Early photosynthetic eukaryotes inhabited low-salinity habitats

Two datasets were analyzed: a genomic dataset including 135 highly conserved proteins (9) compiled from a total of 49 cyanobacterial genomes and a second dataset including 26 genes comprising 119 taxa that include both Cyanobacteria and photosynthetic eukaryotes. The first dataset was analyzed using Maximum Likelihood (ML) in a two-step process: (i) the 49 cyanobacterial genomes dataset was used to determine the deep-branching relationships of Cyanobacteria, and (ii) the topology generated in step i, here referred to as “genome constraint” (SI Appendix and SI Appendix, Fig. S1), was used as a backbone constraint for a second ML analysis that used 26 genes but included an extra set of 70 photosynthetic eukaryotes for a total of 119 taxa (as previously implemented) (9, 10, 27). The 119 taxa dataset was also analyzed using a Bayesian approach without the backbone genome constraint. Results from the two ML analyses and the Bayesian analysis returned results that are mostly congruent (SI Appendix and SI Appendix, Figs. S2 and S3). All analyses recovered well-supported monophyletic groups consistent with recent phylogenomic studies of Cyanobacteria (9, 10, 20, 28, 29). Trees were rooted using Gloeobacter, because this taxon has been previously shown to be the deepest branch within Cyanobacteria (9, 10, 29), even when including the group’s recently discovered nonphotosynthetic closest relatives: the Melainabacteria (30, 31), a group of heterotrophic soil and gut bacteria (32).

Previous studies have shown that the chloroplast lineage diverged among early branching Cyanobacteria lineages (20, 22, 29, 30, 33–36). All analyses strongly support the monophyly of plastids in photosynthetic eukaryotes, with the exclusion of the independently acquired Paulinella plastid (SI Appendix and SI Appendix, Figs. S2 and S3). All of our analyses also support Gloeomargarita as the most closely related cyanobacterium to the Archaeplastida (Figs. 1 and 2, SI Appendix, and SI Appendix, Figs. S2 and S3), consistent with recent phylogenomic studies (23). Gloeomargarita and the Archaeplastida diverged from early branching Cyanobacteria before the diversification of the Microcyanobacteria and Macrocyanobacteria (Fig. 1).

Our ML analyses show Cyanophora (Glaucocystophyta) as the sister of all of the other Archaeplastida (Fig. 1, SI Appendix, and SI Appendix, Fig. S2), with the Rhodophyta (red algae) and the Viridiplantae (green algae and land plants) as monophyletic sister groups. In contrast, our Bayesian analyses show Cyanophora diverging after the Rhodophyta and before the emergence of crown groups Chlorophyta and Streptophyta (Fig. 1, SI Appendix, and SI Appendix, Fig. S3). Simplified topologies in Fig. 2 illustrate the two alternative hypotheses with regard to deep-branching relationships within the Archaeplastida. In our analyses, long branches, such as Chlorokybus and Mesostigma, do not emerge as the monophyletic sister group of the Streptophyta (a division including several orders of nonmarine green algae and embryophytes), congruent with recent plastid-based phylogenies using a large sampling of cyanobacteria taxa (20) but contrary to other previous studies, in which these two lineages appear as basal and monophyletic within the Streptophyta (8, 37, 38). Furthermore, our ML analyses including only eukaryotes indicate that outgroup taxon selection does not influence the position of Chlorokybus and Mesostigma (SI Appendix and SI Appendix, Fig. S4). By including only Gloeomargarita, the position of Chlorokybus and Mesostigma is also independent of taxon selection, but the placement of Cyanophora changes (Fig. 2, SI Appendix, and SI Appendix, Fig. S5).

Divergence times were estimated by applying a Bayesian approach in Phylobayes 4.1 (39) to the two alternative tree topologies generated under ML (SI Appendix and SI Appendix, Fig. S2) and Bayesian analyses (details on how these trees were inferred are in Phylogenomic Analyses and Materials and Methods; SI Appendix and SI Appendix, Fig. S2). All analyses applied 18 calibration points across Cyanobacteria and photosynthetic eukaryotes (Table 1). Six calibrations were used for Cyanobacteria, and 12 were used for photosynthetic eukaryotes. Divergence times were estimated under three different relaxed molecular clock models: log normal (40), Cox–Ingersoll–Ross model (CIR) (41), and uncorrelated gamma multiplier (UGAM) (42). Results were consistent across all models and trees (Table 2, SI Appendix, and SI Appendix, Tables S3 and S4). Sensitivity experiments were performed to evaluate the effect of different root ages on our results (i.e., origin of oxygenic photosynthesis at 3 Bya) (SI Appendix and SI Appendix, Table S3). Overall, using an older maximum age root (3 Bya) pushes divergence times farther back but does not significantly change our conclusions.

