Abiotic sources of fixed nitrogen sustained early ecosystems for several hundred million years after the origin of life

Abstract

Nitrogen (N) plays a crucial role in controlling biological productivity. However, it remains unknown how Earth’s earliest ecosystems accessed bioavailable forms of nitrogen. Here, we present genomic evidence that the last universal common ancestor (LUCA) had genes for importing ammonium into the cell, but the first organisms with all three catalytic nitrogen fixing genes emerged at least 1 billion years later. Similarly, enzymatic pathways for accessing nitrogen from urea and nitriles appear to predate biological N₂ fixation. Our results imply that Earth’s earliest biosphere was maintained by environmental sources of ammonium and other N-bearing compounds, possibly derived from a combination of processes such as hydrothermal activity, photochemistry, rock weathering, lightning, or impact events. Biological N₂ fixation may have emerged in response to an increase in biological nutrient demand or due to declining abiotic supplies of ammonium, urea, and nitriles.

Genomic evidence suggests early life relied on ammonium and other N molecules before biological nitrogen fixation evolved.

INTRODUCTION

Nitrogen (N) is a critical nutrient for all life as we know it and limits primary productivity in large parts of the modern biosphere (1, 2). Today, the major source of bioavailable N (ca. 98%) is biological N₂ fixation by nitrogenase enzymes. Before the onset of biological N₂ fixation, abiotic processes must have supplied fixed N to the biosphere. Computational models and experiments have uncovered a variety of potential abiotic N sources that could have been present on early Earth (Table 1). However, the magnitudes of these fluxes are uncertain and difficult to verify. Furthermore, direct evidence for biological utilization of abiotically generated N species is so far lacking. It is therefore unknown for how long these abiotic sources were able to sustain Earth’s early biosphere and if they were limiting primary productivity (3). The sedimentary rock record has been used as tentative evidence for abiotic N input to the Eoarchean ocean (4), but strong hydrothermal and/or metamorphic alteration render this conclusion uncertain.

Table 1. Early Archean fluxes of fixed nitrogen (N) in mol·year⁻¹ based on dinitrogen (N₂) gas as starting material.

Flux type	Magnitude (mol·year−1)	Products	Supply rate	References
Lightning	109–1010	N oxide(s)*, minor hydrogen cyanide, ammonium, organic compounds including urea	Sporadic	(4, 14, 132, 133)
Volcanic eruptions	109–1011	N oxide(s)*	Sporadic	(47)
Impact shock waves	1010–1011	N oxide(s)*	Sporadic	(134, 135)
Photochemistry†	1011–1012	Hydrogen cyanide	Constant	(52, 136)
Hydrothermal vents	109–1012	Ammonium	Constant	(44, 45)
Crustal weathering	108–109	Ammonium	Constant	(46)
Modern biological N2 fixation	1013	Ammonium	Constant	(137)

^*Can undergo abiotic reduction to ammonium, coupled to iron oxidation (8).

^†Using values with high CH₄ levels, postdating the origin of life.

The most important inorganic N compound for life on early Earth would likely have been ammonium (NH₄⁺), which is the preferred form of fixed N for most microbial organisms today (5, 6). Ammonium has the same redox state as N in amino acids, meaning that no redox transformations are required for ammonium incorporation into biomolecules. In its unprotonated form, ammonia (NH₃) can diffuse passively across cell membranes, but very little NH₃ exists at cellular pH and in neutral or acidic environments, because the pKa (where Ka is the acid dissociation constant) of the ammonium/ammonia equilibrium is 9.25 (7). The efficiency of passive ammonia uptake has also been questioned because although it lacks charges, ammonia is polar (dipole moment of 1.47 D) and may therefore struggle to cross biological membranes without enzymatic support. Perhaps as a result, ammonium is often imported directly from the environment through dedicated uptake enzymes belonging to the Amt/Mep/Rh superfamily of nutrient transporters (7). Ammonium can be generated biologically via N₂ fixation, urea catabolism, or nitrile catabolism. However, before the expansion of biological N₂ fixation, hydrothermal vents may have been major sources of ammonium on early Earth (Table 1). In addition, N oxides generated during lightning, meteorite impacts, and volcanic eruptions could have been converted into ammonium through reactions with ferrous iron in seawater or on mineral surfaces (8, 9). Once inside microbial cells, the ammonium can be described as “assimilated,” but it does not support biological growth and development until the N inside it is incorporated into amino acids via additional enzymes of the glutamate dehydrogenase pathway or glutamine synthetase glutamate synthase (GOGAT) pathway (10). Enzymes of the latter pathway are nearly ubiquitous among prokaryotes, and estimated to date back to the last universal common ancestor (LUCA) (11). Therefore, we assume that if an organism had genes to import or generate ammonium, then it could also use that ammonium to support growth and development via GOGAT or GDH pathways.

Other potentially important N compounds are nitriles and urea. Nitriles (molecules with C─N triple bonds), including cyanide, may have been generated by ultraviolet (UV) radiation and in association with impact shock events (see Table 1). This class of compounds plays a prominent role in some origin-of-life models [e.g., (12, 13)]. Urea has been detected in laboratory-based lightning experiments (14) and could also have contributed to prebiotic reactions (13). If these models are accurate, then one might expect continued biological utilization of nitriles and urea beyond the origin of life. However, no data currently exist to test this idea.

Here, we leverage genomic records of enzymatic ammonium uptake, cyanide catabolism, and urea catabolism through time to determine whether these abiotic N species could have been relevant sources of N before the emergence of biological N₂ fixation. This approach rests on the assumption that ancient enzymes had the same functions as their modern counterparts, but this assumption is justified by the enzymes’ high substrate specificity. The genomic record thus overcomes some of the uncertainties associated with ancient sedimentary rocks. Our results provide fundamental insights into the ability of planet Earth to sustain its biosphere for the first several hundred million years.

RESULTS

Ammonium was bioavailable in the Hadean

To estimate when genes for biological ammonium uptake and production emerged and radiated across the tree of life, time-calibrated species trees representing three models of substitution rate inheritance [namely, Cox-Ingersoll-Ross (CIR), lognormal (LN), and uncorrelated gamma multipliers (UGAM)] were reconciled with phylogenetic trees for the primary family of ammonium uptake genes (namely, amt/mep/rh) (see Materials and Methods). Estimates made from different clock models vary, with UGAM predicting generally younger species divergences than LN and CIR (fig. S1), but they all estimate with 53 to 62% probability that the LUCA of all life on Earth encoded a gene for amt/mep/rh (Fig. 1 and figs. S2 and S3). LUCA is estimated to have existed at 4.38 to 4.39 billion years ago (Ga) [confidence intervals (CIs) = 4.40 to 4.34 Ga]. This is very close to the oldest evidence for liquid water on the planet’s surface at 4.40 Ga (15, 16), which we applied as an older bound for life. It is also slightly earlier than the date of ~4.2 Ga (CI = 4.33 to 4.09 Ga), which has been estimated for LUCA based on analyses of pre-LUCA gene duplicates (17). The second most probable origination of amt/mep/rh genes (with a 37 to 46% probability) is in the last bacterial common ancestor (LBCA), estimated to have existed at 4.31 to 3.67 Ga (CIs = 4.39 to 3.51 Ga). Put together, this suggests that fixed N in the form of ammonium and/or methylammonium was being incorporated into early microbial organisms as far back as the Hadean.

