Biological use of molybdenum and tungsten stems back to 3.4 billion years ago Authors: Aya S Klos, Morgan S Sobol, Joanne S Boden, Eva E Stüeken, Rika E Anderson, Kurt O Konhauser, Betül Kaçar Published: 2026 ABSTRACT AbstractThe biological significance of the transition metal molybdenum (Mo) lies in its function at the catalytic center of several enzymes that drive a wide spectrum of redox reactions underlying global biogeochemical cycles, yet a paradox persists. While modern life ubiquitously relies on Mo, geochemical evidence suggests that its availability in early Earth’s anoxic oceans was extremely limited. Modern organisms can use Mo down to trace levels; however, the rates of Mo-dependent metabolisms slow down when Mo availability decreases, posing fundamental questions about the extent to which changing Mo abundances shaped the evolution of molybdoenzymes, and when early life began harnessing Mo. Here, we confront this evolutionary enigma by reconstructing the temporal and ecological emergence of molybdoenzymes, their transport systems, and biosynthetic pathways. In parallel, we examine biological tungsten (W) usage due to shared chemical properties and cofactor biosynthetic pathways with Mo. We provide molecular dating evidence of Mo/W utilization back to the Eo- to Mesoarchean (~3.7–3.1 Ga). These findings challenge prevailing assumptions about trace metal availability on the early Earth and underscore the profound antiquity and adaptability of Mo-based biochemistry in shaping early microbial evolution.Subject terms: Element cycles, Phylogenetics Introduction The transition metal molybdenum (Mo) displays an enigmatic evolutionary history in biology in relation to the environment1,2. Mo participates in key biogeochemical transformations of carbon (C), nitrogen (N), and sulfur (S), and previous studies have proposed that many of these metabolic pathways have ancient histories1,3. The catalytic functions of Mo are mediated through its incorporation into enzymes via one of three cofactors, which are assembled by complex biosynthetic pathways that would have needed to have existed to enable Mo utilization4–6. These cofactors include the widely utilized molybdopterin cofactor (Moco), the binuclear [CuSMo] cluster, and the [Mo:7Fe:9S:C] metallocluster (FeMoco). Moco-using enzymes encompass the majority of molybdoenzymes and are further divided into families distinguished by structural and biochemical differences, described in more detail below. FeMoco is uniquely associated with Mo-dependent nitrogenase, the enzyme that catalyzes the reduction of nitrogen gas (N2) to ammonia (NH3) during a process known as N2-fixation, a key step in the global N cycle because it constitutes the major source of naturally fixed N to the biosphere. Phylogenetic analyses suggest that nitrogenases date back to the mid-Archean over 3 billion years ago7,8, consistent with geochemical data9,10. The environmental availability of Mo is thought to have fluctuated markedly throughout Earth’s history in response to progressive oxygenation of seawater11,12. Prior to the origin of oxygenic photosynthesis and subsequent accumulation of O2 in the atmosphere around 2.45 billion years ago13 (a period termed the ‘Great Oxidation Event’ (GOE)) dissolved Mo is thought to have been exceedingly low (<5 nM compared to 105 nM in modern seawater14). This scarcity is thought to have stemmed from the suppression of oxidative weathering of Mo-bearing sulfide minerals on land, which today represent the primary source of Mo to the oceans15. Studies show that organisms can uptake16 and use Mo at trace levels, and a growing body of evidence suggests that Mo was used by life in early Earth environments4–6. Nevertheless, Mo limitation exerts constraints on a system’s capacity for essential Mo-dependent metabolisms17,18 and can consequently restrict primary productivity under certain conditions19,20. Additional data are needed to reconcile the relationships between Mo availability, microbial metabolisms, and environmental redox conditions on the early Earth to assess the importance of Mo-limitation for the growth and evolution of the biosphere during the Archean. In this study, we address this conundrum via a comprehensive computational examination of Mo uptake, storage, intracellular trafficking, cofactor biosynthesis, and catalysis. In parallel, we examine the history of biological usage of tungsten (W) in transport and catalysis due to shared chemical properties between Mo and W21, similarities in the biosynthesis of their respective cofactors22, and reports of W replacing Mo in certain enzymes23. First, we use phylogenomic tools to map the distribution of Mo/W usage across modern organisms and their ecological context. Second, we reconstruct the evolutionary history of Mo-/W-utilizing proteins across the modern tree of life. Our aim is to identify the earliest biological processes that used Mo, gauge the impact of environmental change on biological Mo usage, and outline key sequential stages in the evolution of Mo-based enzymes alongside W-based enzymes over time. Results Mo proteins are widely distributed across life Enzymes that incorporate Mo or W complexed by pterin molecules to form the active site cofactor (Moco or Wco, respectively) can be grouped into four main families: dimethyl sulfoxide reductases (DMSOR), xanthine oxidases (XO), sulfite oxidases (SO), and aldehyde:ferredoxin oxidoreductases (AOR). These families are distinguished by structural variations based on specific modifications to the metal atom ligand sphere or pterin backbone during cofactor assembly (Fig. 1). Within the XO family of enzymes, aerobic carbon monoxide dehydrogenases (CODH) contain a binuclear [CuSMo] cluster distinct from the mononuclear clusters of other Moco-using enzymes24. These enzymes are grouped with other XO family members due to structural similarities in the Mo coordination environment. Full schematic of biological Mo-/W-uptake, trafficking and incorporation into selected molybdo-/tungsto-enzymes. Bacterial homologs that catalyze steps in the Moco synthesis pathway are shown in (a) with corresponding homologs in humans, plants, and fungi (Aspergillus nidulans and Neurospora crassa) detailed in (b). Chemical structures for the FeMo cofactor (FeMoco) and the unmodified form of the Mo cofactor (Moco or Mo-MPT for molybdopterin) incorporated by the SO family are also shown. The AOR, DMSOR, and XO families of enzymes require additional modification of Moco (or Wco for W-bound pterin) prior