All of our analyses indicate that the chloroplast lineage is the sister group of the recently discovered cyanobacterium Gloeomargarita (node 1) (Fig. 1) from which it diverged ∼2.1 Bya, with age estimates, including 95% credibility intervals, ranging from 2.3 to 1.9 Bya. Similar molecular divergence times have been previously reported based on slow-evolving small subunit (SSU) rRNA and rbcL (35) (Fig. 3). The chloroplast lineage diverged from its cyanobacterial relatives at a time when they exhibited only small cell diameters (10, 11) and likely inhabited low-salinity environments characteristic of fresh waters (11) (Fig. 4). The common ancestor of archaeplastids (node 2) (Fig. 1) likely evolved during the Paleoproterozoic Era—with the 95% credibility interval for this node ranging from 2.1 to 1.6 Bya; these results agree with other recent studies (8, 35, 43). Previous molecular clock analyses placed the first endosymbiotic event within the Late Paleoproterozoic Era (43), the Mesoproterozoic Era (37, 44), or even the Neoproterozoic Era (45); these studies, however, either did not include cyanobacterial lineages or if they did, relied on only a few cyanobacterial strains (Fig. 3 shows a comparison of previous molecular clock estimates). Bayesian stochastic character mapping analyses reveal that the ancestor of photosynthetic eukaryotes first evolved in low-salinity environments (Fig. 4) (this conclusion is robust to the phylogenetic placement of Chlorokybus and Mesostigma).

Although the common ancestor the Archaeplastida (node 2) evolved in the Paleoproterozoic Era, crown groups, such as the Rhodophyta (node 4), the Chlorophyta (node 5), and the Streptophyta (node 6), originated later, most likely between 1,200 and 900 Mya (Table 2). The Cyanophora lineage diverged from other photosynthetic eukaryotes (node 3) more than 1,500 Mya. The 95% credibility interval for the ancestors of the Rhodophyta (node 4) ranges between 1.3 and 0.9 Bya; these estimates overlap with recent studies including a large number of taxa, encompassing nuclear and plastid genes, from the Rhodophyta (43) and are younger than estimates from studies including nuclear genes estimating their origin during the Early Mesoproterozoic (8). Well-supported marine lineages within the Rhodophyta can be paleontologically traced to marginal marine habitats at ∼1.1 Bya (Figs. 1 and 3, Table 2, SI Appendix, and SI Appendix, Tables S3 and S4). In this study, the age divergences of the Chlorophyta (node 5) and the Streptophyta (node 6), excluding Chlorokybus and Mesostigma, are similar, with 95% credibility intervals ranging between 1.3 and 0.9 Bya and between 1.1 and 0.67 Bya, respectively; these estimates are younger than previous studies placing the origin of these two major groups in the Early Mesoproterozoic Era (8) but overlap with previous studies that propose Neoproterozoic origins (37, 44). Chlorophyte fossils in ∼800-My-old rocks are consistent with these estimates (18).

Palmophyllophyceae (node 8) is the deepest branch within the Chlorophyta, which is consistent with recent studies; this lineage contains species found in a wide range of marine habitats, including planktonic unicells and loose colonies, as well as benthic macroscopic thalli. The 95% credibility interval for the origin of the Palmophyllophyceae (node 8) ranges between 1.0 and 0.45 Bya. Furthermore, another early branching group of chlorophytes, labeled “prasinophytes” in Fig. 1, consists mainly of marine phytoplankton. Although the species included in this study are monophyletic (Fig. 1), phylogenomic studies including a broader taxonomic sampling of the Chlorophyta show the traditionally recognized Prasinophyceae to be paraphyletic (46). The clade included in our analyses is both ecologically important and paleontologically documented; again, our preferred divergence age of ∼750 Mya for this group (node 8; 1.0–0.49 Bya, including the 95% credibility interval) is consistent with the fossil record (14). Our analyses indicate that relatives of the Palmophyllophyceae and the prasinophytes would have been some of the first Chlorophyta (green algae) to radiate into marine habitats in Neoproterozoic oceans (Fig. 4). The lineage now represented by the plastid of Paulinella diverged as well from marine SynPro clades during the Neoproterozoic Era.

