Sedimentary provenance, mild hydrothermal alteration and metasomatism have affected the major and minor element distribution in these rocks, and therefore need to be addressed. Regarding sedimentary provenance, mapping has shown that the turbidites analyzed in this study sit on top of mafic volcanic rocks (Rosing et al., 1996), and our Th/Sc, U/Th and Fe/Al ratios are indeed consistent with a strong contribution of mafic detritus. These element ratios were selected because they have previously been shown to be good discriminators for distinguishing between mafic, ultramafic and felsic provenance (Ptáček et al., 2020). All three ratios fall in between those of average upper continental crust and average oceanic crust (Rudnick and Gao, 2014; White and Klein, 2014), which suggest mixing of material from sources of similar compositions. The relatively high Ni/Co ratios may further indicate contributions of ultramafic material, such as komatiite (Figure 5A), consistent with previous studies of other Archean siliciclastic rocks (Ptáček et al., 2020). This interpretation is overall in line with previous trace element work on metasedimentary rocks from the Isua Supracrustal Belt (Kamber et al., 2005).

In the case of U/Th and Fe/Al, input of mafic or ultramafic detritus has important implications for the utility of these elements as redox proxies. We briefly address this topic here, because previous work found geochemical evidence of biological oxygen production in rocks from this same geological unit (Rosing and Frei, 2004). A high Fe/Al ratio above ∼0.5 in sedimentary rocks, i.e., elevated relative to upper continental crust, is typically interpreted as evidence of anoxic conditions during the time of deposition that favored the accumulation of authigenic iron minerals, including sulfides, carbonates or oxides (Lyons and Severmann, 2006; Raiswell et al., 2019). However, in settings with significant contributions of iron-rich siliciclastic material or detrital iron sulfides or iron oxides, such as mafic rock-forming minerals and their weathering products, this empirically defined threshold is no longer applicable (Stüeken et al., 2017; Stüeken et al., 2020). It is therefore not possible to infer redox conditions during the time of deposition on the basis on Fe/Al ratios. Other commonly used redox proxies include Mo abundances and U/Th ratios (Tribovillard et al., 2006). Both Mo(VI) and U(VI) are soluble under oxic conditions and can thus become enriched in anoxic sediments as Mo(IV) and U(IV), respectively. However, we find no evidence for enrichments in either of these proxies above the level of detrital background. Molybdenum would be expected to correlate with TOC, because adsorption of thiolated Mo to organic matter is the major pathway for Mo burial in sediments (Wilde et al., 2004; Helz et al., 2011). The absence of such a correlation in our data (Figure 4C) prohibits any inferences about Mo levels in the water column during the time of organic matter deposition. We can therefore not confirm previous suggestions for the presence of biogenic O₂ production during the time of deposition of these meta-sedimentary rocks (cf. Rosing and Frei, 2004).

It is, however, likely that these rocks have undergone some degree of hydrothermal alteration. Previous workers documented chalcopyrite in this geological unit (Rosing, 1999), and our good correlation between Cu and S (Figure 4B) is likely evidence for the presence of chalcopyrite. Copper [both Cu(I) and Cu(II)] is most soluble in saline and/or hot fluids, such as hydrothermal effluents and relatively insoluble in cold seawater (Zhong et al., 2015). The Cu and S abundances in our samples are overall relatively low, and it is conceivable that Cu sulfides are detrital in origin. However, given the association of Cu with S, we cannot rule out minor hydrothermal overprinting. This conclusion is also consistent with our sulfur isotope data which plot close to the average upper mantle value of −1 ± 0.5 ‰ (Labidi et al., 2012) and may thus reflect contributions of hydrothermal H₂S derived from magmatic processes (Seal, 2006). Such hydrothermal fluids may have also introduced slightly elevated amounts of Mo into some of the samples, which also covaries with S (Figure 4D). Whether this hydrothermal activity was syn- or post-depositional cannot be resolved from our data, but its implications require consideration in the interpretation of the nitrogen isotope data.