to enzyme incorporation. Chemical structures for these modified cofactors are depicted in (c) where asterisks refer to previous observations of the presence of either Mo or W in cofactors, as has been shown for certain enzymes of the DMSOR and AOR families. Unlike other Moco-using enzymes that contain a single metal center (mononuclear), aerobic carbon monoxide dehydrogenases (CODH) incorporate a binuclear [CuSMo] cluster. These enzymes are grouped into the XO family based on similarities in their coordination of the Mo atom. Wco-incorporating tungstoenzymes are specifically found in the DMSOR and AOR families. While some enzymes use W obligately, the similarity in chemical properties between Mo and W can also allow for some degree of interchangeability between these cofactors. This is exemplified by some DMSOR enzymes known to incorporate Wco instead of Moco25. The four major enzyme families described above can further be divided into subfamilies based on further structural and sequence-related differences. In prokaryotes, the incorporation of Mo into molybdoenzymes begins with its uptake from the environment, commonly in the form of molybdate (MoO42−) (though growth on mineral-derived Mo sources such as tetrathiomolybdate or Mo released from molybdenite26 has been demonstrated as well in certain strains). Uptake of this anion by bacteria is most often catalyzed by the high-affinity ABC-type transport system ModABC. In archaea, a dual-affinity ABC-type transport system that is capable of transporting both Mo and W (WtpABC) is reportedly the most common molybdate transporter27. Additionally, TupABC, a transporter specifically used for tungstate (WO42-) uptake, is found in both bacteria and archaea28. In eukaryotes, the mechanism for molybdate uptake is only broadly characterized. Two different types of molybdate transporters have been identified: MOT1 in land plants and green algae16,29, and MOT2 in green algae and animals, including humans30. In addition, an MOT1 family homolog (Nc-MOT-1) identified in the filamentous fungus Neurospora crassa has been characterized for its role in Mo uptake31. Following uptake, MoO42- can undergo a complicated series of maturation steps orchestrated by proteins encoded by the nif genes to form FeMoco for nitrogenase. Alternatively, it can follow a separate complex biosynthetic pathway and ultimately become incorporated into molybdopterin to form Moco for all other molybdoenzymes. The final Mo-incorporating cofactor is subsequently inserted into the active site of its respective enzymes. For enzymes in the DMSOR, XO, and AOR families, an additional maturation step is required before Moco/Wco incorporation is complete. The core genes involved in Moco biosynthesis pathways are conserved across all domains of life32. We screened select genomes from a previously constructed trees of life33 to identify genes encompassing the core machinery for Mo(/W) transport, storage, and Mo(/W)co biosynthesis (see methods for protein selection). We also included genes representing all four molybdo/tungstoenzyme families, as well as those involved in nitrogenase and FeMoco assembly, resulting in a total of 102 protein groups. Expanding on previous studies34, we confirm the wide distribution of genes encoding molybdo/tungstoenzymes in addition to genes for Mo/W transport, storage, and cofactor biosynthesis across the tree of life. These genes are found among taxa encompassing a broad range of ecological conditions, including high and low temperatures, varying degrees of O2 tolerance, acidic and alkaline pHs, variable habitat types, and habitat diversities (Fig. 2). We find that aerobes (organisms that live in the presence of O2) tend to encode more Mo genes compared to anaerobes (organisms that live without O2) (P < 2.2 × 10−16, r = 0.53). In contrast, anaerobes tend to encode more W genes compared to aerobes, though this difference is smaller than that observed for Mo genes (P < 2.4 × 10−5, r = 0.22, Supplementary Fig. 2). Other environmental parameters (temperature, pH, habitat diversity) display weak correlations with Mo/W genes (R2 < 0.1 in all cases, see Supplementary Fig. 2), and we reiterate the importance of understanding that these correlations describe general patterns rather than ecological rules. Broadly, we find that organisms with numerous different Mo genes tend to be associated with more temperate, pH-neutral conditions, and a greater variety of host environments, whereas organisms with numerous W genes tend to be associated with hotter conditions and fewer different host environments (Supplementary Fig. 2). This is concurrent with previous literature establishing the importance and predominance of tungstoenzymes in hyperthermophilic archaea35, as well as biochemical characterization of tungstoenzymes supporting their suitability to catalyzing low potential redox reactions under high temperature, anaerobic conditions25. Phylogenetic distribution of Mo-related genes and their associations with host ecological attributes. Here, we describe Mo-related genes as those involved in all aspects of biological Mo-usage, including transport, trafficking, storage, cofactor synthesis, and Mo-incorporating enzymes. Mo-related gene counts (b) and host ecological attributes (c) are mapped to a species tree of life33 (a). Tips on the phylogenetic tree are colored by taxonomy (see Supplementary Fig. 1 for color legend). Simplified taxonomic clade labels are on the right. The legend (d) shows gradients for quantitative variables and categories for ecological traits. Missing data is denoted in gray shading for the ecological trait mapping. The most widely distributed genes across the analyzed genomes are those involved in Moco biosynthesis, which is essential for the active sites of all molybdoenzymes except nitrogenase. This finding, alongside previous evidence of the pathway’s conservation32,34, underscores the fundamental importance of the Moco biosynthesis pathway for Mo- and W-utilizing organisms (W-cofactors rely on the same pathway for synthesis22). The next most widely distributed genes encode active membrane transporters for Mo (ModABC) and W (TupABC). The latter is found across nearly all major archaeal lineages but in only select bacterial lineages (Fig. 2). This distribution aligns with previous reports suggesting that Tup transporters are more common in Mo-utilizing archaea compared to bacteria27. Among Moco-dependent enzymes, the most widely distributed ones among prokaryotes generally belong to DMSOR. Conversely, the