations can occur at equilibrium, in which case the reactant and product can freely exchange isotopes. This is typically the case of aqueous Fe(II)-Fe(III) exchanges, which are particularly rapid (i.e., a few seconds; Welch et al., 2003). Alternately, unidirectional processes and off-equilibrium reactions are accompanied by kinetic isotope fractionation. Mineral precipitation under specific conditions can be kinetically controlled, as illustrated in the case of hematite (Skulan et al., 2022) and iron sulfide precipitation (Guilbaud et al., 2011). The largest iron isotope fractionations are observed during redox reactions, iron oxidation (Croal et al., 2004) and reduction (Crosby et al., 2005). Iron isotope fractionations of various physico-chemical processes have recently been reviewed in Dauphas et al. (2017) and Johnson et al. (2020).

Box 2: Comparison of Fe isotope analytical methods
Iron isotope analyses were first performed by thermal ionization mass spectrometry (TIMS). These early works suffered from poor thermal ionization efficiency of Fe and variable instrumental mass fractionation during measurement (Johnson et al., 2020).
Nowadays, multi-collector inductively coupled-plasma mass spectrometry (MC-ICP-MS) is widely used for iron isotope analysis. Benefits of this method include high ionization efficiency of iron by argon plasma, and stable instrumental mass fractionation which can be corrected for by sample-standard bracketing (Dauphas and Rouxel, 2006). Sample preparation for MC-ICP-MS measurement of iron isotopes involves mineral separation from the matrix, dissolution and separation of Fe by ion-exchange chromatography. Mineral separation can be performed chemically, by sequential extraction of the various iron-bearing mineral phases (Huerta-Diaz and Morse, 1990;Poulton and Canfield, 2005), or by mineral grain separation such as rock crushing and grain handpicking (Rouxel et al., 2005). With this approach, Fe isotope composition can be measured with a precision of 0.1 /amu .
Alternatively, iron isotope analyses can be performed by in situ measurement on mineral grains using secondary ion mass spectrometry (SIMS) and ns-or fs-laser ablation MC-ICP-MS (LA-MC-ICP-MS). These techniques allow for sputtering of the mineral of interest at the sample surface and subsequent isotope analysis. For iron isotope analysis by SIMS, the primary ion beam is typically focused to a 15 μm-diameter spot, sputtering the sample surface over a zone of 3 to 10 μm and forming a pit from less than 1 μm to a few μm in depth (Marin-Carbonne et al., 2020;Decraene et al., 2021a). Reproducibility better than 0.2 (2σ , 56 Fe/ 54 Fe) can be achieved (Galić et al., 2017;Marin-Carbonne et al., 2020;Decraene et al., 2021a). The laser spot diameter during fs-LA analysis is similar to that of the primary ion beam, however the pit depth can be up to tens of μm, resulting in 6-fold higher sample consumption with a similar reproducibility (Zheng et al., 2018).

Data compilation
Both published and new in situ Fe isotope compositions of iron sulfides from Archean to Proterozoic were compiled (see the supplementary materials). The measurements were performed by secondary ion mass spectrometer (SIMS), or laser ablation multicollector inductively coupled plasma mass spectrometer (LA-MC-ICP-MS). New data were measured on samples from two IPGP drillcores (TCDP1 and PPDP2), from the BARB3 drillcore (ICDP, 2008) and the Francevillian Group. All samples are described in the supplementary materials. The compilation primarily includes the age, size and morphology of pyrite together with the lithology of the host rock (Table S1). The metamorphic grade and organic matter content were also compiled when available.