In the form of ammonium (NH₄⁺), nitrogen has the same charge and size as K⁺ and Rb⁺ and therefore partitions into similar mineral phases (Busigny and Bebout, 2013). This notion explains our dataset where the abundances for TN and K show a strong correlation (R² = 0.89; Figure 3C). It is thus conceivable that one major source of N are detrital mineral grains that were eroded from K-bearing igneous rocks (e.g., Hall, 1999). However, organic δ¹³C values as well as previous hydrogen and nitrogen detections in the graphite in these rocks point toward the presence of biomass during the deposition of the turbidites (Rosing, 1999; Hassenkam et al., 2017). This graphitized biomass is thought to be derived from living organisms that were thriving on the seafloor or within the water column. When biomass gets buried in sediments and undergoes diagenesis, ammonium is released into porewaters, where it may accumulate to high concentrations in the mM range (e.g., Rosenfeld, 1979; Boudreau and Canfield, 1988). At such high concentrations, significant amounts of ammonium can substitute into K-bearing minerals, such as illite (Müller, 1977; Schroeder and McLain, 1998). This mechanism thus effectively transfers N from organic matter into silicate minerals and has even been invoked to explain elevated N abundances in granitoids (Hall, 1999). Hence if the graphite in our samples does indeed represent ancient biomass, then it is very likely that N was initially introduced in the form of organic molecules and later transferred to potassic minerals during diagenesis and/or metamorphism. The correlation between TN and K could therefore represent a diagenetic artifact rather than provenance.

However, we cannot rule out that the N-poor samples in our set contain significant contributions of detrital nitrogen. In fact, mixing between a detrital and authigenic component could well explain the correlation between TN and δ¹⁵N that is difficult to explain by metamorphic effects alone. We can calculate this detrital endmember, if we make an assumption for its isotopic composition. As a first plausible guess, we assume that it has a composition of −5‰, i.e., similar to Earth’s upper mantle and mafic crust (Cartigny and Marty, 2013). Two-endmember mixing of (i) detrital nitrogen with a composition of −5‰ and (ii) authigenic nitrogen with a composition of +6.1‰ (i.e., the mean value of the N-rich samples) would thus imply that the authigenic component of the three N-poor samples makes up between 61 and 72% while the rest is detrital; the detrital fraction would make up 1.6 ppm on average (range 1.4–1.7 ppm). It is likely that also the N-rich samples contain a similar detrital component, but if we correct the measured data under the assumption that 1.6 ppm of the total nitrogen is detrital in origin, the resulting δ¹⁵N values increase only slightly from a mean of +6.1 to +6.9‰. Across all samples, the authigenic component would thus have a composition of +6.7 ± 0.8‰ (Figure 3A). Hence detrital contributions are probably negligible for these relatively N-rich samples in our sample set (4–8% of total N).

Importantly, using a value of −5‰ for the detrital contribution, based on the composition of mantle-derived magmatic rocks (Cartigny and Marty, 2013), requires that this detrital component has not been affected by metamorphism in the same manner as the authigenic nitrogen in the same samples (see section “Metamorphic Effects on Nitrogen Geochemistry”). This assumption is plausible, because detrital minerals eroded from magmatic rocks have already been exposed to high temperatures and pressures and may therefore be resistant to metamorphism. Of course, magmatic rocks do undergo metamorphism as well, but magmatic N is likely fully lattice bound as the high temperatures under which the minerals formed would volatilize loosely bound material. Therefore, metamorphism of magmatic materials likely results in a smaller loss of N compared to materials in which N is predominantly organic-bound. Hence detrital magmatic minerals are expected to lose relatively less N during thermal metamorphism. However, if we take the more radical view that all three of the N-poor samples are 100% composed of detrital N with an average of 4.8 ppm and an isotopic value of +2.3‰ (i.e., allowing for metamorphic alteration and isotopic enrichment of detrital minerals), it would mean that the N-rich population contains between 76 and 87% authigenic N, and this authigenic N would have an isotopic value of +7.0 ± 0.8‰. This result is very similar to the value of +6.7 ± 0.8‰ calculated above, meaning that the uncertainty about the size and isotopic composition of the detrital component does not impact our overall conclusions. We will therefore proceed with the assumption that the authigenic N component in these samples falls around a mean δ¹⁵N value of 6–7‰.

