Living organisms can be studied in immensely more detail than can fossils. Even without fossils, comparative phylogenetic studies now provide remarkably detailed and accurate reconstructions of evolutionary relationships across the whole tree of life and rigorously inferred ancestral states for most of it. However, reconstructing the history of life requires integrating insights from palaeontology and neontology: comparative study of extant organisms. Without palaeontological evidence the tree lacks a time-scale; inferences would lack the important control of the temporally layered evidence in the rocks; and we would lack direct knowledge of entirely extinct groups. Temporal evidence from palaeontology also helps place the root on the universal tree, and valuably checks inferences about the temporal direction of changes in many parts of it. Moreover, because we all sometimes make interpretative mistakes, each discipline offers the other a valuable external perspective to help correct them.

Molecular and cell biological evidence on the tree of life is much more congruent with the fossil record than often supposed; most apparent conflict arises unnecessarily from interpretative errors by neontologists and palaeontologists (Cavalier-Smith 2002a,b). In particular, using steranes as a biomarker for eukaryotes is problematic, as several phylogenetically diverse bacteria can make sterols, including some actinobacteria, which I consider ancestral to both eukaryotes and archaebacteria (i.e. to all neomura); I recently argued that no pre-Neoproterozoic fossils are confidently assignable to any eukaryote phylum or even indubitably eukaryotic (Cavalier-Smith 2002a). Understanding early eukaryote evolution has greatly improved since then, especially evidence for the root of the tree (Stechmann & Cavalier-Smith 2002, 2003; Richards & Cavalier-Smith 2005) and for the new supergroup Rhizaria, which includes abundant fossil foraminifera and radiolaria (Nikolaev et al. 2004), while molecular cell biology has advanced so greatly that the origin of eukaryotes is more specifically explicable than ever before. Cavalier-Smith (2006) placed the topology of the bacterial part of the universal tree of life on firmer grounds and explained how several transitions within it can be confidently polarized by strong evolutionary arguments. I concluded that the root of the tree of life is within negibacteria: Gram-negative eubacteria possessing two envelope membranes. On that interpretation the single surface membrane in posibacteria, archaebacteria and eukaryotes is a derived evolutionary condition that arose by losing the OM (figure 1). That the first cells were double-enveloped negibacteria and that their OM was lost only once in the history of life are not new ideas (Blobel 1980; Cavalier-Smith 1987a,b, 1991a, 2001, 2002a, 2004a), but is increasingly favoured by these new polarizations of major evolutionary transitions and new evidence for monophyly of Posibacteria.

Figure 1 The tree of life emphasizing major revolutions in cell structure. The most basic distinction in cell biology is between negibacteria, with a cell envelope of two distinct lipid bilayer membranes, and unimembrana, with but one surface membrane. Contrary (more ...)

In this paper I attempt to map the tree of life onto the fossil record, arguing that they are broadly congruent, and sketch a synthesis. I discuss evidence for a three-phase history of life.

Before discussing the tree, I outline three major dichotomies in cell organization among bacteria, each almost as important as the distinction between bacteria and eukaryotes for understanding the major steps in cell evolution, but less generally appreciated. I use ‘bacteria’ as in classical terminology for all prokaryotes; it is misleading and confusing to restrict ‘bacteria’ to eubacteria alone, excluding archaebacteria from the concept.

I shall argue that endogenous biological limitations to megaevolutionary innovations in cell structure probably dominated timing of the most fundamentally innovative changes in the biosphere. The immense difficulty of the neomuran revolution and eukaryogenesis accounts for the late origin of plants (by cyanobacterial enslavement), animals (by evolving epithelia and connective tissue), fungi (by evolving tip-growing chitin walls) and chromalveolates (by secondary enslavement of a red alga), and soon thereafter their Cambrian explosive radiations. The relatively early origin of cyanobacteria and proteobacteria, and far later origin of methanogenic archaebacteria, was probably responsible, respectively, for the Palaeoproterozoic and Neoproterozoic snowball Earth episodes by altering greenhouse gas balances.

Figure 2 Fundamental contrasts and continuities between eubacteria and neomura in surface chemistry and protein secretion. The older eubacteria (below) typically have walls of murein peptidoglycan and lipoprotein. Neomura have a surface coat (eukaryotes) or wall (more ...)

Despite the notable differences between eubacteria and neomura, most cell properties are not nearly as sharply distinct (Cavalier-Smith 2002a); both groups share over a thousand fundamentally similar characters, ruling out separate origin, sometimes postulated (Koga et al. 1998; Woese 1998; Martin & Russell 2003)—their common ancestor was a complex, highly developed cell (Cavalier-Smith 2002b; Peretó et al. 2004), definitely a eubacterium (Cavalier-Smith 2006a). Two key cellular innovations only made the ancestral archaebacterium: novel acid-stable flagella and novel acid- and heat-stable isoprenoid ether lipids, both secondary adaptations to hyperthermophily (Cavalier-Smith 2002a). Their unique flagella evolved from type IV secretion proteins previously used to make eubacterial pili (Cavalier-Smith 2006a). As eubacteria make isoprenoids, the archaebacterial novelty was to add them to glycerol phosphate with new stereochemistry to make isoprenoid-ether glycerophosphate lipids; the enzyme doing this is related to one found in posibacteria (only), and thus involved no major innovation (Peretó et al. 2004); some archaebacteria even retained the key enzyme making acyl ester lipids. Neither change involved really novel proteins.