Fig. 1. Ammonium-uptake enzymes stem back to LUCA or LBCA, whereas N-fixing enzymes evolved in more recent lineages.

The most likely origins of nifH (yellow circles), nifD (pink circles), and nifK (turquoise circles) estimated with the CIR clock model are annotated with colored circles (sized in proportion to origination probability) on a time-calibrated tree of life where branch thicknesses are scaled to represent origination probabilities of ammonium-uptake genes (namely, amt/mep/rh). The grey star and associated annotation indicates the earliest lineage to host all three catalytic N fixing genes (namely, nifH, nifD, and nifK, host is defined as ≥50% presence probability for each gene). Text “nifD & nifK” indicates the first lineage with genes encoding both N₂-interacting subunits of nitrogenase (defined as ≥50% presence probabilities of nifD and nifK). GOE, Great Oxygenation Event; Had., Hadean; Phan., Phanerozoic. Blue rectangles indicate bacterial and archaeal clades (E., Euryarchaeota; D., DPANN) and phyla (from top to bottom: B.I, Bacillota I; B.A, Bacillota A; B.D, Bacillota D; B.C, Bacillota C.; B.B, Bacillota B; B.G, Bacillota G; Acti., Actinobacteriota; B.E, Bacillota E; Ar., Armatimonadota; Er., Eremiobacterota; Cya., Cyanobacteriota; Ma., Margulisbacteriota; Therm., Thermotogota; Caldi., Caldisericota; Chloro., Chloroflexi.; Patesc., Patescibacteriota; Pseud., Pseudomonadota; Myx., Myxococcota; Des., Desulfobacterota; Nit., Nitrospirota; Acid., Acidobacteriota; Pla., Planctomycetota; V., Verrucomicrobiota; Om., Omnitrophota; Bac., Bacteroidota; W., WOR-3; Spi., Spirochaetota; El., Elusimicrobiota; Tp., Thermoplasmatota, Ha., Halobacteriota; Th., Thermoproteota; Mi., Microarchaeota; N., Nanoarchaeota; Ae., Aenigmatarchaeota). The CIR clock model is presented to aid comparison with other studies that use CIR for their main figures.

Biological N₂ fixation post-dates evidence for environmental ammonium

To compare these results with the timing of when microbial communities became capable of fixing N from dinitrogen gas (N₂), we tracked the rise and spread of the catalytic molybdenum nitrogenase genes (namely, nifD, nifK, and nifH). These genes encode the structural proteins of nitrogenase, with the enzyme encoded by nifH acting to shuttle electrons to the nitrogenase catalytic center and active site encoded by nifD and nifK. All diazotrophic organisms studied to date encode all three subunits (18). Previous work has found that molybdenum nitrogenases arose before iron or vanadium nitrogenases (19–22). Our results indicate an extremely low probability that catalytic molybdenum nitrogenase genes originated in the same ancestral lineages as ammonium uptake genes (namely, ≤7% origination probability for nifD, nifK, or nifH in LUCA versus ≤4% origination probability for nifD, nifK, or nifH in LBCA) (table S1). Previous phylogenetic analyses have also indicated that LUCA lacked nitrogenase genes (17, 23, 24). Instead, our results, like others [e.g., (23)] show that the three nitrogenase structural proteins (encoded by nifD, nifK, and nifH) are more likely to stem back to more recent lineages (origination probabilities of 95 to 100% for all genes; table S2).

Our models were unable to resolve in which ancestral organisms nifD, nifK, and nifH arose because multiple lineages share high origination probabilities. These include several early Terrabacteria, Gracilicutes, and Archaea (Fig. 1 and figs. S5 and S6). Two of our three clock models (namely, CIR and UGAM) suggest that nifH is most likely to have originated in Archaea (origination probabilities total 66 to 81% for Archaea versus 14 to 27% for Bacteria). Whereas all of our analyses suggest that nifK and nifD are more likely to have originated in Bacteria (origination probabilities total 91 to 100% for Bacteria versus up to 7% for Archaea except for nifD with UGAM that has a 54% probability of originating in Bacteria compared to a 46% probability of originating in Archaea) (table S1 and Supplementary Text).

Timing of nitrogenase origination

It is challenging to discern the precise origin of nitrogenase given that it consists of several subunits with different distributions across the tree of life. For example, although nifD and nifK are generally found in the same species (84 species host both genes, whereas only two species host one of the two genes), nifH is found alone in additional species (18 in these analyses; data file S1), as has been noted elsewhere (21, 25). By reconstructing individual gene trees for nifH, nifK, and nifD, we find evidence that some nifH homologs evolved along different evolutionary trajectories to other nitrogenase subunits because the topology of our bayesian nifK and nifD phylogenies are similar (figs. S2 and S3) and different to that of the nifH phylogeny (fig. S4). Perhaps for this reason, our results often find high origination probabilities for both nifD and nifK, but not nifH in the same lineages (Fig. 1 and figs. S5 and S6). Both nifD and nifK genes are known to have evolved from a shared ancestor that duplicated to create two genes. One gene became nifK, while the other became nifD (21, 24, 26), but the two genes were dated separately for this analysis. The first lineage to host the two N₂-interacting components of nitrogenase encoded by nifD and nifK (defined as a ≥50% presence probability of both genes) was an ancestral bacterium [likely the most recent common ancestor (MRCA) of Nitrospirota and Nitrospinota] that existed in the late Paleo- or Mesoarchean at 3.37, 3.21, or 2.81 Ga (CIs = 3.49 to 2.31 Ga; table S3).

All modern diazotrophic microbes fix N with all three structural subunits of nitrogenase (nifD, nifK, and nifH), but the first microbial lineage to encode all three catalytic nif genes (defined as a >50% presence probability for nifD, nifK, and nifH) is predicted to have been an ancestral bacterium (either a Desulfobacteriota or Bacillota E) that existed more recently in geological time, specifically at 2.88 to 1.87 Ga, and potentially as far back as 3.04 Ga (because CIs = 3.04 to 1.70 Ga) (table S4).

Urea has been bioavailable as a N source since the Mesoarchean

A diverse array of Bacteria, Fungi, and plants extract N from urea using urease enzymes (27), which hydrolyse urea into ammonia. The urease enzymes of most bacteria are composed of three subunits, namely, alpha, beta, and gamma encoded by ureC, ureB, and ureA genes, respectively. Accessory genes are required to assemble urease from its subunits and insert nickel cofactors into its active site (27) but were not investigated here. In agreement with previous findings [e.g., (27)], we find that ureases are present in a wide range of bacteria, including several Cyanobacteria, Pseudomonadota, and Actinobacteria among others, but only one Archaea: Halorientalis persicus of the Halobacteriota phylum (data file S1). Specifically, 87 of the 1024 genomes investigated here (equivalent to 8%) encode all three catalytic urease subunits. A further 30 genomes encode two of the three subunits (specifically Beta and Gamma), and a further 33 encode only one subunit (either Gamma or Beta). The first organism most likely to have encoded all three subunits is estimated to have been a Terrabacterium [either the most recent common ancestor of Cyanobacteria and Sericytochromatia (CIR) or the most recent common ancestor of Cyanobacteria and Actinobacteria (UGAM and LN)] that existed between 3.87 and 3.28 Ga (CIs = 3.75 to 3.12 Ga) (table S5).