rarest Mo-related genes include certain members of the SO and XO family, W-dependent AOR family, as well as nitrogenase, and Mo-storage genes (MosAB), though the discovery and inclusion of new enzyme groups not well-characterized at present may shift these patterns. Diversity of Mo/W-related protein families We performed a comprehensive assessment of protein similarity among Mo/W-related genes and their homologs, testing four different log E-value cut-offs to filter the results of homology searches and test the sensitivity of our results to different thresholds (Supplementary Figs. 4–7). The diverse functions of Mo/W-related proteins are mirrored by their diverse ancestries and origins. We find limited sequence homology between most major categories of Mo-related protein families. In other words, they occupy distinct clusters within the sequence similarity network. Notably, our analyses demonstrate limited homology between the major Moco/Wco-dependent enzyme families—DMSOR, XO, SO, and AOR—with strong homology of enzymes detected within these families. These data suggest that either a shared ancestry is no longer detectable due to the large degree of evolutionary distance or that Moco/Wco-incorporating enzymes evolved independently multiple times throughout life’s history as has been previously proposed2. Even more broadly, the ability of any protein to interact with Mo for the purposes of transport, biosynthesis, or catalysis, may have evolved more than once. Ultimately, this complex history gave rise to distinct enzyme families covering a wide range of substrates and redox potentials. This diversity may reflect the suitability and versatility of Mo for incorporation into a variety of different enzymes. Within each functional category (e.g., transport, cofactor synthesis), individual Mo/W proteins form distinct clusters, including the Moco/Wco-dependent enzyme families (XO, DMSOR, SO, AOR), nitrogenase proteins, and different subunits of Mo/W transporters. Distinct clusters also emerge among certain Moco synthesis proteins, reflecting shared catalytic roles – such as transferase activity – and evolutionary histories. Many Mo/W-related proteins have identifiable homologs that are not Mo/W-related, offering a window into their evolutionary origins. For example, the molybdate binding A subunit of the Mo transporter (ModA) is a member of a broader clade of substrate-binding proteins that includes those specific to compounds like SO42− and phosphate (e.g., H2PO4-)36. This supports the hypothesis that this subunit shares an evolutionary origin with non-Mo related transporters. Estimated timing for gene events of Mo/W proteins Our analyses have characterized the modern distribution and occurrence patterns of Mo/W-related enzymes, outlining their potential evolutionary history. To contextualize these analyses within geological time, we performed gene tree-species tree reconciliations to estimate the timing of the rise and proliferation of Mo-/W-related proteins. Chronograms from a previous study37 dating genes involved in phosphorus cycling were used for this analysis. Briefly, these chronograms were reconstructed by anchoring a species tree in time using eight fossil calibration points and one of three clock models, namely Cox-Ingersoll-Ross (cir), lognormal (ln) or uncorrelated gamma multipliers (ugam), resulting in three different age-calibrated trees (see methods for details). Below, we report date ranges that consider results from all three clock models tested. Caution is warranted when interpreting estimates from gene-tree-species-tree reconciliations. For example, a gene event can occur anywhere along a branch of the species tree, resulting in a wide range of possible ages. Here, we report earliest dates for full branch lengths (Supplementary Table 1), however, a summary of midpoint dates (calculated as the mean of dates corresponding to the left and right nodes of a branch) for HGT and duplication events can be found in Supplementary Figs. 8-15. Dates corresponding to loss events were excluded as these cannot be interpreted as evidence of gene usage. Speciation events occur when a lineage containing a gene diverges into two species, and the gene is vertically inherited from the parent lineage to the two daughters. Thus, speciation events are reported with dates corresponding to the node they occurred on, consistent with previous studies37. We find that older events tend to be more temporally constrained because most older branches in the species tree are shorter than more recent branches, as previously reported38. Plots depicting the full temporal range of gene events (spanning the range between left and right node dates) are available in Figs. S17–S22. Our results suggest that the utilization of Mo/W in enzymes extends back to the Eo- to Mesoarchean (Fig. 3). Estimates of the earliest gene events associated with individual genes of Mo/W enzymes date back to ~3.7–3.3 Ga depending on the clock model with the LN model predicting the earliest estimates and the UGAM model predicting the most recent estimates (Supplementary Figs. 8–15). Enzymes associated with the earliest gene events include those involved in Moco biosynthesis, for which all core enzymes in the modern pathway likely existed by the Mesoarchean to Paleoproterozoic (~3.1–2.2 Ga). Similar evidence suggests that all subunits of the modern functional W transport system (TupABC) have existed since ~3.1–2.0 Ga, while all subunits of modern Mo transport systems (ModABC and/or WtpABC) are estimated to have emerged around the Mesoarchean to early Proterozoic (~2.8–2.0 Ga and ~3.0–1.9 Ga, respectively). Evolutionary synthesis model for “building” of biological Mo/W-utilization across Earth history. Shapes at the start of bars correspond to the midpoint date (or right node date for speciation events) of the earliest gene event recorded for genes within a given group. Events occurring on terminal branches were not considered here. For six groups, faded shapes at the start of bars distinguish the earliest gene event for a single gene within the group, while the second, darker shape corresponds to the earliest point by which all important genes within the group had a recorded gene event. These subgroups of important genes include: all three subunits encoding ABC-type Mo(/W) transporters (ModABC, TupABC, WtpABC), subunits essential for FeMoco synthesis (NifBEN) and nitrogenase activity (NifHDK), and 9 genes essential for Moco synthesis as described by Mendel & Leimkühler (2015)6. Dates were taken from reconciliation results using the CIR clock model with an HGT cost of 3 and should be interpreted as estimates for the earliest phylogenetic evidence of Mo/W-related genes per category. Results from the symmetric median reconciliation were used for all groups except “Moco synthesis” and “AOR family” groups, where random reconciliations were only available for the proteins with the earliest recorded gene events within these groups (see methods for further discussion). A plot showing changes in levels of atmospheric O2, adapted from Lyons et al. (2014)81 (see ref. in “Methods”) are shown at the bottom, alongside estimates for marine Mo concentrations derived from Johnson et al. (2021)14. The gray shading refers to the divergence of Eukaryotic Moco synthesis, XO family, and SO family genes after the evolution of Eukaryotes. Blue spheres represent Mo-usage, while pink spheres represent W-usage by proteins within each group. Among molybdoenzymes, the DMSOR and XO families are the most ancient, with their earliest gene events dating back to ~3.7–3.1 Ga. This is consistent with previous reports of the antiquity of these enzyme families based on phylogenetic separation between bacterial and archaeal domains1,39 and patterns suggesting vertical inheritance of these enzymes34. Gene events for these molybdoenzymes remain sparse in the Archean but become increasingly frequent in the Proterozoic. This pattern is consistent with the widespread use of molybdoenzymes across different lineages later in time, potentially in response to the establishment of a fully functional Moco synthesis system. Specifically, the catalytic subunit of formylmethanofuran dehydrogenase (FwdB, FmdB) had some of the oldest gene events (~3.7–3.1 Ga), consistent with previous evidence supporting the antiquity of this enzyme and its associated metabolic pathway (methanogenesis)3,39,40. We also find early evidence of a putative selenate reductase subunit (~3.4–2.6 Ga) in support of inferences made for the antiquity of usage of selenium compounds in bacteria41. These observations are followed by the incorporation of Mo into additional enzymes during the Proterozoic, as shown by the more recent gene events of other molybdoenzymes (e.g., xanthine dehydrogenases XdhB, XDH, XdhA, YagR; nicotinate dehydrogenase NdhLM, NicB; assimilatory nitrate reductase NasA; sulfane dehydrogenase SoxC; sulfite oxidase SO) (Supplementary Table 1). These enzymes enable modern organisms to utilize alternate compounds as sources of C, N, S, and/or energy, suggesting a potential role for Mo in the expansion of metabolisms following oxygenation of Earth’s atmosphere and surface ocean. Our findings corroborate previous phylogenomic8,39 and geochemical9 studies that date Mo-nitrogenases to an early origin on an anoxic Earth. We also identify maturases (NifB and NifE), which are part of the machinery for FeMoco synthesis in modern organisms, existing as early as the Mesorchean (~3.0–2.7 Ga). These proteins share close homology with the catalytic subunits of nitrogenase (NifK and NifD), and phylogenetic studies have proposed that the latter may have emerged from maturase-like predecessors42. However, our results cannot discern the evolutionary sequence of events for nifEB and nifDK genes. Rather, they indicate that all four genes share ancient origins tied to anoxic conditions on the early Earth. Taken together, we show an ancient emergence of molybdoenzymes across different enzyme families, suggesting that Mo and W were used by early life to catalyze diverse biochemical pathways. Several important representative enzyme groups for molybdo-/tungstoenzyme families, such as sulfite dehydrogenases (SorA) and plant nitrate reductases (NR) for the SO family, were necessarily excluded from the analysis due to a low number of curated gene sequences identified from genomes. Inclusion of these genes may reveal additional pathways for ancient Mo-usage or more recent evolutionary innovations in molybdoenzymes. Nevertheless, our reconciliation results remain consistent with the established importance of Mo and W in modern organisms persisting from the Mesoarchean to the present. Discussion Our results confirm the widespread use of Mo across modern organisms inhabiting a variety of different environments and suggest that similar broad-scale Mo-usage patterns were already established in early organisms dating back to ~3.1–2.2 Ga with the emergence of a core Moco biosynthetic pathway. The ability of modern organisms to use Mo despite Mo-limited conditions has been well-established; however, Mo availability can also exert constraints on the rates and scope of Mo-dependent metabolisms in a system. We present molecular dating evidence that corroborates previous findings from phylogenetic studies1,34,39 suggesting an ancient origin for specific Mo-using pathways and supports the conclusion that bulk-level Mo scarcity in the early Earth oceans did not preclude the emergence of Mo-dependent enzymes. Considering these results, we remain mindful of the possibility that ancient enzymes may have had alternate functions or structures that are no longer observed in extant systems, which can be clarified with future efforts characterizing enzymes with genes derived from ancestral sequence reconstruction. In addition, reconciliation results must be carefully interpreted in the context of available data. The detection of gene events heavily relies on the topology of the chronogram and gene tree used, thus the discovery and inclusion of new lineages or gene sequences may alter interpretations and reveal new insights into the evolutionary history of molybdoenzymes in future analyses. In light of these limitations, we focused on identifying broad-scale trends and the relative timing for the emergence and proliferation of genes across the tree of life. Reconciliation results indicate that some of the earliest Mo-/W-related proteins to evolve were involved in Moco biosynthesis. The ancient origin of this biosynthesis pathway is well-supported by the preservation of species tree topologies in the gene trees of these enzymes and their conservation among all domains of life32,34. Indeed, the same key enzymes forming the prokaryote Moco/Wco-biosynthesis pathway have close eukaryotic homologs which catalyze the same fundamental reactions. Previous studies34 have speculated on the presence of a Moco biosynthesis pathway in the last common universal ancestor (LUCA), based on its presence and conservation across the different