SIMS analyses
Iron isotope compositions were measured with a Cameca ims 1270 ion microprobe at CRPG-CNRS (Nancy, France) for TCDP samples, at UCLA (Los Angeles, USA) for BARB3 samples, and with a Cameca 1280 HR at the SwissSIMS (Lausanne, Switzerland) for the samples from the Francevillian Group, following the procedures described in Marin-Carbonne et al. (2011) and Decraene et al. (2021a). Pyrite grains were sputtered by a 16 O − primary beam of 10 nA intensity focused to a spot of about 10 to 15 μm. 54 Fe + and 56 Fe + were measured in multi-collection mode with two off-axis Faraday cups, with a mass resolving power of ∼7,000. The isobaric interference of 54 Cr on 54 Fe was monitored at masses 52 and 53, but chromium levels were negligible in all samples. The internal precision for δ 56 Fe values was typically better than ± 0.1 (2σ ), and the external reproducibility based on multiple measurements of our pyrite reference material (Balmat with δ 56 Fe = -0.399 ) was better than ±0.2 (2σ ).

Statistical analysis
The database presented in this study comprises isotope compositions reported in 12 scientific publications, in addition to our new data (see supplementary material). All data are considered, without being filtered, although the contributions of each publication are unequal among the studies and/or geological formations (from 16 to more than 200 reported δ 56 Fe values). Iron isotope compositions are displayed as distributions, in the form of histogram and often probability density. The former allows to appreciate the amount of data per bin, while the latter is necessary as it reflects both measurement values and errors, which are different for SIMS and LA-MC-ICP-MS. Gaussian function was used for probability density calculation (Equation (2), where μ is the expected value and σ the standard deviation).
For histograms and probability density distributions, a small bin size was carefully chosen, so that the distributions do not vary with bin boundaries. However, the small bin size adopted here can induce minor peaks that might not be representative. Histograms and probability density distributions are shown with 0.15 bin size, in a range from -5.5 to +5.5 . Finally, the nonparametric Mann Whitney U test was used to evaluate the probability of two independent non-gaussian distributions to be equal.

Results
About 2000 Fe isotope compositions of synsedimentary or diagenetic pyrite grains were compiled, with values between -4.87 and +5.04 ( Fig. 1, Table 1). This wide range is roughly constant throughout the studied interval (3.8 to 1.8 Ga). The widest range is observed between 2.8 and 2.6 Ga, concomitantly with the GOE, but it can simply reflect the large number of available data (1/3 of the data were measured in 2.8 to 2.6 Ga-old pyrite). The medians show no specific time evolution (Fig. 1), strongly contrasting with those compiled for bulk δ 56 Fe values of pyrites (Heard and Dauphas, 2020).
Information on the pyrite grain size was not available in all previous publications. Among the grains for which the size was specified, most of them are millimeter-sized (Fig. 3). These large grains have δ 56 Fe values ranging from -4 to +4 , with most values clustering between -2 and +2 . δ 56 Fe values of 0.1 to 1 mm-sized grains range from -3 to +1.5 . Mid-size (50 to 100 μm) grains are mostly positive, with values up to +3 , while small (<50 μm) grains span a δ 56 Fe from -2 to +4 .
Pyrite compiled in the present study experienced various metamorphic grades. The range of δ 56 Fe values in amphibolite facies is slightly narrower compared to prehnite-pumpellyite and greenschist facies, but the corresponding number of data is also lower (316 vs > 700; Fig. 4).
Fe isotope composition of pyrite depends on host-rock lithology (Fig. 5). The distributions observed in stromatolites and cherts are bimodal and similar (Mann Whitney U =102113, p=0.86, twosided). In contrast, carbonates and BIFs show a monomodal and more restricted range from -3.4 to +0.57 and from +0.25 to +2.35 respectively. Interestingly, carbonates display mostly negative values whereas BIFs exclusively contain positive values. It is worth noting that the number of analyses for these lithologies is the smallest (n=45 for carbonates, and n=90 for BIFs). Detrital sediments show a large multimodal range of distribution from -4.18 to +5.04 .