Genes encoding each individual catalytic subunit probably originated in earlier lineages before they united together in a single genome. This is because there is a 36 to 74% probability of ureC originating in a lineage that predates the first lineage to encode all three urease subunits, a 67 to 100% chance of ureB originating before the first lineage to encode all three subunits, and a 97 to 100% chance of ureA originating before the first lineage to encode all three subunits depending on which clock model is applied to estimate timing. These earlier origins (ignoring lineages with <10% origination probabilities) span from 4.15 to 3.57 Ga for ureC, 4.39 to 3.41 Ga for ureB, and 4.39 to 3.60 Ga for ureA. Whether these ancestral lineages could successfully extract N from urea with only a subset of the three catalytic urease genes remains unknown. Together, our findings indicate that microbial organisms have been extracting N (and carbon) from urea for at least 3.28 Ga and since the start of the Mesoarchean.

Microbial nitrile catabolism for 3.2 Ga

Nitriles (compounds with C≡N triple bonds, including cyanide) are catabolized by diverse bacteria, fungi, and plants (28, 29). Although their function in bacteria remains largely unknown, nitrilases likely play a role in detoxification (30), and the ammonium that is released when nitriles are catabolized (29) can serve as a source of N. In line with previous findings [e.g., (28, 29)], we find that class I nitrilase genes are relatively rare (present in 83 of the 1024 genomes analyzed), although widespread in diverse phyla of Bacteria (including Desulfobacterota, Cyanobacteria, and several Bacillota phyla), and several Eukaryotes (data file S1). There is a very low probability that LUCA could degrade nitriles (including cyanide) using class I nitrilases, because origination probabilities in LUCA range from 1% (LN) to 2% (UGAM). Similar results have been reported previously (28). Instead, there is a higher probability of nitrile-degrading nitrilases originating in more recent lineages because several have low, but noteworthy origination probabilities ≥10%. Their ages range from 3.82 Ga to 3.14 Ga (CIs span 4.06 to 2.98 Ga; table S6). This predates previous estimates of 1.88 to 1.02 Ga made using alternative molecular clock approaches (28). Where Schwartz et al. (28) implemented relaxed molecular clocks directly onto a gene tree of class I nitrilase enzymes using calibration points from plants, animals, and fungi, we implemented relaxed molecular clocks on a species tree and reconciled the resulting chronology with gene trees of class I nitrilases to estimate their origin. Our approach has the benefit of using older calibration points spanning from 3.46 to 0.91 Ga [as oppose to 0.75 to 0.25 million years ago (Ma)] to more accurately date older Proterozoic and Archean events.

Relative importance through time

To gain insight into how important each N source was to the biosphere through time, we must have a sense of how widespread each N acquisition strategy was across the lineages that are known to have existed in different time periods. When interpreting these results, note that other lineages could have existed and subsequently gone extinct, but we do not have information about these lineages, nor do we know how they acquired N, because they did not leave traces in genomic records. Ammonium uptake genes are more widespread across the tree of life today than nifD, nifK, and nifH genes for N₂ fixation, and our results estimate that this was true in all past geological eras as well (Fig. 2 and figs. S7 to S9). Since their emergence in LUCA (or LBCA), ammonium uptake genes have been present in a mean of 57 to 60% of known lineages (57% CIR, 60% UGAM, and 58% LN), whereas nifD, nifK, and nifH have each been restricted to less than or equal to 33% of known lineages with a mean of 6 to 9% (7 to 8% with CIR, 4 to 5% with UGAM, and 5 to 6% with LN). How abundant these lineages were remains unknown. Some modern metabolisms are restricted to few microbial groups but still impart sizeable effects on the planet (e.g., oxygenic photosynthesis is only present in one phylum of bacteria and some eukaryotes, but this relatively small subset of all lineages that exist today maintain the vast majority of oxygen in the atmosphere).

Fig. 2. Proportion of microbial lineages with genes for ammonium uptake (black line) and biological N₂ fixation (pink, turquoise, and yellow lines representing nifD, nifK, and nifH, respectively) in different time periods.

Proportions represent a mean of up to three values representing the number of lineages with the gene divided by the number of lineages in the species tree in a given time bin. Each value was calculated using results from a different clock model (either CIR, UGAM, or LN). Values were not counted if the time bin contained less than three lineages. Error bars represent maximum and minimum proportions. Genes are assumed to be present in a lineage when their presence probabilities are ≥50%.

On the whole, our results indicate that most of microbial lineages can import ammonium from the environment, and these constitute a far larger number of lineages than those able to fix N₂ or use N from urea or nitriles. We also observe that there is a drop from 100% of lineages containing amt/mep/rh genes for ammonium import ~4.2 Ga to ~50% of lineages with amt/mep/rh ~ 2.7 Ga (Fig. 2). Rather than suggesting a decrease in the number of lineages able to import ammonium, this instead mirrors an increase in the absolute number of lineages that are known to have existed, from just one at ~4.4 Ga to near 100 lineages at ~2.7 Ga (CIR) or ~1.7 Ga (UGAM and LN) (fig. S1). Therefore, our data indicate that some of the new lineages that appeared between ~3.6 and ~2.7 Ga lacked ammonium uptake genes. They must have been accessing N from other sources, such as amino acids (6) because genes for catabolizing nitriles appear toward the end of this window (fig. S10), and genes for catabolizing urea exhibit a similar decline (fig. S11).

Genes associated with the biological extraction of ammonium from urea and nitriles have been less widespread than ammonium uptake genes for most of the planet’s history. Class I nitrilases have existed in less than or equal to 33% of known lineages since their origin at 3.82 to 3.14 Ga, more often being present in 6 to 7% of known lineages based on means of 26 to 29 time bins, each spanning 100 Ma (ranges apply due to differing results from different clock models). The three genes for catabolizing urea have always been somewhat more widespread than class I nitrilases for accessing N from nitriles and nif genes for fixing N, but they have remained less widespread than amt/mep/rh for ammonium uptake. This is because ureA, ureB, and ureC have always been present in less than or equal to 57% of known lineages, with a mean of 10 to 22% since their origins in the Mesoarchean (data file S2). Therefore, our results indicate that environmental ammonium has been important for a greater diversity of microbial species than urea, nitriles, or dinitrogen gas throughout most of Earth’s history.