The lack of a terrestrial fossil record has contributed to our lack of understanding about life in early nonmarine habitats. However, novel hypotheses about life in early terrestrial environments have emerged based on phylogenomic and trait evolution analyses. Recently sequenced cyanobacterial genomes have improved the resolution of phylogenomic analyses and the deep-branching relationships within Cyanobacteria (11, 20, 29, 47). Our phylogenomic analyses, including a wide range of plastids from photosynthetic eukaryotes, are in agreement with some of the most recent large-scale multigene phylogenies (11, 20, 22, 29). All of our analyses agree with previous studies showing that the Cyanobacteria lineage that led to the chloroplast diverged early in cyanobacterial history (Fig. 1) (20, 22, 30, 33, 48).

Molecular phylogenetics, trait evolution, and relaxed molecular clock analyses have recently helped with the interpretation of the early fossil (9, 10, 47) and geochemical records (27). Broad taxonomic studies of prokaryotes imply a terrestrial/freshwater ancestry for Cyanobacteria (49, 50) and their early diverging lineages (9, 11, 51). The plastid lineage, at its time of divergence, likely exhibited small cell diameters (1–2 µm), because the radiation leading to the Macrocyanobacteria [a group of cyanobacteria exhibiting cell diameter >3 µm and observable in the fossil record by 1.9 Bya (52) (Fig. 1)] had yet to occur (11). Our Bayesian stochastic mapping analyses provide strong support for the hypothesis that the common ancestor of Cyanobacteria and the plastid lineage (node 1) (Fig. 1) inhabited low-salinity environments (Fig. 3). Interestingly, most deeply diverging cyanobacterial lineages use sucrose to cope with changing environmental conditions (e.g., mild changes in osmotic pressure, desiccation tolerance) (53, 54). A similar mechanism was likely inherited by the plastid lineage at the time of the first endosymbiotic event (Fig. 1). Compatible solute genes (i.e., glucosylglycerol, glucosylglycerate, and glycine-betaine) involved in regulating osmotic stress in high-salinity environments (53, 54) evolved in marine lineages (11, 55) after the divergence of the chloroplast lineage.

All of our analyses strongly support the monophyly of plastids (except for that of Paulinella) in photosynthetic eukaryotes (SI Appendix and SI Appendix, Figs. S2 and S3), consistent with previous studies reporting a single endosymbiotic event in the ancestor of the Archaeplastida (34, 37, 44, 56–58). Our analyses provide two alternative hypotheses for deep-branching relationships among the Archaeplastida (Fig. 2). Like most previous studies, our ML analyses suggest that Cyanophora (Glaucocystophyta) is the sister of all other plastids (SI Appendix and SI Appendix, Fig. S2) followed by the divergence of the Rhodophyta and the Viridiplantae (green algae and land plants). In contrast, our Bayesian analyses show Cyanophora branching after the Rhodophyta (SI Appendix and SI Appendix, Fig. S3). It is worth highlighting that, although most previous phylogenetic studies place Cyanophora as the deepest branch with the Archaeplastida (20, 23, 37, 46), other studies have postulated alternative phylogenetic placements for this lineage (8, 35, 36).