The δ 56 Fe temporal evolution is represented in Fig. 6 for various intervals including Eoarchean (>3.6 Ga), Paleoarchean (3.6-3.2 Ga), Mesoarchean and Neoarchean (3.2-2.8 and 2.8-2.5 Ga). Mesoarchean and Neoarchean intervals were merged because only 11 data were available for Mesoarchean pyrites, at 2.9 Ga. The Paleoproterozoic was subdivided as Siderian and Rhyacian (2.5-2.05 Ga), and Orosirian and Statherian (2.05-1.6 Ga). Data in the Eoarchean and Paleoproterozoic time bins only include 1 or 2 locations, while data in the Paleo-, Meso-and Neoarchean intervals comprise 8 different locations, therefore being statistically more robust. During Eoarchean, a negative and two positive modes are observed, around -1, 0.75 and 1.8 . In the Paleoarchean, negative values are mostly comprised between 0 and -2 , with a mode centered around -1.5 , and one main positive mode is observed at 0.75 . Between Mesoarchean and Neoarchean, the negative mode at -1.

Discrepancy between bulk and in situ δ 56 Fe evolution
The large and constant range observed for the secular evolution of in situ pyrite δ 56 Fe values ( Fig. 1) is at odds with the shift from negative to positive values visible in the bulk δ 56 Fe compilation from Archean to late Paleoproterozoic (Heard and Dauphas, 2020). For pyrite older than 2.96 Ga, a comparison between bulk and in situ δ 56 Fe values is meaningless, due to the difference in number of data (4 vs 690). The distinct negative δ 56 Fe excursion between 2.7 and 2.5 Ga determined from bulk analyses has no equivalent in in situ data. This discrepancy could arise from a loss of spatial δ 56 Fe heterogeneity when preparing samples for bulk analyses (Yoshiya et al., 2015b). Alternatively, it may reflect a sampling bias. Large pyrite grains are routinely analyzed in the case of bulk measurements, as those are easily extracted from their matrix. The compilation presented by Heard and Dauphas (2020) almost exclusively includes millimeter-scale pyrite grains, largely hosted in black shales (see Rouxel et al., 2005;Archer and Vance, 2006;Hofmann et al., 2009;Busigny et al., 2017;Eickmann et al., 2018). . Hence, grain size could be an additional cause for sample bias, although not detectable from the data compiled herein (Fig. 3). These aspects should be further explored in the future, by combining both in situ and bulk isotopic analyses on the same rock samples.

Limited influence of metamorphism on pyrite iron isotope composition
Pyrite samples considered in the present compilation experienced various metamorphic grades, from prehnite-pumpellyite to amphibolite facies conditions. Hence, the effect of post-depositional metamorphism on δ 56 Fe values must be assessed. At high temperature, isotope exchange can proceed and eventually reach equilibrium. This type of exchange homogenizes isotope ratios, leading to narrow Fe isotope distributions (Hyslop et al., 2008).
The large δ 56 Fe variability observed in pyrite from all metamorphic grades (Fig. 4) indicates that it has not been reset, as also suggested in previous SIMS studies (Whitehouse and Fedo, 2007;Galić et al., 2017;Marin-Carbonne et al., 2020). However, decreasing standard deviation with increasing metamorphic grade may reflect a slight influence of metamorphism. For the amphibolite facies, most data were obtained from a small number of samples from the Isua Greenstone Belt (3.8 Ga, Greenland) and the Talvivaara Formation (1.95 Ga, Finland), which may not be widely representative. Interestingly, both modes are present in the δ 56 Fe distribution of pyrite from the Isua Greenstone Belt (Fig. 6a, b). This bimodality can hardly be explained by a unique process, such as metamor- phism. Therefore, we conclude that primary pyrite δ 56 Fe signatures were likely preserved in Archean and Proterozoic rocks, even in those which experienced amphibolite-facies metamorphism.