domains of life. Our methods do not provide the resolution needed to make such inferences, however, we estimate that the core enzymes constituting the modern Moco synthesis pathway trace their origins back to the Mesoarchean, suggesting that a complete Moco synthesis pathway may have existed as early as that time (Fig. 3). This coincides with the emergence of molybdoenzymes from different families and Mo(/W)-transport systems. Collectively, these results indicate that ancient organisms may have already been investing in Mo(/W) to use for cellular functions on an anoxic Earth, in support of previous evidence suggesting the antiquity of Mo/W-related enzymes1,34. The geologic record of black shales has been interpreted as evidence of low Mo concentrations in Archean seawater, raising the question of how distinct Mo-enzyme families emerged during this time. One potential explanation addressing this apparent mismatch in geochemical and phylogenomic records may be that suitability to biological functions was a driving factor for choosing Mo. Mo, when incorporated into enzymes, can catalyze reactions spanning a wide range of redox conditions and substrates1,25. Consistent with this idea, our analyses support an ancient phylogenetic origin for Mo-utilizing enzymes involved in a variety of essential metabolisms important to C, N, and S cycling (Supplementary Figs. 8–15). Therefore, the antiquity of several distinct Mo-enzyme families suggests that early life could have capitalized on the redox flexibility of Moco, enabling access to a wide-range of substrates and conferring a selective advantage to those capable of using Mo in metabolic reactions. We propose that the functional versatility of Mo-based enzymes has persisted through deep time, reflected in today’s widespread phylogenetic and environmental distribution of molybdoenzymes (Fig. 2). Notably, our screen of 1,609 genomes across the tree of life shows the presence of these enzymes among diverse temperatures, pH conditions, and lifestyles, supporting the wide ecological breadth of molybdoenzymes. Thus, Mo likely played a pivotal role in the earliest energy-harnessing reactions that fueled life long before the GOE. Throughout the Archean, Mo could have been sourced from submarine hydrothermal vents43,44 and released into seawater as either dissolved molybdate45 or molybdate sorbed onto iron sulfide nanoparticles46 as previously speculated1. The widespread utilization of Mo during the Archean suggests that marine Mo were present in higher concentrations than previously thought47, thereby providing a crucial driver for the early emergence of Mo-enzymes. Intriguingly, the emergence and diversification of molybdoenzymes preceding and during the GOE (Supplementary Figs. 8-15), when oxidative weathering began delivering increased amounts of Mo from land to the ocean, suggests that Mo-based biochemistry may have partly evolved in response to a shift in Mo supply, i.e., from hydrothermal vents47 to riverine input. Alternatively, many of these molybdoenzymes are associated with aerobe-associated genomes, and/or involved in the cycling of nutrient species with high redox potentials, consistent with a potential link between environmental redox conditions and nutrient pathways, such as S cycling (DmsA, SoxC, SoeA, SO) or NO2- and NO3- reduction (NxrA, NarZ, NarG, NasA) (Supplementary Figs. 8-15). Thus, the increased availability of nutrient species of high redox potential following oxygenation of the atmosphere may have facilitated the emergence and evolution of molybdoenzymes that could harness these substrates in metabolism. This observation may be specific to molybdoenzymes or potentially reflect an overall trend in the rise of oxidoreductases alongside changes in atmospheric oxygen levels. Mo storage proteins (MosAB) were the only proteins in our analysis whose earliest gene events always post-date the Archean regardless of which clock model is applied (~2.2-1.1 Ga). This presents a curious case as to why these proteins would have emerged during a time of relatively higher predicted marine Mo (though likely still lower than in the modern oceans14). A recent study suggested that competition for Mo between N2-fixers and NO3- reducers, both relying on molybdoenzyme-catalyzed metabolisms, generated a selection pressure following the GOE that drove molecular innovations tied to Mo-usage48. The relatively late evolution of MosAB storage proteins may reflect a Mo-conserving strategy developed by organisms facing Mo scarcity exacerbated by intensified competition for Mo following the GOE. Of note, the distribution of MosAB in the tree of life is restricted to select taxa (Fig. 2). This may reflect a later potential advantage of Mo storage for organisms inhabiting specific low-Mo environments49 such as highly productive euxinic waters due to increased SO42- availability after the advent of oxidative weathering of crustal pyrite50. Alternatively, Mo storage proteins could have been broadly lost in most organisms following the concomitant increase in seawater Mo concentrations driven by higher O2 levels. Alternate theories addressing the geochemical and phylogenomic discrepancies on Mo availability lean towards W-dependent metabolisms preceding Mo-dependent ones51. Similar to Mo, W can bind pyranopterin to form Wco, which is structurally analogous to Moco. In modern oxygenated marine systems, Mo is significantly more abundant than W, but this disparity decreases under sulfidic and anoxic conditions51. Given their similar chemistry as well as the reported interchangeability between Mo and W in certain metalloenzymes23, and their inverse trends in response to O251, some theories propose that the earliest molybdoenzymes initially utilized W before incorporating Mo51. Our reconciliation data don’t test this directly, but they do support the Archean origins of catalytic subunits of both Mo- and W-using enzymes, including obligate W-dependent AOR and facultative W-dependent Fmd/Fwd and Fdh (both of which have known Wco- and Moco-incorporating homologs52,53), as well as Mo-specific enzymes, such as xanthine dehydrogenase and sulfite oxidase. This demonstrates the concurrent importance of both Mo and W for early life, with evidence of Mo/W-transport systems (e.g., Wtp, Mod, and Tup) evolving between the Archean and Paleoproterozoic. In modern organisms, Mod transporters share similar binding affinity for both molybdate and tungstate, whereas Tup is tungstate-specific and Wtp has slightly higher affinity for