Our ML hypothesis shown in Fig. 2 is consistent with ancestral traits (e.g., peptidoglycan) being retained by the Glaucophytes and later lost by all other algae and land plants (37). Furthermore, this topology supports an evolutionary scenario in which phycobilisomes, the light-harvesting antennae anchored in the thylakoid, are an ancestral trait present in Cyanobacteria that has been retained in the Glaucophytes and red algae but lost in green algae and land plants. Given this hypothesis, another likely ancestral feature is the turgor-resistant peptidoglycan wall around the “cyanelle” in the Glaucocystophytes (59). This feature helps minimize energy cost for active water efflux involved in volume regulation by creating low osmolarity in the cytosol (60) and higher osmolarity in the cyanelle related to the higher concentrations of HCO₃⁻ required for functioning carboxysomes in an inorganic carbon-concentrating mechanism (CCM). Although the glaucocystophyte Cyanophora paradoxa has a CCM (61), its complete genome sequence shows no carboxysome-related genes (62). This is consistent with the hypothesis proposed by Raven (59) for the function of the peptidoglycan wall of the cyanelle, requiring the ribulose-1.5-bisphosphate carboxylase-oxygenase (Rubisco) containing a central body of the cyanelle with high HCO₃⁻ concentrations for its CCM function, despite being a pyrenoid rather than a carboxysome. However, the Bayesian archaeoplastid phylogeny in Fig. 2 shows glaucocystophytes branching after the rhodophytes, and therefore, if the implicit assumption is correct (57) that the glaucocystophyte CCM is ancestral within the archaeplastids, this trait must have been lost independently in ancestral red and green algae. Another problem with the ancestral CCM suggestion (57) is the absence of any molecular genetic markers for the glaucocystophyte “pyrenoid” (63), knowledge of the enzymes kinetics of their glaucocystophyte Form IB Rubisco, or uncertainties of CO₂ and O₂ availability at the time of chloroplast origin (64). In any event, our inferences about the environmental setting of archaeplastid origins are robust to the placement of Cyanophora.

Within the Viridiplantae, our analyses show that Chlorokybus and Mesostigma branched before the Streptophyta and the Chlorophyta as previously found by Shih et al. (20); however, other studies based on plastid genes show relatively low support for this node (8, 36). For our plastid dataset, it seems that the position of Chlorokybus and Mesostigma is independent of outgroup taxon selection (SI Appendix and SI Appendix, Fig. S4). In contrast, nuclear gene-based phylogenies provide high statistical support for the inclusion of Chlorokybus and Mesostigma within the Streptophyta (38, 65). Chlorokybus and Mesostigma are long-branched taxa and therefore, likely prone to long-branch attraction, and this might explain incongruence in their branching relationships. More genomic studies are needed to fully resolve the deep-branching relationships of the charophyte green algae (66).

It is worth emphasizing the distinction between the moment in which the plastid ancestor diverged from other lineages of Cyanobacteria (node 1) and the last common ancestor of extant plastids (node 2). Our Bayesian relaxed molecular clock analyses suggest that the plastid lineage diverged from other Cyanobacteria as early as ∼2.1 Bya (node 1) (Fig. 1 and Table 2); all analyses, including clock models, converge to similar age estimates. This date is broadly similar to previous studies, including values in the order of ∼2.4–2.7 Bya based on conserved encoded proteins (8) and ∼2.1 or 2.4 Bya based on SSU RNA data (35) (Fig. 2). In contrast, the median age for the primary endosymbiotic event that led to the origin of the archaeplastid chloroplast is in the order of ∼1.9 Bya (node 2) (Figs. 1 and 2). Assuming that no plastid-bearing archaeplastid lineages have gone extinct (i.e., that the last common archaeplastid ancestor was also the first photosynthetic archaeplastid), our analyses suggest a time interval of ∼0.2 By between the split of the plastid lineage from Gloeomargarita (node 1) and the origin of the first photosynthetic eukaryote (node 2). Alternatively, photosynthesis may first have become established in now extinct stem group archaeplastids.

Molecular clock studies have generated a large range of age estimates, spanning over a billion years, for the origin of the Archaeplastida (Fig. 3). Our analyses, which include six cyanobacteria calibration points, suggest that photosynthetic eukaryotes originated relatively early, perhaps ∼1.9 Bya (with the 95% credibility interval for this node ranging from 2.1 to 1.6 Bya). Molecular clock studies vary widely in terms of types of data (e.g., nuclear or plastid genes, rRNA, proteins) and taxa selection. Some estimates based on protein data suggest divergence at ∼0.9 Bya (56); others based on the α and β subunits of ATP synthase and the elongation factor thermo unstable (45, 67) propose ∼1.1 Bya, and by implementing the SSU rRNA and rbcL, still others predict ∼1.3 or 1.2 Bya (35) (Fig. 2). Some of the oldest ages are in the order of ∼2.1 Bya using a set of conserved proteins and RNA genes (8). It is difficult, if not impossible, to compare molecular clock studies, because discrepancies might be caused by taxon sampling, types of substitution and clock models used, fossil calibrations, and accuracy in the implementation of molecular clock methodologies. However, extensive taxon sampling is a significant determinant of accurate phylogenetic estimation (68); here, by including Gloeomargarita, we have more accurately constrained the age of the Archaeplastida. In general, studies that include cyanobacterial relatives provide older ages for photosynthetic eukaryotes (Fig. 2) (8, 35).