Pyrite records partial iron oxidation
Strictly positive δ 56 Fe values are observed in pyrite from BIFs ( Fig. 5), most likely reflecting Fe-(oxyhydr)oxides reduction in the sediments and sulfidation. In detail, aqueous Fe(II) was first supplied to the ocean by hydrothermal vents with δ 56 Fe of -0.5 to +0.1 (Rouxel et al., 2005), or by continental weathering (Decraene et al., 2021b). Fe(II) from the deep waters was transferred upwards by upwelling and ocean circulation, where it could be partially oxidized by photo-oxidation, by photosynthetically produced O 2 , and/or by anoxygenic photosynthesis (Dauphas et al., 2004;Bekker et al., 2010;Konhauser et al., 2017). Aqueous Fe(III) readily precipitated as iron (oxyhydr)oxide, with an apparent isotope fractionation of +1 to +1.5 (Bullen et al., 2001;Croal et al., 2004;Nie et al., 2017), producing positive Fe isotope compositions as commonly measured in BIF Fe (oxyhydr)oxides (Planavsky et al., 2012;Dauphas et al., 2017). In this context, BIF pyrite grains could form through complete heavy Fe (oxyhydr)oxides sulfidation, thus retaining their initial positive isotope composition. Positive δ 56 Fe modes are also present in the distributions of stromatolites and cherts, and we suggest that the corresponding pyrite particles derive from heavy Fe-(oxyhydr)oxides. In summary, positive Fe isotope compositions in BIFs, stromatolites and cherts from various geological formations point to partial aqueous Fe(II) oxidation in the Archean oceans, a phenomenon which was often suggested in the literature (e.g., Ostrander et al., 2022). The compilation presented here implies that this mechanism likely existed as early as 3.8 Ga, in good agreement with bulk rock analyses of metamorphic Iron Formations from Greenland (Dauphas et al., 2004;Czaja et al., 2013).  (a, c, e, g, i) and the corresponding probability density plots (b, d, f, h, j). The numbers of values (n) and geological formations (f) from which pyrite originates are indicated on the histograms. The vertical grey bar on the right panels represents a δ 56 Fe value of 0 and is drawn to facilitate comparison between plots. Stromatolite distribution includes data measured in pyrite grains hosted in stromatolitic carbonate, stromatolite and stromatolitic chert (see data_compilation.xlsx in supplementary materials). Similarly, the chert distribution includes data measured in pyrite hosted in (black or white) chert, chert with barite, amphibole-rich cherts. The detrital sediment distribution includes data measured in pyrite hosted in (green, grey, black, calcareous or tuffaceous) mudstone, black shale, diamictite, (siliceous, calcareous or tuffaceous) sandstone, quartz-pyrite pebble, carbonaceous shale, carbonate laminated siltstone, quartz-rich clastic sedimentary rock and conglomerate.

Pyrite records Fe kinetic abiotic and microbial processes
Iron isotope composition in pyrite from all ages displays a negative δ 56 Fe mode ranging from -2 to -1 ( Fig. 6). Three main hypotheses could potentially explain such negative δ 56 Fe values in pyrite and are discussed below.
First, Rouxel et al. (2005) proposed that partial oxidation and precipitation of large amounts of seawater Fe(II) as isotopically heavy iron (oxyhydr)oxides would leave a pool of light residual dissolved Fe(II) from which pyrite precursors could precipitate. This model implies that pyrite and iron (oxyhydr)oxide should record parallel distillation trends from heavy to light isotope compositions (Heard and Dauphas, 2020). As described in subsection 4.3, pyrite with positive δ 56 Fe values probably retains the composition of its iron (oxyhydr)oxide precursor. Thus, the shift in opposite directions of the positive and negative δ 56 Fe modes between the 3.2-2.5 and 2.5-2.05 Ga intervals (Fig. 6) does not support this formation mechanism. However, data from the latter interval come from only two different locations. Therefore, distillation of the seawater Fe(II) reservoir as a contributor to pyrite formation mechanism cannot completely be ruled out.