The rocks investigated in this study have undergone metamorphic alteration up to mid-amphibolite facies (Rosing, 1999; Ramírez-Salazar et al., 2021), and it is well known that metamorphism strongly impacts N abundances and isotopic ratios in sedimentary rocks (Haendel et al., 1986; Bebout and Fogel, 1992; Boyd and Phillippot, 1998; Jia, 2006; Palya et al., 2011). The measured data do therefore not represent primary values. However, it is possible to quantify the degree of metamorphic alteration and to estimate the pre-metamorphic starting value, because the isotopic fractionation factors associated with partial N volatilization during metamorphism have been constrained in previous studies. For a temperature around 527°C, Hanschmann (1981) estimated α = 1.0073 for N-loss as NH₃ (i.e., for the reaction NH₃–NH₄⁺) and α = 1.0049 for N-loss as N₂ (for the reaction N₂–NH₃), where α = (¹⁵N/¹⁴N)_residue/(¹⁵N/¹⁴N)_volatile (summarized by Haendel et al., 1986). These fractionation factors were found to be applicable to Archean rocks by Pinti et al. (2001). As the nitrogen speciation in our scenario is not known, we plot the maximum and minimum fractionation factors (Figure 6). We used a standard Rayleigh distillation equation (δ¹⁵N_metamorphic/1000 + 1)/(δ¹⁵N_initial/1000 + 1) = F^{(1–α )}) where F = fraction nitrogen remaining, δ¹⁵N_initial = initial δ¹⁵N value, and δ¹⁵N_metamorphic = metamorphic δ¹⁵N value. Either F or δ¹⁵N_initial can be calculated by assuming reasonable bounds for the respective other parameter.

At first, we fixed δ¹⁵N_initial and calculated F. A plausible assumption for δ¹⁵N_initial may be the composition of biological N₂-fixing organisms, which dominate δ¹⁵N signals of younger Archean sedimentary rocks (Stüeken et al., 2016). We therefore set δ¹⁵N_initial to 0‰ and calculated how much nitrogen would need to remain in the system (F) such that the calculated metamorphosed values (δ¹⁵N_metamorphic) match our observation (+6.7‰ for average authigenic N). The results suggest that about 24–40% (average 33%) of TN would have needed to be retained to shift authigenic δ¹⁵N from 0 to +6.7‰ with a fractionation factor (α) of 1.0049–1.0073 (Figure 6A). However, such a high retention is inconsistent with previous studies that documented down to only 10% nitrogen retention at amphibolite facies (Haendel et al., 1986). The average TN retention found by Haendel et al. (1986) was 20%, while the maximum was around 10%. In a second calculation we therefore assumed that only 10% of TN has remained in the sample (i.e., F = 0.1) and varied δ¹⁵N_initial until the isotopic composition of the residual measured nitrogen (δ¹⁵N_metamorphic) was equal to the calculated authigenic δ¹⁵N value for each sample. The results point toward a starting composition of −4.6 ± 0.8‰ for α = 1.0049 and −10.1 ± 0.8‰ for α = 1.0073 with a mean of −7.3 ± 1.1‰ (Figure 6B). This value is close to the value of −7.1‰ that Pinti et al. (2001) reconstructed from a single sample of the same geological unit, using a stepwise combustion technique paired with argon isotope measurements. If we assume 20% N retention, our initial δ¹⁵N values would fall between −1.2 and −5.2‰, which does not overlap with the value derived by Pinti et al. (2001). To further constrain TN retention, we looked at the reconstructed C/N ratios, which were calculated by changing TN while holding TOC constant. In reality, some TOC is likely to have been lost as well during metamorphism, and so the reconstructed values represent an upper limit of the true initial C/N Redfield ratio. In the case of 10% retention, reconstructed C/N ratios fall around an average of 15 for the N-rich sample population, which is close to the Redfield ratio (C/N = 7–10) of microbial biomass (Godfrey and Glass, 2011; Algeo et al., 2014). For the case of 20% retention, the C/N ratios would be 30 on average and therefore further removed from the expected value of microbial biomass. This offset would be even larger if we account for loss of some TOC during metamorphism. A scenario with only 10% nitrogen retention, i.e., 90% TN loss during metamorphism, therefore appears more likely. In any case, the reconstructed C/N ratios vary widely between samples (range 2–226 for the 10% retention scenario, median 20, Figures 6C,D), but this variability may simply reflect diagenetic ammonium migration from organic-rich to organic-lean laminae, which is commonly observed in younger sedimentary successions (e.g., Koehler et al., 2019b). Overall, paired with the one data point collected by Pinti et al. (2001), our data suggest a starting δ¹⁵N value that was significantly less ¹⁵N-enriched than in younger Archean sedimentary rocks.