Within photosynthetic eukaryotes, there is a deep divergence between the Viridiplantae and the Rhodophyta (node 3), and this split has been estimated to have occurred ∼1.9 Bya. Other estimates for this node have been on the order of ∼1.4 Bya based on DNA and 1.67 Bya for protein (58). There is a significant lag between the common ancestor of the Archaeplastida (node 2) and the origin of major clades of photosynthetic eukaryotes, such as the Rhodophyta (node 4), the Chlorophyta (node 5), and the Streptophyta (node 6; excluding Chlorokybus and Mesostigma). Although basal lineages of extant red algae are nonmarine, both fossils and our molecular clocks indicate that reds had colonized marine environments by ∼1.1 Bya (Fig. 1). Moreover, previous studies have further tested that inclusion and exclusion of Bangiomorpha as a calibration point have little effect on age estimates for the rhodophytes (43, 44). Our molecular clock analyses and fossils further agree that chlorophyte green algae colonized the marine realm during the Neoproterozoic Era, with several lineages subsequently regaining terrestrial and freshwater aquatic environments. More specifically, fossils and molecular clock analyses indicate that prasinophytes emerged as ecologically important constituents of the marine phytoplankton late in the Neoproterozoic Era.

Broadly, then, our study suggests that, although the plastid lineage diverged from other Cyanobacteria during the Paleoproterozoic Era, it took nearly a billion years for major extant clades to diversify and gain ecological prominence. The younger divergence age of Paulinella ∼0.5 Bya from its closest relatives is consistent with suggestions that its significantly greater genome retention indicates a much earlier stage of endosymbiosis (69).

Recent years have seen significant improvements in our understanding of eukaryotic origins. It is now evident that eukaryotes do not constitute a third primary lineage of life, at least not as originally proposed by Woese and Fox (70). Instead, phylogenomic analyses minimally agree that the eukaryotic lineage evolved from within the Archaea (71–74). Furthermore, it is uncontroversial that the establishment of the crown eukaryotes involved a stable endosymbiosis between an archaeon [related to the recently discovered Lokiarchaeota (75)—the cellular host] and an alpha-proteobacterium [the ancestor of the mitochondrion (72–74, 76)]. A topic of current debate in eukaryotic evolution is whether an amitochondriate protoeukaryotic lineage ever existed. Bioenergetic arguments have been marshalled to disfavor this view (77), suggesting that only a mitochondriate cell can acquire the complexity observed in living eukaryotes, but these arguments are not universally accepted (78, 79). A recent genomic study (80) suggested that the mitochondrion might have entered the protoeukaryotic lineage late, when the stem eukaryotic cell was already cytologically rather complex (80). However, these results have been shown to represent a methodological artifact (81). Irrespective of how eukaryote-like the stem eukaryotes were, all crown eukaryotes are mitochondriate, and therefore, their last common ancestor must also have contained mitochondria (82). It is not yet clear when the protomitochondrial endosymbiosis was established (76). However, because the free-living ancestor of the mitochondrion was an alpha-proteobacterium, at the very least, this event must have postdated the separation of the alpha-Proteobacteria from their sister lineage (the group including the beta- and gamma-Proteobacteria and possibly other less well-known lineages, like the Acidithiobacillia and Zetaproteobacteria) (72). Given the capacity of mitochondria for aerobic respiration, this might well have happened after the Great Oxygenation Event.

Our analyses help to constrain the origin of eukaryotes in so far as they suggest that the endosymbiotic event between a crown eukaryote and a cyanobacterium resulted in the origin of the Archaeplastida by ∼1.9 Bya. This is consistent with currently available fossil evidence and the view that eukaryotes, minimally, must postdate the diversification of Archaea and the node separating the alpha-Proteobacteria from their sister lineage. How much older the base of the eukaryotes might be from the base of the Archaeplastida is unknown and will, in part, depend on the topology of the eukaryotic tree, which is still not fully resolved (83).