Secondly, non-redox kinetic reactions could account for a large part of the negative pyrite Fe isotope compositions (Guilbaud et al., 2011;Heard and Dauphas, 2020). Variable pyrite precipitation rate can induce large isotopic fractionation range (Guilbaud et al., 2011;Mansor and Fantle, 2019). δ 56 Fe values as low as -3 are obtained when the precipitation rate is faster than the isotope exchange (Mansor and Fantle, 2019). Pyrite precipitation rate depends on Fe/S ratio, whose variation controls the kinetic vs equilibrium extents. In the early Archean ocean, pyrite precipitation was likely slow, due to limited sulfide availability (Mansor and Fantle, 2019). Increased sulfate delivery imparted by the GOE (Lyons et al., 2014) would have increased pyrite precipitation rate and could potentially explain the shift of the negative δ 56 Fe mode to more negative values between 3.2 and 2.05 Ga. However, large variations of S isotope composition were reported in Archean stromatolites, pointing to local enrichments in sulfate (Bontognali et al., 2012;Marin-Carbonne et al., 2018). Following Mansor and Fantle (2019), such conditions should lead to markedly negative pyrite Fe isotope compositions in stromatolites, which are not observed (Fig. 5). For other lithologies, like cherts, carbonates and detrital sediments, the sulfate content is poorly constrained (Halevy, 2013). Therefore, a kinetic control cannot be excluded in these environments. Kinetically precipitated pyrites are expected to show a correlation between particle size and Fe isotope composition (Mansor and Fantle, 2019), which is not observed from the present compilation (Fig. 3). Similar conclusions were drawn by Decraene et al. (2021b) on micropyrite from the Tumbiana Formation. Kinetic precipitation should produce a specific isotope profile within a single pyrite grain. Precipitation rate is predicted to slow down as the reaction proceeds (Mansor and Fantle, 2019), thus δ 56 Fe values should gradually increase from light to heavy isotope composition, from inner to outer parts of the pyrite crystal. Such Fe isotope profiles are not observed in pyrite from Mendon black cherts (Marin-Carbonne et al., 2020). We therefore conclude that some of the pyrite grains may be produced by kinetic precipitation, but this process cannot be considered as the sole driver for pyrite formation. Lastly, isotopically light pyrites could precipitate from partial reduction of iron (oxyhydr)oxides, either abiotically or under the influence of DIR bacteria (Crosby et al., 2007;Johnson et al., 2008;Frierdich et al., 2019). Abiotic reduction is associated with isotope fractionation between -2.6 and -3.1 (Wiederhold et al., 2006;Frierdich et al., 2019), not significantly different from DIR fractionation, -2.9 (Crosby et al., 2007). Accordingly, iron isotope fractionation alone cannot be used to assess pyrite biogenicity. Stromatolites often contain micropyrite closely associated with organic matter, likely corresponding to remains of microbial mats (Bontognali et al., 2012;Marin-Carbonne et al., 2018;Lepot et al., 2019;Decraene et al., 2021b). In our compilation, stromatolites originating from four different locations display a bimodal δ 56 Fe distributions with a difference of -2.5 (Fig. 5), consistent with DIR microbial activity, of pyrite with positive δ 56 Fe record the composition of primary Fe (oxyhydr)oxides. It may also explain the bimodal δ 56 Fe distribution observed in cherts (Fig. 5). Specifically, organic carbon-rich cherts show mostly negative δ 56 Fe distributions (Fig. 7). Although organic carbon in cherts is rarely measured and these results might therefore not be representative, the latter suggest that organic matter availability controls Fe reduction. Several experimental works showed that organic matter might catalyze Fe(III) reduction and pyrite formation, by providing an active surface for pyrite nucleation and therefore increasing the corresponding reaction rate Rickard et al., 2007;Rickard, 2012;Duverger et al., 2021). Consequently, organic matter availability could drive Fe isotope fractionation during pyrite precipitation. The negative Fe isotope compositions measured in carbonates could also have been produced by DIR (Fig. 5), as previously suggested by Yoshiya et al. (2012) for the Tumbiana Formation.
In summary, the negative δ 56 Fe values cannot be explained by a single process. Partial reduction of Fe (oxyhydr)oxides followed by pyrite precipitation, alongside kinetic precipitation from dissolved Fe(II), could both have contributed to pyrite formation during the Archean and Paleoproterozoic.