Although we find that genes for ammonium uptake are more common among microbes than some other N acquisition strategies, we note that our finding that 57% of extant lineages encode Amt/Mep/Rh enzymes differ from previous reports of these genes being near ubiquitous in sequenced bacterial genomes (7). Our search methodology correctly identifies homologs in species that have previously been reported to use ammonium transporters from the Amt/Mep/Rh family (table S7), so differences (between the near ubiquity of ammonium transporters in sequenced bacterial genomes, and their absence from 43% of the genomes analysed here) may be a result sequencing bias. Most (~90%) of sequenced genomes are from just 20 bacterial strains (31), with diverse samples of the type investigated here (to capture the full diversity of life) representing a small minority of public databases. As a result, our findings support the premise that amt/mep/rh are widespread among prokaryotes and elaborate further by suggesting that they are absent from some bacterial phyla, including Patescibacteria and WOR-3 (fig. S12).

DISCUSSION

Reconciling genomic and geochemical records

Nitrogen is one of the most important components of biological molecules, so its relative environmental abundance acts as a key control on the size of the biosphere. Understanding the range and scope of strategies used to acquire N through Earth history can give us insights into which N sources may have been most important to the biosphere over time and thus what factors have controlled primary production over time. The results of our phylogenetic analyses suggest that microbial communities were genetically capable of importing N from environmental pools long before the appearance of any of the nif genes used for biological N₂ fixation. Ammonium appears to have been the oldest N source for Earth’s earliest biosphere, but the ability to catabolize nitriles and urea also emerged early, potentially before life acquired the ability to metabolize N₂ gas (Fig. 3). The onset of biological N₂ fixation marks an important reference point in the evolution of the N cycle because this metabolism made the biosphere independent from planetary processes that generate bioavailable N forms. While our analyses indicate that biological N₂ fixation with nitrogenase was not present in LUCA, there are a number of ways to interpret the data.

Fig. 3. A history of microbial N sources through time.

Dark colored bars indicate time periods when all three clock models agree that the metabolism existed, light colored bars represent the oldest age predicted by any of the three clock models, and dashed lines represent the oldest confidence intervals. Isotopic evidence of biological N₂ fixation (BNF) is presented in black (see main text for references), with the gray area representing more putative isotopic evidence of biological N₂ fixation. Paleoarc. represents the Paleoarchean, Mesoarc. represents the Mesoarchean, Neoarc. represents the Neoarchean, Neoptz. represents the Neoproterozoic.

Most microbial species that fix N₂ (all except thermophilic Archaea, which lack nifN) generate NH₄⁺ from N₂ using a full set of nifD, nifK, and nifH as well as the maturases nifE, nifN, and nifB (18). Our results estimate that the first N₂-fixing lineage with all three catalytic subunits (namely, nifD, nifK, and nifH) existed at 2.88, 2.79, or 1.87 Ga (depending on the clock model), which is consistent with some other evolutionary biology studies estimating that molybdenum nitrogenases evolved at 2.11 to 1.72 Ga (24) or ~2.33 Ga (21). If nifD and nifK evolved from maturase-like ancestors, as has been previously suggested (26) and contested (21, 24), then this first lineage with nifD, nifK, and nifH (the complete set of catalytic N₂-fixing genes) may also have had nifE and nifN. However, some of these estimates, particularly the younger Paleoproterozoic estimate generated with the uncorrelated clock model, are incompatible with putative 2.1-Ga-old fossils of N₂-fixing cyanobacteria (32) and the existing tenet that biological N₂ fixation (as an oxygen-sensitive process) evolved in anoxic settings (33) before oxygen accumulated in the global atmosphere.

A Neoarchean to Paleoproterozoic origin of biological N₂ fixation (between 2.88 and 1.87 Ga) would also be inconsistent with some of the earlier sedimentary N isotope records (δ¹⁵N = [(¹⁵N/¹⁴N)_sample/(¹⁵N/¹⁴N)_air − 1] × 1000) of N₂ fixation at 3.2 Ga. Biological N₂ fixation with Mo-based nitrogenase (Nif) imparts very small isotopic fractionations [average of −1.4± 1.0 ‰; (34)], and these values have been documented from well-preserved sedimentary rocks back to 3.2 Ga and possibly 3.4 Ga (35–37). Biological N₂ fixation may extend closer into this range because a recent evolutionary biology study estimated that all three catalytic subunits were present in an ancestral bacterium that existed ca. 3 Ga (23). This slightly predates our earliest estimate of 2.88 Ga due to the application of methods for interpolating between species tree nodes to get higher resolution age estimates and potentially also due to differing selection of clock models and species. However, it does not quite reach back into the older isotopic evidence for N₂ fixation. This gap between early geochemical signatures of N₂ fixation and the evolution of some N₂-fixing genes has been noted previously [e.g., (22)].

As a result of these discrepancies, we propose that the first N₂-fixing lineages were able to convert N₂ into NH₄⁺ without nifH. Genes encoding NifH, NifD, and NifK have been assumed to have followed the same evolutionary trajectories due to their connected roles [e.g., (38)], but nifH exists in a number of organisms that do not fix N (25). There might therefore be different reasons (or selective advantages) to inherit and keep nifH. In this study, we relaxed the assumption that nifH, nifD, and nifK follow the same evolutionary trajectories and found that nifH emerged before other nitrogenase genes in different time periods (Table 2) and lineages (Fig. 1 and Supplementary Text), similar to the results of other studies with the same assumption (20, 23). Of the six genes encoding molybdenum nitrogenase, only nifD and nifK, encoding dinitrogenase/component I, interact directly with N₂. The first lineage to host these two genes existed at 3.37, 3.21, or 2.81 Ga, depending on the clock model used (table S3). Two of these estimates are contemporaneous with early geochemical records of biological N₂ fixation at 3.2 Ga. As these organisms would have lacked nifH, whose gene product donates electrons to dinitrogenase/component I to allow them to fix N₂, an alternative reductase would have been required. Several nonbiological catalysts (including cadmium sulfide, cobaltocene, polycarboxylates, and polyaminocarboxylate ligated Eu²⁺) can reduce dinitrogenase/component I in the absence of NifH (39–41). Most limit the dinitrogenase complex to catalyzing the conversion of nitrite (NO₂⁻), azide (N₃⁻), or hydrazide (N₂H₄) to ammonium. However, cadmium sulfide can promote biological N₂ fixation from dinitrogen gas to ammonium at 63% of the adenosine 5′-triphosphate–coupled reaction rate provided by NifH (42). Cadmium is rare in seawater (43), so the natural relevance of this mechanism remains speculative. If alternative enzymatic reductases existed in the past and were less efficient than nifH similar to cadmium sulfide, one could imagine that they would easily have disappeared from the genomic record when microbes with nifD and nifK acquired nifH in the Neoarchean to Paleoproterozoic. Therefore, further laboratory-based research into whether N₂ fixation is possible with alternative enzymatic reductases is warranted.

Table 2. Estimated dates (in Ga) for nodes with ≥ 10% origination probability for N-fixing genes.

CI, confidence interval.