Little is known about early life on land. Nonmarine environments are generally sites of erosion rather than deposition and therefore, relatively uncommon in all but the youngest sedimentary basins. Moreover, nonmarine sedimentary rocks are not easily differentiated in the absence of environmentally diagnostic plant or animal fossils, and those facies most reliably recognized as nonmarine on the basis of physical features (alluvial fans, braided stream deposits) tend to be poorly fossiliferous (84). Lacustrine and floodplain environments are known from Archean successions, and stromatolites and ¹²C-depleted organic matter in these rocks document early microbial ecosystems driven by photosynthesis (85). Significantly, shales associated with alluvial and fluviatile sandstones in the Late Mesoproterozoic successions from Scotland and midcontinent North America (86, 87) contain abundant and modestly diverse microfossils reasonably interpreted as eukaryotic (88). The taxonomic affinities of these fossils are unclear, and photosynthetic microorganisms may or may not be present in the assemblage. Also, at least for the Scotland example, the nonmarine origin of fossiliferous shales remains to be shown. Nonetheless, it is reasonable to conclude that eukaryotic organisms had gained a firm foothold in nonmarine environments by 1,000 Mya. Thus, the meager fossil record of nonmarine environments is consistent with phylogenomic inferences without adding independent evidence for nonmarine algal origins. Paleoenvironmental studies may help more, suggesting as they do that eukaryotes in the terrestrial realm may have been shielded, at least in part, from the persistent subsurface anoxia (especially in shallow lakes and streams) (89), episodic euxinia, and trace metal limitation (90) that potentially limited eukaryotic diversification in Mesoproterozoic oceans. Our analyses are consistent with paleontological data suggesting that, although archaeplastids originated in nonmarine environments, both red (node 6) and green (node 7) algae radiated in the oceans during the Late Mesoproterozoic and Early Neoproterozoic intervals, respectively (Fig. 1).

All sequence data were obtained from GenBank (www.ncbi.nlm.nih.gov). Sequence data come from 49 cyanobacterial genomes as previously described in the works by Blank and Sánchez-Baracaldo (9) and Sánchez-Baracaldo (11). To establish the deep-branching relationships of Cyanobacteria, our dataset comprised 136 proteins; molecular markers included are evolutionarily conserved, had a minimum number of gene duplications, and were present in all cyanobacterial taxa (10, 91). A second alignment with a broader taxonomic sampling of photosynthetic eukaryotes (70 taxa) and cyanobacteria (49 taxa) was used to infer the evolutionary relationships of Cyanobacteria and photosynthetic eukaryotes; this dataset included 26 genes. SI Appendix, Table S1 contains a list of genes shared across all taxonomic groups included in this study. Each gene was aligned independently using SATé 2.2.3 (92), a multiple sequence alignment and phylogenetic reconstruction program. Single-gene alignments generated in SATé were imported into Mesquite v. 2.75 (93) to obtain “nexus” and “phylip” format files for subsequent analyses. Single alignments were later concatenated into a single-nexus format file using Sequence Matrix v 100.0 (94). With the concatenated matrices generated here, we performed two sets of analyses describe below.

ML and Bayesian analyses were implemented to determine the phylogenetic relationships between Cyanobacteria and photosynthetic eukaryotes. For the first set of analyses, we performed ML analyses in two stages. We generated genome analyses from 49 cyanobacteria genomes to determine the deep-branching relationships of Cyanobacteria in RAxML GUI v.1.1 (95). The topology generated at this stage was later implemented as genome constraint in RAxML GUI v.1.1 (95) for the second dataset assembled including 26 genes (protein sequences) and a total of 70 photosynthetic eukaryotes. A total of 136 protein-coding genes (with 50,187 amino acids) were used to establish phylogenetic relationships of Cyanobacteria. ProTest v.2.4 (96) was used to estimate the best model of evolution for the protein set. To analyze the protein sequences, we implemented the LG +G model (gamma-distribution with four rate categories).

A second dataset was analyzed under the CAT-GTR+G model in Phylobayes MPI 1.7a (97). This dataset consisted of 26 genes shared across 49 cyanobacteria and 70 photosynthetic eukaryotes. Convergence of Bayesian analyses was tested using the software Tracecomp (in Phylobayes) to estimate effective sample sizes and relative differences of key parameters and the software Bpcomp to calculate mean distance between trees from independent chains. To test whether taxon selection of the cyanobacteria outgroups had any effect on the branching relationships of the Archaeplastida, we performed ML analyses including only eukaryotes and the closest known cyanobacteria (Gloeomargarita).