Temporal evolution of micropyrite δ 56 Fe values
The evolution of bulk δ 56 Fe values in sedimentary pyrite through time has been proposed to reflect Fe/S ratio variations in seawater (Guilbaud et al., 2011;Mansor and Fantle, 2019;Heard and Dauphas, 2020). In particular, the shift from negative to positive δ 56 Fe values from Archean to Paleoproterozoic could reflect an increase in sulfate delivery to seawater, imparted by continental oxidative weathering following the GOE (Canfield, 1998;Lyons et al., 2021). Positive correlation between Archean and Paleoproterozoic pyrite δ 56 Fe and S isotope compositions of sulfide led Heard and Dauphas (2020) to suggest a primary control of the Fe/S ratio in the ocean on Fe isotope signatures in pyrite.
Our compilation of in situ data shows a different pattern (Fig. 1). The quasi-systematic prominence of the positive mode suggests that pyrite was mostly formed via Fe (oxyhydr)oxide reduction, starting 3.6 Ga ago. As oxidative weathering after the GOE likely content. Experimental works suggested that organic matter could catalyze pyrite formation by providing active surfaces for pyrite nucleation (Rickard et al., 2007;Rickard, 2012;Duverger et al., 2021). The upper plot (a) comprises pyrite δ 56 Fe values from cherts containing 0.82 to 1.51 wt% organic carbon. The middle plot (b) includes δ 56 Fe values measured in black cherts for which no specific information about the organic carbon content was given. We suggest that they can reasonably be considered as organic matter-rich host rocks. In the lower plot (c), δ 56 Fe values were measured in pyrites associated with cherts presenting TOC values from less than 0.05 to 0.38 wt%. The numbers of values (n) and geological formations (f) from which pyrite originates are indicated. Pyrite particles in organic matter-poor and organic matter-rich (TOC > 0.8 wt%) cherts have one location in common (Mendon Formation). The three other locations are distinct. The grey bar represents a δ 56 Fe value of 0 and is drawn to facilitate comparison between plots.
increased the input of Fe (oxyhydr)oxide with δ 56 Fe close to 0 , the occurrence of 0 pyrite grains in stromatolites, cherts or BIFs post-dating the GOE would confirm the proposed pyrite formation mechanism from Fe (oxyhydr)oxide quantitative reduction (Section 4.3). However, only few microanalyses were carried out on such samples (only 18 analyses in 2.32 Ga old stromatolites).
Furthermore, δ 56 Fe values close to 0 become more significant between 3.2 and 2.05 Ga. These values were measured in pyrite grains hosted in detrital sediments (Fig. 5), and they are similar to the crustal average δ 56 Fe value . Several studies suggest that the lithosphere might have stabilized between 3.2 and 2.5 Ga, leading to the development and thickening of the continental crust (Dhuime et al., 2015;Cawood et al., 2018). This would in turn have led to the onset of continental weathering, that was evidenced in formations as old as 3.2 Ga (Hessler and Lowe, 2006). Delivery of continental Fe(II)-bearing species through rivers before the GOE (Hao et al., 2017) could have supplied the ocean detrital pyrites delivered to the ocean. Pyrite grains from various formations (3.1 to 2.6 Ga) in Zimbabwe and Kaapvaal cratons were described as detrital by Hofmann et al. (2009), based on petrographic evidence, ages greater than the depositional age, and multiple S isotope analyses. They used Fe and S isotope compositions to constrain the pyrite origin, and suggested that δ 56 Fe values clustering around 0 , associated with small variations in 34 S/ 32 S ratios and S mass independent fractionations close to 0 reflect a crustal origin. Moreover, Agangi et al. (2015) measured Fe isotope compositions in 2.7 Ga pyrite, that they interpreted to be of detrital origin. Inclusion-rich pyrites have a high δ 56 Fe average (2.75 ), compared to concentrically laminated pyrites (0.2 ). A detrital origin of pyrite cannot be confirmed from the present compilation as only Fe isotope compositions are available and no obvious pyrite shape or size can be used as detrital indicator.
Interpretation of δ 56 Fe distributions in the other time intervals is uncertain and not discussed herein, as the small number of pyrite localities may induce a lack of representativity.