Gene	Clock model	Origin probability	Median age/Ga	Oldest CI/Ga	Youngest CI/Ga
nifD	CIR	25%	3.222	3.387	3.075
		15%	3.296	3.459	3.158
		11%	3.331	3.496	3.182
	UGAM	32%	2.938	3.134	2.720
		14%	3.375	3.567	3.172
	LN	27%	3.367	3.486	3.247
		18%	3.559	3.675	3.438
		14%	3.372	3.502	3.236
nifH	CIR	33%	3.191	3.280	3.097
		11%	3.484	3.548	3.461
		10%	3.307	3.388	3.226
	UGAM	50%	2.938	3.134	2.720
	LN	18%	3.372	3.502	3.236
		16%	3.002	3.126	2.867
nifK	CIR	19%	3.296	3.459	3.158
		13%	3.209	3.374	3.071
		12%	3.222	3.387	3.075
		12%	3.331	3.496	3.182
		10%	3.053	3.215	2.899
	UGAM	19%	3.375	3.567	3.172
		12%	2.979	3.186	2.769
		11%	3.592	3.790	3.397
	LN	32%	3.367	3.486	3.247
		16%	3.175	3.310	3.034
		15%	3.193	3.330	3.051

Abiotic sources of ammonium on early Earth

Our phylogenetic analyses reveal a gap of several hundred million years between the emergence of biological assimilation of ammonium or methylammonium from the environment and enzymatic conversion of N₂ into ammonium within cells regardless of whether that required all three catalytic nitrogenase genes or just nifD and nifK. This long gap suggests that ammonium was freely available in the environment and able to sustain Earth’s earliest biosphere for a prolonged period of time until biological N₂ fixation emerged. This ammonium could have been generated by a variety of abiotic sources, such as hydrothermal reduction of N₂ by catalytic minerals within oceanic crust (44, 45), continental weathering of ammonium-bearing minerals (46), and Fe²⁺-coupled reduction of N oxides (8, 9) generated by lightning or volcanism (4, 47, 48). All of these processes have been proposed for early Earth and may also operate on other terrestrial planets. Hence, our results combined with those of others such as the tracing of glutamine synthetases and nitrite reduction enzymes back to LUCA by de Carvalho Fernandes et al. (11) and Moody et al. (23) suggest that fundamental planetary processes may be able to sustain a (small) biosphere with N over geological timescales.

Sedimentary δ¹⁵N data from before 3.2 Ga, when abiotic N sources were likely dominating, are generally difficult to interpret due to metamorphic and/or hydrothermal overprinting, which may have added or removed N and introduced isotopic scatter [reviewed by (35)]. However, it is also possible that these scattered data from the Eo- and Paleoarchean (before 3.2 Ga), which range from −4 to +11‰ (49), reflect at least in part microbial utilization of abiotic N sources, as suggested by our genomic results. Attempts to reconstruct primary δ¹⁵N values from highly metamorphosed rocks at 3.7 Ga yield negative values down to −7‰ that could be consistent with lightning or photochemical N input, for example (4). Alternatively, the assimilation of ammonium provided by hydrothermal processes could have played a role and contributed to these signatures (50).

It has been suggested that primary productivity on early Earth, before the onset of photosynthesis, was perhaps three to five orders of magnitude lower than it is today (3). Hence, relative to the rate of modern biological N₂ fixation, which is on the order of 10¹³ mol N year⁻¹ (51), abiotic source fluxes of 10⁸ to 10¹⁰ mol N year⁻¹ would perhaps have been sufficient to maintain early life on a global scale. The required range of abiotic N input for such a small biosphere is consistent with all proposed abiotic N source fluxes, even at their lower limits (Table 1).

Early nitriles and urea

Some models propose that nitriles (molecules with a C≡N triple bond) were critical for the origin of life because nitrile molecules can undergo reductive homologation to form precursors of important biomolecular building blocks (such as RNA, proteins, and lipids) in the presence of UV radiation (12). Sources of nitriles include photochemical reactions in the presence of high CH₄ levels (52) and asteroid impacts (53). Asteroids themselves contain cyanide and nitriles, and when they hit the ground, additional nitrile compounds are generated due to the elevated pressures and temperatures (54). While we cannot distinguish between nitrile sources using our methods, our data provide an encouraging indication that nitriles were indeed present or at least being metabolized by early microbial life on the Mesoarchean Earth. This finding lends credence to the involvement of nitriles in prebiotic chemistry and indicates that nitriles may have continued to support life beyond its origins.

Likewise, we find early evidence of urea utilization that may also stem back to prebiotic conditions. Urea is generated in small amounts during lightning reactions (14), in space (55), via the degradation of some cyanides by UV light (56), and (if ammonia is present) via spontaneous formation on water droplets such as fog or sea spray (57). As a result, urea may have contributed to the origin of life by reacting with nitriles (specifically cyanoacetaldehyde) to create two of the five bases found in RNA and DNA (13). Similar to nitriles, our data suggest that urea continued to play a role in microbial metabolisms long after the establishment of the biosphere.

What drove the emergence of N₂-fixing genes in the Paleo- to Mesoarchean or Paleoproterozoic?

As abiotic processes were evidently capable of sustaining Earth’s biosphere for several hundred million years, what drove the emergence of N₂-fixing genes in the Paleo- to Mesoarchean or Paleoproterozoic? There is no consensus on what selective pressures may have favored the evolution of biological N₂ fixation (33), but one possible explanation is an increase in productivity, driven by the rise of photosynthesis (22). The rise of anoxygenic phototrophs to ecological dominance may have increased primary productivity by a factor of 10 compared to ecosystems dominated by purely chemoautotrophic metabolisms (58–60). Another increase by a factor of 10 may have been achieved by the emergence of oxygenic phototrophs (namely, Cyanobacteria) (3). The antiquity of anoxygenic phototrophs is not well constrained, but the origin of Cyanobacteria can be traced back to as early as 3.3 Ga (61–63). If abiotic sources of fixed N were at the lower end of proposed fluxes (Table 1), then the increased demand for bioavailable N resulting from such an increase in primary productivity caused by the evolution of Cyanobacteria would have been difficult to sustain, perhaps leading to a selective advantage for organisms that had nif genes. This advantage must have been substantial, given that biological N₂ fixation is energetically costly (33). This interpretation does not require that early phototrophs themselves were capable of fixing N₂ because they could have benefitted from biogenic ammonium released into the environment by other N₂ fixers within the ecosystem.

Alternatively, it is possible that the emergence of nif genes was driven by a decline of abiotic N sources. Trace element data in chemical sedimentary rocks (cherts and iron formations) suggest a progressive decline in hydrothermal activity from the Eoarchean (before 3.5 Ga) into the Mesoarchean (after 3.2 Ga) (64). The magnitude of this decline has so far not been quantified, but it is conceivable that it curtailed the supply of hydrothermal ammonium, which was one of the largest sources of abiotically fixed N on early Earth (Table 1). It is also possible that a shift in atmospheric composition affected the rate of N oxide production during lightning (48). On the anoxic Archean Earth, the O atoms required for the formation of N─O bonds during lightning would have been largely derived from CO₂, and hence, a decline in atmospheric pCO₂ over time could have suppressed NO_x production. This would in turn have limited the supply of ammonium formed from NO_x reduction via Fe²⁺ oxidation (8). Likewise, a decline in CH₄ levels could have suppressed the generation of HCN by photochemistry (52). However, climate reconstructions of the Paleoarchean are not yet precise enough to test these hypotheses.