A pre-metamorphic δ¹⁵N value around −1 to −10 ‰ in these Isua turbidites (Pinti et al., 2001, this study) may not necessarily represent the composition of biomass during the time of deposition. As noted above, we cannot rule out that these rocks have been exposed to at least mild hydrothermal alteration over their complex metamorphic history, as suggested by the presence of possibly hydrothermally-derived chalcopyrite. However, it is unlikely that post-depositional fluid circulation led to these light isotopic values by either subtraction or addition of nitrogen. First, experimental work has shown that the interaction between hot fluids and organic-bound N imparts minimal isotopic fractionation (<1‰) (Boudou et al., 2008). Hydrothermal leaching of N from these rocks is therefore unlikely to have induced such a large isotopic perturbation. Second, hydrothermal fluids are N-poor unless they circulate through sedimentary packages and mobilized ammonium from older organic matter (Lilley et al., 1993), for which there is no evidence in this field area. And even if older sedimentary rocks existed and provided a source of nitrogen to hydrothermal fluids, which was then added to our sample set, those older rocks would themselves have needed to contain isotopically light nitrogen. Hence there is no obvious mechanism by which hydrothermal fluids could have shifted initial δ¹⁵N values downward by several permil.

Other studies of hydrothermally altered Precambrian rocks show strong isotopic perturbations, where organic-bound nitrogen has become isotopically depleted by up to 15‰ (Godfrey et al., 2013); however, in that case, the silicate-bound nitrogen fraction incorporated the complementary heavy nitrogen pool, such that the bulk rock value was within 2–3‰ of contemporaneous unaltered strata from the same basin. As we measured bulk rock values rather than kerogen isolates, this mechanism can therefore not explain our data. Therefore, we find no evidence for significant hydrothermal alteration of the ammonium contained in these samples, suggesting that the pre-metamorphic δ¹⁵N value was between −1 and −10‰.

If primary δ¹⁵N values were as high as −1‰, as inferred for 20% TN retention and a relatively small isotopic fractionation factor (section “Metamorphic Effects on Nitrogen Geochemistry”), this may be evidence for the presence of biological N₂ fixation as far back as 3.7 Ga. This interpretation would be consistent with phylogenetic data indicating an early evolution and ecological radiation of this metabolism, possibly dating back to the last universal common ancestor of life on Earth (Weiss et al., 2016; Parsons et al., 2020). However, if the lighter values are correct, which agree better with the previous estimate by Pinti et al. (2001) and are consistent with reconstructed C/N ratios, this may point toward a distinct geobiological N cycle at the time. First, such a low δ¹⁵N value could reflect a different source of N to biological communities compared to later periods in Earth’s history where δ¹⁵N rarely drops below −2‰ (Stüeken et al., 2016). One possible explanation may be biological N₂ fixation with a so-called alternative nitrogenase. These enzymes contain either V or Fe instead of Mo in their catalytic center and impart isotopic fractionation down to −8 ‰ during the conversion of N₂ to NH₄⁺ (Zhang et al., 2014). They are rarely expressed in natural environments today, but Mo scarcity in the early Archean ocean may potentially have favored V- or Fe-based nitrogenase (Raymond et al., 2004; Scott et al., 2008). However, the scarcity of similarly light δ¹⁵N values throughout the rest of the Archean then becomes puzzling. Furthermore, phylogenetic data suggest that V- and Fe-based nitrogenases may not have radiated until the late Proterozoic (Parsons et al., 2020), making it unlikely that they were important players in the Eoarchean N cycle. Alternatively, isotopically light N may have been derived from atmospheric rainout of lightning products or HCN. HCN can form during photolysis in the upper atmosphere (Tian et al., 2011), and experimental data revealed an isotopic composition of −15 to −25‰ (Kuga et al., 2014). The composition of lightning products such as NOx is so far poorly constrained, but some existing measurements from the modern atmosphere show light values around −5 to −15‰ (Moore, 1977). Atmospheric N-bearing molecules could thus be a plausible explanation for our results. Phylogenetic data are consistent with early utilization of NOx species, possibly derived from lightning (Parsons et al., 2020). If so, it would imply that the atmospheric nitrogen source declined in relative importance between the Eoarchean and younger Archean (3.2 Ga onward) where such light values are no longer observed and N metabolisms are dominated by anaerobic pathways (Parsons et al., 2020). A third possible explanation for light δ¹⁵N in biomass is partial assimilation of ammonium from a large, dissolved ammonium pool in seawater. Culturing experiments with modern microorganisms show fractionation factors of −4 to −27‰ if ammonium concentrations exceed ca. 10–20 μM (Hoch et al., 1992). However, given the high metabolic costs of converting N₂ into ammonium and the inefficiency of abiotic ammonium accumulation in seawater (Stüeken, 2016), it is unlikely that such a large ammonium pool existed. In summary, the meaning of these light values in Eoarchean rocks, if correct, remains elusive, but atmospheric rainout of isotopically light bioavailable N species is at present perhaps the most plausible explanation.