In summary, microscale δ 56 Fe distributions in pyrite point to Fe (oxyhydr)oxide precursor as early as 3.6 Ga ago. In addition, an influence of continental weathering might be recorded in Fe isotope signatures from 3.2 Ga. Further investigations of the δ 56 Fe evolution through time would increase the present database and possibly strengthen the tentative conclusions drawn here.

Conclusions
In situ Fe isotope compositions in Archean and Paleoproterozoic pyrite show distinct trends from those obtained by conventional bulk measurements of pyrite δ 56 Fe values. The origin of the discrepancy between in situ and bulk Fe isotope signatures is not clear, and should be addressed in future works, for instance by coupling in situ and bulk techniques on the same samples. The δ 56 Fe distributions in various lithologies point to several pyrite formation mechanisms: total and partial Fe (oxyhydr)oxide reduction, that could be abiotic and/or potentially microbially mediated, and kinetic pyrite precipitation from dissolved Fe(II). These mechanisms might have existed since since 3.8 Ga, as suggested by the consistent evolution of the δ 56 Fe distributions throughout early Archean to Paleoproterozoic. The high occurrence of pyrite with positive δ 56 Fe values indicates that Fe(III) reduction was a dominant pyrite formation pathway. Importantly, this study underlines the lack of microanalyses in specific lithologies and time intervals, that hinders robust interpretation of the isotope signature records.

CRediT authorship contribution statement
JMC, VB and JD designed the study, JMC and MND conducted the SIMS analyses, JD compiled the data. All authors have contributed to the data interpretation. JD wrote the manuscript with important contributions of all co-authors.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
Data are available in the Supplementary materials. benefited from comments from two anonymous reviewers, as well as editorial handling by Boswell Wing.
Juliette Dupeyron obtained her MSc at the Institut de Physique du Globe de Paris (IPGP, France) in 2021. She is currently a PhD candidate at the University of Lausanne (UNIL, Switzerland) and IPGP, where she studies pyrite formation in modern anoxic environments. She is supervised by Johanna Marin Carbonne (UNIL), Jasmine Berg (UNIL), and Vincent Busigny (IPGP).
Marie-Noëlle Decraene graduated from the University of Lorraine (France) in 2018 and recently obtained her PhD in Earth sciences at the University of Lausanne (Switzerland). Her research focuses on the origin of pyrite in ancient microbialites. She is interested in the ability of micrometric pyrite to record and preserve microbial processes, by measuring spatially resolved Fe isotopes in Archean and Phanerozoic sedimentary rocks.
Johanna Marin Carbonne completed her PhD in geosciences at the CRPG-CNRS (University of Lorraine, France) in 2009. She was first a researcher at UCLA (USA, 2010(USA, -2012, then at the Institut de Physique du Globe de Paris (France, 2012(France, -2014. She joined the faculty of the University Jean Monnet (Saint-Etienne, France) and the laboratory Magma and Volcanoes (Clermont Ferrand, France), and moved to the University of Lausanne (UNIL) in 2018. She is currently associate professor in stable isotope geochemistry at UNIL and the director of the SwissSIMS, an ion probe national facility owned by the University of Lausanne, University of Bern, University of Geneva and ETHZ. Her research interests include Early Earth surface condition reconstructions, redox biogeochemical cycles through time, microbialite and stromatolite formation and microbe mineral interactions. Vincent Busigny is a Professor of Geochemistry at the University Paris Cité and the Institut de Physique du Globe de Paris. After completing his PhD in 2004 in Paris under the supervision of Marc Javoy, Pascal Philippot and Pierre Cartigny, he worked with Nicolas Dauphas at the University of Chicago before returning to France. His research emphasizes the use of stable isotopes (Fe, N, C, H, O, S) as tracers of mineral-fluid-life interactions. He currently focuses on the reconstruction of Precambrian environments and ecosystems based on three different approaches: study of ancient sedimentary rocks, laboratory experiments, and modern analogues of ancient environments. Other applications include the behavior of fluid-mobile elements during alteration and subduction of oceanic lithosphere to determine chemical evolution of terrestrial surface and mantle reservoirs.