tungstate but overall lower affinity for either anion compared to the Mod and Tup transporters54. Whether ancient transporters exhibited similar substrate affinities to their modern counterparts has yet to be experimentally characterized and tested across Mo/W concentrations predicted for different periods in Earth history. Nevertheless, the limited sequence identity between Wtp, Mod, and Tup transporters, combined with their distinct ligand specificities28 support the hypothesis that these transport systems evolved independently to meet specific metal needs of early life. The eukaryotic Mo-transporters MOT1 and MOT2 were excluded from our analyses due to the lack of KEGG ortholog numbers corresponding to these proteins. They represent a separate family of transporters, distinct from the prokaryotic ABC-type Mo(/W)-transporters, and their restriction to eukaryotes may suggest a separate evolutionary innovation within the eukaryote clade. The working conditions of molybdo-/tungsto-enzymes, particularly in relation to temperature, differ25, as reflected in the ecological separation of W- and Mo-utilizing organisms in contemporary environments (Supplementary Fig. 2). On the chemical level, tungstoenzymes require a higher temperature to attain optimal catalytic activity, which may explain their ubiquity in contemporary thermophilic archaea27. In addition, W-based redox reactions generally operate at relatively lower redox potentials and over a narrower range than molybdoenzymes25. Combined with evidence that W can antagonize molybdoenzymes and vice versa55, it is possible that Mo- and W-based enzymes evolved concurrently rather than sequentially (i.e., W before Mo), allowing early organisms to catalyze different biochemical reactions. In fact, the ability to use both transition metals may have even provided a selective advantage for organisms living in systems limited in either metal56; however, this advantage would be restricted to reactions with high enough redox potential covered by both Mo- and W-using homologs of an enzyme. Taken together, our analysis suggests that Mo utilization in biological systems originated during the Eo- to Mesoarchean, with Moco biosynthesis and molybdoenzymes playing essential roles in early metabolisms, potentially due to the redox flexibility of the latter. This expands on previous phylogenetic studies supporting the antiquity of Mo-using pathways1. Following the GOE, Mo-dependent enzymes diversified to facilitate metabolisms of oxidized substrates. Competition for Mo drove adaptive strategies such as Mo storage proteins, with their limited distribution reflecting evolutionary trade-offs linked to environmental Mo availability. Most Mo/W-related proteins investigated in this study have their earliest gene events estimated to date back to the Archean and the period coinciding with the GOE. These enzymes catalyze diverse pathways, underscoring the critical role of Mo and W in early biological productivity and (more broadly) nutrient cycling on the early Earth. Sufficient concentrations of Mo supported Mo-dependent biological processes prior to the oxidation of Earth’s atmosphere and oceans45,57 . Combined with these catalytic functions, our findings contribute to a growing understanding of an anoxic early Earth that supported Mo-based biochemical versatility, revealing a biosphere already dependent on a metal whose abundance had yet to emerge. Methods Genome and protein dataset curation for phylogenetic distribution of Mo-genes Genome and protein sequence data were curated from 1,609 bacterial, archaeal, and eukaryotic organisms. These selected organisms are a subset of 3,083 genomes included in a previously constructed reference tree of life33. Organisms were selected based on the availability of protein sequence data associated with each genome assembly (downloaded from NCBI58, August 2023). In other words, genomes of taxa with no associated protein sequence data available on NCBI were excluded from our analyses. This limitation warrants caution regarding the bias introduced toward model organisms and other well-sequenced lineages in results generated from our dataset. As described below, our dataset retains similar taxonomic diversity compared to the original dataset including a similar proportion of Archaeal lineages (n = 101, 6.3% vs n = 230, 7.5% in the original dataset); however, new patterns may emerge with the inclusion of previously uncharacterized lineages, particularly among Archaea where W usage is particularly prevalent. As a final step, all bacterial and archaeal genomes including MAGs and isolate genomes were filtered for quality (completion ≥ 90%, contamination ≤ 5%) of associated genome assemblies, predicted by CheckM259 using default settings. To ensure that no significant changes in taxonomic representation occurred, we compared the proportions of each phylum in the original dataset of genomes (n = 3,083) and our subset of genomes (n = 1609) and found similar patterns of representation were retained in our subset. Notably, the top five taxonomic groups represented in both datasets were the same with comparable proportions: CPR (26.4% and 23.9%), Proteobacteria (20.1% and 22.9%), Firmicutes (7.4% and 8.9%), Actinobacteria (5.6% and 6.7%), and Bacteroidetes (4.6% and 5.7%) for the original dataset and the subsampled dataset, respectively (see Supplementary Table 2). Protein functional annotation Nitrogenase (Nif/Vnf/AnfHDGK) subunit proteins from the curated genome dataset were manually annotated with KEGG Orthology identifiers (“KO numbers”) as follows. Candidate nitrogenase sequences were identified by HMMER v.3.3.2 (hmmsearch) using KOfam HMM profile60 queries (downloaded July 2023) against a local database of all protein sequences associated with the collected genome assemblies (NifH: K02588, NifD: K02586, NifK: K02591, VnfG: K22898). hmmsearch was performed with full-sequence (-E) and domain (--domE) E-value thresholds of <1 × 10−5. Target sequences from each hmmsearch run were aligned to previously published nitrogenase protein sequence datasets7,42 by MAFFT v.7.49061 (alignment strategy was automatically selected by the software using the –auto flag) and used to build maximum likelihood phylogenetic trees by FastTree v.2.1.1162 (WAG model) with 1000 bootstraps. The WAG model was chosen from one of two available models (WAG or JTT) in the Geneious FastTree plug-in software, where WAG has been shown to outperform JTT63. Manual annotations of nitrogenase sequences were made based on the