In summary, we reveal the first indirect evidence that abiotic processes operating on the surface of the early Earth were able to maintain a sufficient supply of bioavailable N to sustain microbial ecosystems for several hundred million years after the origin of life. These N compounds included ammonium as well as urea and nitriles, all of which have been documented in laboratory experiments and models simulating early Earth conditions (Table 1). Based on these simulations, abiotic N₂-fixing processes have long been invoked for prebiotic scenarios that led to the formation of amino acids and nucleobases. Our results show that abiotic conversion of gaseous N₂ into bioavailable forms sustained the biosphere long after the emergence of life and extended well into the Archean eon.

Our findings have implications for the habitability of other terrestrial planets with active volcanism and hydrological cycles at their surface because these conditions can evidently be sufficient for supplying bioavailable N—an essential nutrient for carbon-based life—through a variety of abiotic mechanisms. Together, our results suggest that abiotic sources of ammonium may be sufficient for the development of a substantial surface biosphere, with important implications for understanding the coevolution of life and planetary biospheres over time.

MATERIALS AND METHODS

Species tree reconstruction

Taxa selection

Our evolutionary tree of life included 1025 strains (data file S1) that were chosen to represent the full diversity of Bacteria and Archaea. To select representative strains, we included one genome from each order as defined by the Genome Taxonomy Database (GTDB) version 220 (65, 66). Where multiple genomes were available for each order, we selected the genome with the highest completeness as estimated by CheckM2 (67). All genomes with less than or equal to 90% completeness or more than or equal to 5% contamination as estimated by CheckM2 (67) were discarded.

The resulting taxon selection was then balanced to maximize phylogenetic signal following the methods of (68). Briefly, this involved removing phyla with fewer than five representatives and downsampling overrepresented phyla. Overrepresented phyla are defined as those with more than “S” representatives, calculated using the equation below, where “N” represents the number of genomes present in the phylum and “Quant” represents the 0.9 quantile of the number of genomes in each phyla before downsampling. This quantile was calculated in R (R Core Team, 2022) using the “quantile” function

$$S=N\left[1-{\left(\frac{N-\mathit{\text{Quant}}}{N}\right)}^2\right]$$

Three overrepresented phyla were identified (namely, Pseudomonadota, Patescibacteria, and Chloroflexota) and downsampled to ensure that molecular clocks would be able to estimate an accurate age for the origin of the crown group of the phylum. To do this, the full GTDB phylogenetic tree of Bacteria was downloaded, rooted on the most recent common ancestor of a Gracilicute (namely, the Pseudomonadota with GTDB accession RS_GCF_000297215.2) and Terrabacteria (namely, the Actinobacteria with GTDB accession RS_GCF_000384115.1) based on the findings of (69), and labeled in TreeViewerCommandLine version 1.3.1 (70) to highlight genomes flagged for inclusion. Of these, random genomes were manually removed to reach the balanced number of genomes (S) required for that phylum. If the removal of the selected genome would have changed the MRCA of the phyla in question, that genome was kept so a valid date for the crown ancestor of each phylum could be generated in future steps.

We added several additional genomes in order to calibrate the molecular clock with microfossils and geochemical records spanning a wide range of phylogenetic groups and geological time intervals. These additional genomes included nine nuclear eukaryotic genomes, four mitochondrial genomes, four plastid genomes, six cyanobacterial genomes, two anoxychlamydial genomes [a clade within Chlamydiota that donated three subunits of hydrogenase maturase to stem eukaryotes (71)] and two chromatiales genomes [an order of Gammaproteobacteria which have left okenane biomarkers in the geological record (72)]. If any of these genomes belonged to the same order as another genome in the tree, the other genome was removed to maintain the balanced taxon selection achieved earlier (with the exception of genomes that required two or more representatives from the same order to apply the calibration).

Marker gene selection

We created our tree of life using an alignment of 28 protein-coding genes, selected based on their ubiquity and absence of interdomain transfers in their evolutionary history. First, we compiled a list of 79 marker genes that had been used in several previous tree-of-life studies [namely, (68, 73–77)]. From this list of 79, we discarded any genes that had experienced interdomain horizontal gene transfer as tested and reported by (17). For the remaining 30 genes that had not been tested by (17), we reconstructed rough maximum-likelihood phylogenetic trees, using amino acid sequences. Homologs of each marker protein were identified in each of the 1025 genomes using HMMER version 3.4 (78) using HMM profiles downloaded from PFAM (79). The resulting hits were filtered using a custom python script (called filter.py), which selected the best hit for each marker protein in each genome based on the highest scoring alignment region. If that hit was truncated (defined as envelope length < 0.7) or likely to be found due to chance as opposed to homology (defined as bitscore <66 when the subject sequence was a single genome), then it was removed. These were aligned in MAFFT version 7.490 (80) using either the iterative E-INS-i strategy (--genafpair –maxiterate 1000) if less than 200 homolog were present or a faster progressive E-large-INS-1 strategy (--mpi –large –genafpair) if more than 200 homologs were present. Poorly sequenced residues were removed using TAPER (81), and uninformative columns containing gaps in ≥90% of sequences were removed using TrimAl version 1.4.rev22 (82) following the methodology of (68, 83). Rough maximum-likelihood protein phylogenies were then generated in FastTree version 2.1.11 OpenMP (84) with default parameters. These trees were annotated with TreeViewerCommandLine version 1.3.1 (70) to distinguish between homologs from Archaeal and Bacterial genomes based on GTDB taxonomy. Any marker proteins with clades indicative of paraphyly (indicating duplication in the history of the gene encoding that protein) or substantial interdomain horizontal gene transfer were removed from our list. Rare marker genes (present in <50% of genomes) were also removed from the list, unless they were predominantly Archaeal to maintain the representation of Archaea in a dataset where Archaea represented <10% of the total genomes (largely because fewer Archaeal orders have been identified on the GTDB). Together, this left us with a set of 28 marker genes with minimal evidence of paraphyly and interdomain HGT. Twenty four of these genes were found in both Archaea and Bacteria, three were specific to Bacteria and one was specific to Archaea. The few domain-specific proteins assist with accurate phylogenetic positioning and dating within each domain, whereas the multidomain markers inform evolutionary relationships both within and between the domains.