If our reconstruction is correct, and if these values are representative of the Eoarchean (which we cannot confirm from a single outcrop), then this may have implications for the secular evolution of key geological N reservoirs over Earth’s history. Today Earth’s lower mantle, upper mantle and surface reservoirs show striking and intriguing isotopic imbalances. The upper mantle shows an average δ¹⁵N value of ca. −5 ± 3‰, as determined by measurements of diamonds, kimberlite xenoliths and mid-ocean ridge basalt (Cartigny and Marty, 2013). This negative upper mantle value is thought to date back to the Archean as indicated by diamonds from 3.3 to 2.9 Ga which display a mode around −5 ± 3‰ (Richardson et al., 2001; Cartigny, 2005). In contrast, plume sources for ocean island basalts—assumed to originate from lower mantle domains below the upper mantle—are weakly positive in δ¹⁵N, similar to post-Archean crustal reservoirs (Marty and Dauphas, 2003). Since the Great Oxidation Event in the Paleoproterozoic (∼2.5 Ga), crustal materials are dominantly enriched in ¹⁵N (>+2‰) (Zerkle et al., 2017; Kipp et al., 2018), and most of the Archean sedimentary record between 3.2 and 2.5 Ga falls around values of 0‰ (Stüeken et al., 2015; Homann et al., 2018; Koehler et al., 2019a; Ossa Ossa et al., 2019). Isotopically lighter values are rare in open marine settings and mostly restricted to organic extracts, which are distinct from bulk rock values (Beaumont and Robert, 1999; Yang et al., 2019). Hence Earth’s upper mantle and exterior (crust + atmosphere) appear to have displayed an isotopic imbalance for the last 3.2 billion years (Boyd and Pillinger, 1994). This is intriguing, because degassing of the upper mantle and/or the subduction of sedimentary nitrogen should generate a more ¹⁵N-enritched upper mantle (Boyd and Pillinger, 1994).

However, if the positive δ¹⁵N values in plumes from the lower mantle are primary and reflect the initial mantle δ¹⁵N value, then one possible explanation for the isotopic imbalance of the upper mantle is the subduction of isotopically light sedimentary rocks in the Eoarchean This hypothesis is not new (Marty and Dauphas, 2003; Cartigny, 2005; Cartigny and Marty, 2013), but has perhaps lost momentum because several recent studies of Mesoarchean rocks revealed sedimentary δ¹⁵N values 0‰ (Stüeken et al., 2015; Homann et al., 2018; Koehler et al., 2019a; Ossa Ossa et al., 2019). However, our new findings open up the possibility that the Eoarchean N cycle may have been distinct, in particular if it involved a stronger involvement of atmospheric products that were incorporated into ancient biomass. It is therefore conceivable that during the first one billion years of Earth’s history nitrogen in sediments was more akin to the modern upper mantle δ¹⁵N value. If correct, this may fuel the idea that the mantle inherited its light value from Eoarchean sediment recycling. We stress that a geodynamic mechanism for such a model is so far ambiguous, and it remains to be determined if the total volume of isotopically light sedimentary rocks would have been large enough to change the isotopic value of the mantle. For example, the subduction of sedimentary N with negative δ¹⁵N values would only control the upper mantle δ¹⁵N value if the mass of subducted sedimentary N is far greater than the N abundance of the upper mantle to account for loss by degassing. The abundance and distribution of nitrogen throughout the deep silicate Earth is a topic of much debate (Zerkle and Mikhail, 2017), but our results suggest that this may be a worthwhile avenue to pursue in future studies.