clustering of sequences in these trees using previously described nitrogenase protein clades as reference7,42,64, as well as the presence of known, nitrogenase metallocluster ligands (e.g., FeMo-cofactor His442 and Cys275 ligands; site index from Azotobacter vinelandii NifD). We have confirmed that trees constructed from alignments of previously described nitrogenase sequences7,42 using FastTree with the WAG model recapture major nitrogenase clades. All other sequences in the protein dataset were annotated with KO numbers by HMMER v.3.3.2 (hmmscan) against a local copy of the KOfam HMM profile database60 (downloaded July 2023) using the University of Wisconsin-Madison Center for High Throughput Computing (CHTC) servers. hmmscan was performed with full-sequence (-E) and domain (--domE) E-value thresholds of <1 × 10−5. Each protein query was annotated with the KO number target with the lowest full-sequence E-value. It should be noted that KEGG does not distinguish between the Mo-using and W-using orthologs of formylmethanofuran dehydrogenase. Thus, all analyses on the subunit harboring the Mo/W-containing active site (FwdB/FmdB: K00201) of these enzymes reflect combined results for both genes. Identification of Mo/W-related proteins Initial Mo/W-related protein targets were identified by a literature search5,6,27,65–67. These include proteins involved in Mo/W transport, storage, trafficking, and cofactor biosynthesis, as well as representatives of the different molybdo-/tungstoenzymes characterized to date. Proteins annotated as described above were grouped by corresponding KO numbers, aligned by MAFFT61 and used to generate an HMM profile database by hhmake in HH-Suite68. HMM profiles corresponding to Mo-related gene targets identified by the literature search were used as queries to perform HMM-HMM homology search against the generated HMM profile database using hhsearch in HH-Suite. HMM profiles with detected homology to the initial Mo-related HMM profile queries were manually inspected for the presence of Mo-related functional domains, based on Mo-related pfam annotations available for KEGG Orthologs in the KEGG database. Confirmed homologs were used to expand the Mo-related protein target dataset. The final Mo-related protein dataset contained 102 unique KO number targets. Mo-related protein network The HMM-HMM homology search output was used to build a protein homology network using the NetworkX Python library. Edges were drawn using four different homology thresholds: log-eval > 15, 5, 0, and −5. Gene count matrix Mo-related gene counts were mapped onto a previously constructed reference species tree of life33. R packages Ape v.5.7−169 and ggtree v.3.8.070 were used to prune taxa corresponding to genomes removed based on quality and protein data availability, as described above. This resulted in a set of 1,609 genomes present in the species tree. Mo-related gene counts per genome were generated and visualized using a KO number annotation E-value threshold of 1 × 10−10, determined by hmmscan as described above. Ecological attribute prediction and correlation analysis Optimal growth temperature, pH, habitat, and O2 tolerance metadata were collected for each of the selected genomes from Hug. et al. (2016)33 by matching Genbank accession numbers to BacDive71, NCBI BioProject58, or GOLD JGI72 as well as matching NCBI taxon IDs to the phenotypic trait database from Madin et al. (2020)73. The Oxyphen package74 was subsequently used to predict oxygen metabolism for each genome missing associated data for this trait. Spearman’s rank correlation coefficient between environmental trait metadata and number of Mo/W genes were calculated in R. Gene sequence curation for reconciliation analyses To ascertain the timing of gene duplication, loss, horizontal gene transfer (HGT), and speciation events for Mo-/W-related genes, we reconciled phylogenetic trees for each gene of interest against a molecular clock containing 865 genomes selected to represent the full diversity of life on Earth. These genomes were originally selected by Boden et al. (2024)37 to represent the full range of prokaryotic diversity, however, representatives that encoded ptxD genes were selected, when possible, to cover the diversity of phosphite-oxidizing organisms. To our knowledge, there has not been a study reporting special characteristics of the Mo-protein content of phosphite oxidizers that would directly influence our results. Further, archaea encoding ptxD were not detected in this study, thus our dataset of deep-branching archaeal representatives should not be influenced by this bias. Given the wide range of prokaryote diversity covered by the Boden et al. (2024), we proceeded to use the same chronogram genome dataset for our analysis. Initial gene sequence collection for the construction of gene trees was performed by annotating protein-coding sequences from each genome with KO numbers via sequence comparisons to a Kofam HMM profile database, as described above. As an initial curation step, protein sequences corresponding to each KO number were aligned with seed alignments for corresponding TIGRFAM profiles when available, using MAFFT61 v7.505 with the --automated and --seed options. TIGRFAM seed alignments were used for access to publicly available curated protein sequence alignments. TIGRFAM also distinguishes homologous proteins with conserved function (‘equivalogs’), thus we used equivalog seed alignments when available. These seed alignments were manually inspected and supplemented with additional sequences from the literature to ensure full representation of the known diversity of each protein. When TIGRFAM profiles were not available, sequences were aligned with those of experimentally characterized proteins corresponding to respective KO numbers. Alignments were subsequently trimmed to remove column positions comprising >85% gaps with trimAl v1.4.rev1575 (-gt 0.15). Trimmed sequence alignments were then used to construct phylogenetic trees using IQ-TREE v.2.1.476 with 1000 ultrafast bootstraps and the best-fit substitution model chosen by ModelFinder77. Complex mixture models were included in this search for the best-fit substitution model using the following IQ-TREE flag parameter: -madd C10,C20,C30,C40,C50,C60,EX2,EX3,EHO,UL2,UL3,EX_EHO,LG4M,LG4X,CF4. Gene sequences were included if they met the following criteria: (1) the bit score for the sequence’s KO number annotation using hmmscan was better than the score of the corresponding Kofam HMM profile, and (2) the sequence clustered monophyletically with TIGRFAM seed alignments and/or sequences of experimentally characterized proteins obtained from the literature (Supplementary Figs. 23–68). This extra step of sequence removal was taken to ensure gene trees were constructed for proteins with identical biological function. A total of 36 genes lacked a sufficient number of final curated sequences (n ≤ 5) to enable tree construction. Additionally, some prokaryotic enzymes in the Moco synthesis pathway have distinct KO numbers from their eukaryotic homologs. Given the shared ancestry of these enzymes (also verified in our construction of phylogenetic trees using alignments from these homologs), we decided to combine curated sequences of homologous genes for our final construction of gene trees for the reconciliation analysis. Age estimation of Mo-/W-related proteins Final curated protein sequences were subsequently aligned in MAFFT61 v7.505 with L-INS-i and trimmed to remove column positions comprising >85% gaps with trimAl v1.4.rev1575 (-gt 0.15). Phylogenetic trees were constructed from trimmed alignments using IQ-TREE v.2.1.476 with 1000 ultrafast bootstraps and the best-fit substitution model chosen by ModelFinder77 (complex mixture models were taken into consideration for this search using the -madd parameter, as described above). One of our trees (corresponding to MoeA/GPHN) did not converge after 1000 MCMC iterations but was included in the study. The resulting gene trees were reconciled with four previously constructed age-calibrated species trees37 using ecceTERA reconciliation software v.1.2.478 to estimate the timing of gene events. Further details on the construction of chronograms are described in Boden et al. (2024)37. Briefly, a species tree was constructed using 16 different ribosomal proteins from 865 lineages and some topological constraints from previous studies. This tree was anchored in time using PhyloBayes79 with 8 fossil calibration points and a uniform distribution on the root prior and one of three clock models, namely Cox-Ingersoll-Ross (cir), lognormal (ln) or uncorrelated gamma multipliers (ugam). This resulted in three age-calibrated species trees, each made with a different clock model (because there is no current consensus on which is most appropriate for estimating ages in deep time). Default settings were implemented in ecceTERA for reconciliations, specifying amalgamation of Supplementary Table 3) ees with the amalgamate=true parameter and allowing transfers from extinct branches (compute.TD=true). As previously reported37, the default costs for gene events assume that HGT is more “expensive” than speciation, duplication, and loss to reduce genome size variation between parent and daughter lineages (David & Alm, 201180). Consistent with previous methods37, we also report reconciliation results using different HGT costs (2, 4, and 6, alongside the default of 3, Supplementary Fig. 69) to account for scenarios involving greater genome size variation. ecceTERA output was parsed using an adapted version of the recon2xml.pl Perl script provided by ecceTERA software and a set of custom python scripts (several adapted from Mateos et al., 202338). Reconciliation data associating gene events with dates corresponding to nodes of the species tree were visualized in R (Supplementary Figs. 8-15). A few of our gene trees (MoaA/CNX2/MOCS1; MoaC/CNX3/MOCS1; MoaD/CysO/MOCS2A/CNXG; MoeA/GPHN; and Aor) had too many sequences for computation of the symmetric median reconciliations by ecceTERA. Thus, for these genes, we report results from the random reconciliation chosen by ecceTERA. Reconciliation output from both the symmetric median reconciliations (when available) and random reconciliations using all clock models are shown in Supplementary Figs. 8-15. Supplementary information Supplementary Information Peer Review file Acknowledgements We thank Evrim Fer and Amanda Garcia for their assistance in data visualization; Holly Rucker, Kaustubh Amritkar, Ariel Anbar, Lance Seefeldt, and the members of the MUSE ICAR for valuable feedback and discussions; the Center for High Throughput Computing (CHTC) at the University of Wisconsin–Madison for providing computing resources; and the University of Wisconsin–Madison Department of Bacteriology Bioinformatics Core for valuable input. B.K. discloses funding from the National Aeronautics and Space Administration (NASA) Interdisciplinary Consortium for Astrobiology Research: Metal Utilization and Selection Across Eons, MUSE [80NSSC17K0296] with additional support from the Paglia Post-Baccalaureate Research Fellowship from Carleton College (A.S.K.) and the NASA Postdoctoral Program, administered by Oak Ridge Associated Universities under contract with NASA (MSS). Authors also acknowledge additional funding support from a NASA Astrobiology Program Grant [80NSSC18KO829] (REA) and a NERC Frontiers grant (NE/V010824/1) (J.S.B. and E.E.S.). Author contributions A.S.K.: Study design, data acquisition and analysis, writing (original draft), editing; M.S.S.: data acquisition and analysis, writing-editing; J.S.B.: writing-editing, provision of molecular clocks and advice on reconciliation analyses; E.E.S.: editing, R.E.A.: editing, provision of molecular clocks and advice on reconciliation analyses, K.O.K.: writing-editing, B.K.: Conceptualization, study design, data analysis, writing (original draft), editing, supervision, funding. Peer review Peer review information Nature Communications thanks Kevin Oliphant and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability The accession numbers, environmental metadata, gene presence/absence datasets, sequence alignments, and protein phylogenies used in or generated for this study can be found on GitHub at https://github.com/kacarlab/MoW_evo. Code availability Custom scripts used for the data analysis and visualizations of this study can be found at https://github.com/kacarlab/MoW_evo, alongside the data. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-72133-0. References Associated Data Supplementary Materials Supplementary Information Peer Review file Data Availability Statement The accession numbers, environmental metadata, gene presence/absence datasets, sequence alignments, and protein phylogenies used in or generated for this study can be found on GitHub at https://github.com/kacarlab/MoW_evo. Custom scripts used for the data analysis and visualizations of this study can be found at https://github.com/kacarlab/MoW_evo, alongside the data.