Alignments for phylogenetic reconstruction

Homologs of these 28 marker proteins (selected for their ubiquity and lack of interdomain horizontal gene transfer) were used to reconstruct the phylogenetic tree. Rather than using all of the marker gene homologs that had been identified above, we carefully removed homologs that could blur the phylogenetic relationships between strains as a result of long-branch attraction (caused by abnormally high levels of mutation), gene duplication, or horizontal gene transfer. To identify homologs for removal, we reconstructed more accurate protein phylogenies for each marker gene and manually removed sequences which fell on abnormally long branches, did not group with their respective domain (unless they were from eukaryotes), or exhibited branching patterns indicative of paralogs. These accurate protein phylogenies were reconstructed in IQ-TREE version 2.2.5 (85) using complex mixture models and 1000 ultrafast bootstraps (86) with more than the default number of iterations (specified with -nm 5000) to increase the chances of ultrafast bootstrap correlation coefficients reaching convergence (at the default correlation coefficient of ≥0.99). Most protein phylogenies (19 of 28) were reconstructed with the best-fitting substitution model out of C10 combined with F for empirical state frequencies observed from the data and G4 or R4 for substitution rate heterogeneity with LG, Q.pfam (87), or Q.bac (88) (command -madd LG + G4 + C10 + F, LG + R4 + C10 + F, Q.pfam + G4 + C10 + F, Q.pfam + R4 + C10 + F, Q.bac + G4 + C10 + F, and Q.bac + R4 + C10 + F) and the standard models implemented in ModelFinder (89). However, because of time restraints, some genes (7 of 28) were reconstructed using the best-fitting substitution model out of the same set of mixture models and a reduced set of the standard models implemented in ModelFinder (namely, the base models LG, WAG, Q.pfam, and Q.bac). A further two genes used the same reduced set of standard models and mixture models, but without the -nm option to increase the maximum number of iterations. The alignments used as input for these phylogenies were generated as above, but this time all proteins were aligned using the accurate E-INS-I strategy (--genafpair –maxiterate 1000). In total, 64.6% of homologs of marker protein PF00009, 4.8% of homologs from PF00118, 0.5% of homologs from PF00154, and 0.2% of homologs from PF05000 were removed because they were on abnormally long branches, paralogous, or nested among homologs from a different domain.

We then removed genomes that could not be placed accurately in the species tree because they contained fewer than 15 of the resulting 28 marker proteins, unless those genomes belonged to organelles (namely, mitochondria or plastids) or symbiotic phyla [namely, Patescibacteria and DPANN including Nanoarchaeota, Nanohaloachaeota, Micrarchaeota, Iainarchaeota, Aenigmarchaeota, UAP2, and Altiarchaeota (90)], which are expected to have undergone genome reduction. These genomes were removed before the taxon balancing described above was completed. The final balanced genomes encode a mean of 24 marker proteins (standard deviation of 2.5). Mitochondrial genomes encode the fewest marker proteins, with each of the four mitochondrial genomes encoding one, three, five, or seven marker genes.

Phylogenetic reconstruction

Appropriate amino acid substitution models for each marker protein were selected using ModelFinder (89) with the -m MF + MERGE option to find the best partitioning scheme (91), the -safe option to avoid numerical underflow, −rcluster 5 to speed up the computation (a.k.a. make it complete in a matter of months instead of years) by only considering the top 5% of partitioning schemes, and an additional parameter to specify the addition of six complex mixture models (including LG + G + F + C60, LG + G + C60, Q.pfam + G + F + C60, Q.pfam + G + C60, Q.bac + G + F + C60, and Q.bac + G + C60). Two separate runs were conducted, one with the -p option to allow each partition to have its own evolutionary rate, and another with the -Q option to additionally account for heterotachy [as described in (92)]. The resulting partitioning scheme (91) chosen with the -p option had a better (closer to zero) Bayesian information criterion (BIC) score than the best partitioning scheme chosen with -Q. Therefore, an amended version of this partitioning scheme (with the original 13 partitions, but where mixture models including “C60” were modified to include “C10” instead due to computational restraints) was used to reconstruct our species tree in IQ-TREE version 2.2.5 (85), using the “-safe” option to prevent numerical underflow. Branch supports were estimated with 1000 replicates of ultrafast bootstrap approximations (86) the “-bnni” option to reduce the risk of overestimating ultrafast bootstrap branch supports and the SH-aLRT test (93). Three independent replicates were completed, and the tree with the best (a.k.a. closest to zero) log-likelihood value was retained for further analyses. The resulting phylogenies were visualized in TreeViewer v2.1.0 (70) and FigTree v.1.4.4 (94).

Species trees made using different methods sometimes estimate different relationships between phyla [e.g., as demonstrated here (83)], so we incorporated information from previous studies that thoroughly investigated macroevolutionary relationships between large clades of Bacteria and Archaea [including (68, 69, 75, 90, 95–100)] by applying a topological constraint on the relationships between Gracilicutes, Terrabacteria, Patescibacteria (formerly known as the “Candidate Phyla Radiation”), and Euryarchaeota (including Halobacteriota, Thermoplasmatota, Methanobacteriota, Hydrothermarchaeota, and Hadarchaeota) in our tree (fig. S13). This has been applied before in (101).

Divergence time estimation

Our species tree was anchored to a geological timeline by implementing three relaxed Bayesian molecular clocks in Phylobayes version 4.1 (102). Creating a molecular clock with an alignment comprising 4510–amino acid columns and accurate evolutionary models was computationally unfeasible, so the 28 marker protein alignments were randomly subsampled using a custom code (SubsampleAlignmentRandomly.py) to retain 50% of the original alignment columns from each marker protein for molecular dating. Similar random sampling has been used in other molecular clock studies [e.g., (96)] because it ensures that sequence information from all strains and marker proteins are used for dating. Our resulting alignment contained 2264 amino acid positions. Topology was fixed to match the species tree, and substitution rates modeled using a uniform exchangeability matrix and an empirical profile mixture model with 20 profiles (specified using -catfix C20) (103). Divergence times were estimated under one of three models to reflect differing views on the inheritance of substitution rates between mother and daughter lineages, namely, lognormal (LN), CIR (104, 105), and UGAM (106).

The first diversification in our clock, representing the root and the first radiation of LUCA into Bacteria and Archaea, was set with a uniform distribution spanning 4.4 to 3.5 Ga based on two assumptions: (i) Life could not have been present before liquid water was present on the planet’s surface (15, 16), and (ii) LUCA must have emerged before the earliest generally accepted microfossils (107).

We also implemented 14 calibration points based on microfossils of Cyanobacteria, microfossils of algae, and geochemical evidence for methanogenesis and oxygenic photosynthesis on a scale powerful enough to cause the Great Oxygenation Event (GOE) (table S8). Given recent data suggesting that oxygenic photosynthesis evolved before crown Cyanobacteria emerged (108, 109) (the term “crown” is used to refer to the most recent common ancestor of all extant Cyanobacteria), we note that the findings reported in this manuscript are based on our assumption that the crown group of the most successful modern oxygenic phototrophs in the prokaryotic domain of life (Cyanobacteria) must have been present when the atmosphere first became oxygenated at the GOE. This does not refute the possibility that other oxygenic photosynthetic lineages could have existed at this point or that more ancient ancestors in the cyanobacterial stem branch or earlier could have been photosynthetic, but it does assume that they were not efficient or widespread enough to cause oxygenation on a global scale. Further analyses may prove this wrong, but at present, our assumption that crown group Cyanobacteria emerged somewhere between the first whiffs of atmospheric oxygen at 3 Ga and the onset of the GOE at 2.32 Ga is in-line with previous molecular clock analyses dating crown group Cyanobacteria to the Neo- or Mesoarchean (63).