We used 18 fossil calibrations across all Cyanobacteria and photosynthetic eukaryotes (Table 1). Cyanobacteria arguably have the best fossil record of any bacterial group (47). In this study, we applied cyanobacterial fossils that have been previously implemented (9, 11, 33). For the origin of Cyanobacteria, two maximum ages were implemented: 2.7 (98) and 3 By (99). The rise in atmospheric oxygen at 2.32 By was set as the minimum age for the cyanobacterial root (2). Simple filamentous fossils of cyanobacteria, comparable to Oscillatoria, occur at 1.9 Byr (52); placement of this occurrence was implemented based on previous phylogenomic and trait evolution (i.e., filamentous vs. unicellular) studies including Pseudanabaena genomes (11, 29). Furthermore, 1.9-Bya cherts contain large unicells and filaments (52), consistent with other evolutionary studies of cyanobacteria showing that the Macrocyanobacteria, sensu Sánchez-Baracaldo (11), had diverged by this time. Akinetes (100) and multiple fission (101) were be assigned to distinct groups, such as the Nostocales and the Pleurocapsales, respectively. Larger phylogenomic studies of Cyanobacteria including two Pleurocapsa genomes (PCC 7319 and PCC 7327) have established that, although these two strains do not form a monophyletic group, they belong to a well-supported monophyletic group (11); this broader clade was calibrated here using preserved microfossils. Fossil calibrations across photosynthetic eukaryotes (Table 1) are relatively well-characterized, adding an important additional constraint to molecular clocks of cyanobacterial evolution.

Ages were estimated for the topologies generated by the ML and Bayesian analyses as described above using a subset of genes from the 26 genes that encode the following proteins: AtpA, AtpB, PetB, PsaC, PsbA, PsbD, RbcL, and S12. Divergence times were estimated implementing three different relaxed molecular clock models: log normal (40), CIR (41), and UGAM (39) in Phylobayes 4.1 (39). As in the Bayesian analyses, substitutions were modeled using the CAT-GTR+G replacement model. For all noncalibrated nodes, we used a birth–death prior on divergence times and soft-bounded calibrations, allowing 0.05% of the prior density to fall outside the minimum–maximum interval of each calibration. Analyses were performed using two different root priors. The first was a gamma-distributed prior, allowing the 95% of the prior distribution to fall between 2.32 and 2.7 Bya, and the second had the 95% of the prior distribution falling between 2.32 and 3 Bya. The second analysis was performed as a sensitivity test (SI Appendix and SI Appendix, Table S4). Convergence of molecular clock analyses was tested using the software Tracecomp. All analyses were also run without data to visualize the marginal priors on node ages implied by our set of fossil calibrations.

To infer the evolution of habitat, we used Bayesian stochastic character mapping (102). Character states were obtained from Bergey’s Manual of Systematic Bacteriology (103), AlgaeBase (www.algaebase.org), taxa description of genome taxa from the Joint Genome Institute (jgi.doe.gov), previous studies of trait evolution of Cyanobacteria (11), and other Cyanobacteria studies. Habitat was coded using multistate discrete character states as follows: 0 = freshwater [0–0.5 parts per thousand (ppt)], 1 = marine (from 30 up to 50 ppt), and 2 = brackish (0.5–30 ppt). SI Appendix, Table S2 contains the characters states used in this study. Analyses were implemented in SIMMAP v1.5 (104) and used time-calibrated trees generated in Phylobayes 4 (39) for both ML and Bayesian analyses (Fig. 4). Sensitivity tests were performed treating Parachlorella and Pedinomonas as marine and/or freshwater within the chlorophyta (SI Appendix and SI Appendix, Figs. S6 and S7). Branch lengths were used as a direct estimate of rate of evolution rather than a prior on the rate parameter. For multistate characters, the bias parameter between states is specified as simply 1/κ, where κ is the number of states. The overall rate of substitution for both of these classes is a branch length multiplier drawn from a prior gamma distribution. Analyses performed in SIMMAP v1.5 (104) were summarized in R using phylotools 0.1.2 (105).

Data associated with this paper are available to download from the Dryad Digital Repository at https://doi.org/10.5061/dryad.421p2. The uploaded data include aligned protein sequences for phylogenetic and molecular clock estimates, calibration files, and Phylobayes scripts.