Many of these calibrations have been applied previously [e.g. (62, 101, 110–116)]. Here, we build on the calibrations applied in (101) by assuming earlier methanogenesis >3.46 Ga (117–119) and adding additional calibrations for purple sulfur bacteria, crown eukaryotes, anoxychlamdiales, metazoa, and embryophytes.

Twelve independent chains were run for 20,532 (UGAM), 97,115 (CIR), or 113,798 (LN) cycles and considered converged when two replicate chains passed all of the following criteria after discarding the first 3500 (for clocks made with CIR and LN) or 1500 (UGAM) cycles as burn-in: (i) relative differences of all parameters <0.3, (ii) effective sizes of all parameters >50, (iii) effective sample size (ESS) of all parameters >200, and (iv) Rhat of all parameters <1.05. These relative differences and effective sizes were calculated in tracecomp implemented in phylobayes v4.1 (102), whereas Rhat and ESS values were calculated with a custom R script (phylobayes_convergence_statistics_fixed.r). The number of chains to discard as burn-in was chosen by visual assessment of traces viewed in Tracer v.1.7.0 (120).

Identification of N genes

We searched in our molecular clock genome taxa for homologs of genes involved in key pathways for the uptake of ammonium and other N compounds using HMMER3 (78). First, query sequences (in the form of “reviewed proteins”) were downloaded from relevant entries on InterPro 104.0 (table S9) (121). These were aligned with MAFFT v.7.4 (80) using genafpair (E-INS-i) and converted into HMM profiles using hmmbuild with default parameters. The resulting HMM profiles were used to search for homologs in the genomes of our molecular clock (all genomes were concatenated and treated as a single subject) using hmmsearch with an e value threshold of 0.001 (recommended in the HMMER3 user guide to minimize the chances of finding false homologs). Hits shorter than 70% the length of the shortest query sequence were removed. Ammonium transporters and nitrite nitrate symporters require transmembrane helices to transport N compounds into and out of the cell (7, 122), so for these proteins only, hits that did not contain transmembrane helices [estimated by DeepTMHMM (123)] were removed.

A cautious approach to gene identification was taken by discarding all homologs that did not descend from the MRCA of the query sequences. Similar approaches have been used to identify phosphorus-cycling genes (101, 124), compatible solute genes (61), light-harvesting genes (112), and arsenic-resistance genes (125) previously. Here, we identified genes that descended from the MRCA of query sequences by aligning, removing poorly sequence residues, and reconstructing rough species trees using the same methods that were used to look for evidence of interdomain HGT in the marker genes used to reconstruct the species tree. An additional step of rooting the gene trees using minimal ancestor deviation (with the -t option to retain branch lengths shorter than 10⁻⁶) (126)—one of the two best methods of rooting prokaryotic gene families (127)—was added to identify which branch corresponds to the common ancestor of all query sequences. Hits that passed all previous thresholds but did not descend from the MRCA of the query sequences were discarded. An additional step was added to distinguish class I nitrilases with activity on nitrile molecules (with C≡N triple bonds) from 12 other classes of nitrilases that act on different N- and carbon-containing molecules [often amides, (29)]. This involved removing sequences (totalling 1961 of 2051), which did not have a signature Cys-Trp-Glu amino acid motif (beginning at position 2180 in our alignment) unique to class I nitrilases (29), then realigning, and trimming the remaining class I nitrilases.

A different additional step was used for NifH, NifD, and NifK to ensure that closely related bacteriochlorophylls with different functions (21, 24, 128) were removed. This involved adding 138 reviewed amino acid sequences of BchX (from CD02044 and IPR010246) and BchL/ChlL (from CD02033 and IPR010246) to the alignment of NifH, 174 reviewed amino acid sequences of BchN/ChlN (from IPR005970, IPR050293, NF002768.0, and NF005867.0) to the alignment of NifD, and 117 reviewed amino acid sequences of BchB/ChlB (from IPR005969 and IPR016209) to the alignment of NifK before filtering out sequences that did not descend from the MRCA of the relevant N fixing proteins. Fortunately, all bacteriochlorophylls were outgroups of these. Maturases (NifE and NifN, otherwise known as “assembly scaffold proteins”) were not kept for further analyses because they cannot efficiently reduce dinitrogen, although laboratory studies have shown that their ancestors may have been capable of limited N₂ fixation (129).

Evolutionary trajectories of N uptake genes

Gene trees were reconstructed from amino acid alignments generated with the genafpair option (E-Ins-i) in MAFFT (80). Poorly sequenced residues were removed, and gappy regions were trimmed with the same criteria that have been applied throughout. For each gene, phylogenies were reconstructed in either MrBayes version 3.2.7a (130) (nifD, nifK, nifH, and class I nitrilases) or IQ-TREE version 2.2.5 (amt/mep/rh, ureC, ureB, and ureA) if MrBayes trees did not converge within 75 million generations. IQ-TREE was parameterized to use the same complex mixture models and parameters that were applied to identify and remove long branches from most of the marker genes (namely, LG + G4 + C10 + F, LG + R4 + C10 + F, Q.pfam + G4 + C10 + F, Q.pfam + R4 + C10 + F, Q.bac + G4 + C10 + F, Q.bac + R4 + C10 + F with 1000 ultrafast bootstraps), with the exception of -nm 5000, which had to be removed due to computational constraints. An additional option, -wbtl was added to output ultrafast bootstrap trees representing uncertainty in branching relationships. MrBayes was parameterized with a mixed amino acid model prior, invariant sites, and gamma-distributed site rates. Chains were considered converged when the average standard deviation of split frequencies reached <0.01, and the potential scale reduction factor between 1.00 and 1.02 after the first 25% of iterations were discarded as burn-in.

Reconciliation of gene trees with the molecular clock

To estimate the timing for duplication, transfer, loss, and speciation events, gene phylogenies were reconciled with our time-calibrated species trees using AleRax version 1.2.1 (131). AleRax was parameterized to consider genes that might be absent as a result of incomplete genome sequencing (via --fraction-missing-file FractionMissing.txt) and to forbid horizontal gene transfers between lineages that did not coexist in time (via --transfer-constraint RELDATED). Uncertainty in gene tree topology was also considered by inputting a sample gene trees instead of a single topology. This sample consisted of either 1000 ultrafast bootstrap resampled trees for genes whose evolutionary histories were reconstructed in IQ-TREE or 151 to 18,991 posterior trees sampled (every 500 generations) during MrBayes runs for different genes after discarding the first 25% as burn-in. The number of posterior trees sampled is a reflection of the number of gene tree topologies that needed to be sampled before the chains reached convergence.

Supplementary Materials

The PDF file includes:

Supplementary Text

Figs. S1 to S13

Tables S1 to S9

Legends for data files S1 and S2

References

Other Supplementary Material for this manuscript includes the following:

Data files S1 and S2