Classification of proteins into fold superfamilies reflects their fundamental distinctiveness. Only one novel protein fold superfamily is universal among archaebacteria: that of the enzyme making their unique tRNA base, archaeosine; even this is related to five other fold superfamilies (Yang et al. 2005). By contrast 21 novel fold superfamilies arose during the neomuran revolution (two universal in neomura, i.e. never secondarily lost), whereas eukaryotes have 153 fold superfamilies (14 universal) never in bacteria (Yang et al. 2005). The degree of novel protein evolution during eukaryote origins (eukaryogenesis) was far greater than for archaebacterial origins. Treating archaebacteria as a separate kingdom or superkingdom is unjustified. Their differences from other bacteria are greatly exaggerated by quantitative differences in rRNA, ribosomal proteins and DNA-handling enzymes. The perception of archaebacteria as a third form of life, based on rRNA quantitative differences, has been profoundly misleading; in most characters they are no more distinctive than any other eubacterial phylum and they are younger, despite their name (Cavalier-Smith 2002a). Their two subphyla (Euryarchaeota, Crenarchaeota) were both probably ancestrally anaerobic respirers energetically dependent on sulphur compounds, a much more ancient physiology in eubacteria. By combining this with hyperthermophily, archaebacteria expanded the biosphere in a relatively minor way into previously inaccessible hot/acid environments. No unique protein fold superfamilies are found in all eubacteria but absent from neomura (Yang et al. 2005), as expected if eubacteria are paraphyletic (i.e. ancestral to at least one other group) and the only primary domain of life, and neomura derived. The major richness of the 1244 protein fold superfamilies probably arose in stem bacteria before the cenancestor; the greatest bout of protein innovation since then was during eukaryogenesis, reinforcing the classical distinction between eukaryotes and prokaryotes.

Figure 3 Membrane evolution and the fundamental contrast between negibacteria and posibacteria. Earliest cells were negibacteria bounded by two lipoprotein membranes: an inner cytoplasmic membrane (CM) impermeable to small hydrophilic molecules (with alpha-helical (more ...)

The lipid tails of posibacterial lipoproteins are in the CM outer leaflet; their protein part is covalently attached to the thick murein wall by sortase enzymes that recognize cell wall sorting signals near the C-terminus of its precursor, after its export to the periplasmic space by SRP (Comfort & Clubb 2004). Prior to this, sortases cleave the hydrophobic C-terminal sequence that temporarily anchors it to the CM. Hundreds of cell wall proteins, mostly not lipoproteins, are thus anchored to the thick murein of posibacteria before the N-terminal signal sequence is cleaved. Most posibacteria have sortases of 2–4 subfamilies with different substrate specificities (Comfort & Clubb 2004). Sortases are a synapomorphy for posibacteria (comprising subphyla Endobacteria and Actinobacteria), strongly supporting their monophyly (Cavalier-Smith 2006a). They probably evolved to attach the wall firmly to the CM and prevent soluble periplasmic proteins escaping when their cenancestor arose by losing the negibacterial OM.

Endobacteria include ancestrally endospore-forming Gram-positive bacteria, e.g. Bacillus, Clostridium, Staphylococcus, and the parasitic Mollicutes (mycoplasmas, spiroplasmas) that secondarily lost murein, lipoproteins and the sortase machinery, and miniaturized their genomes. Endobacteria also exclude the ancestrally endospore-producing Eurybacteria (e.g. Heliobacteria, Sporomusa), which group with them on most trees (Woese et al. 1985), and so are often also misleadingly called low-GC Gram-positives. Eurybacteria are neither Gram-positive nor posibacteria, but negibacteria with OM and endospores; as discussed below, they are probably ancestral to posibacteria (Cavalier-Smith 2006a). Thermotogales are also Eurybacteria, despite frequent, probably artefactual, grouping on sequence trees with neomura and/or Aquificales (Cavalier-Smith 2006a).

Cladistic analyses of morphological or macromolecular characters sufficiently complex to rule out convergence are often the most reliable way of establishing relationships, though can be confused by secondary evolutionary losses. Cladistic analysis of genetic characters (e.g. shared lateral gene transfers, gene replacements, gene fusions) is also important. Best phylogenetic practice seeks congruence between all types of evidence. One should try to understand reasons for any conflicts and biases and give more weight to the strongest evidence and most reliable methods.

Two more reliable approaches can polarize trees, and thereby root them: transition analysis and the fossil record. Transition analysis of sufficiently complex characters can often polarize evolutionary change, as recently explained for the origin of neomura (Cavalier-Smith 2002a) and numerous transitions within bacteria (Cavalier-Smith 2006a). Section 3c uses the origin of posibacteria to illustrate how transition analysis can polarize a major evolutionary step. Transition analysis can establish ancestor–descendant relationships. It gives a relative time-scale to biological evolution, analogous to that in geology yielded by the principle that younger sedimentary strata generally overlie older ones. But it cannot give absolute times to evolution as radiometric clocks can for igneous rocks. Nor does sequence evolution give an absolute time-scale. The idea of a molecular clock for sequences is misleadingly bad terminology. Rates of nucleotide substitution can vary million-fold among molecules when functions change dramatically (as often following gene duplication), and can change thousands-fold for a single molecule with basically unchanged function (Cavalier-Smith 2002b). Sometimes, for short evolutionary periods and restricted lineages, rates of change may be steadier and useful for rough interpolation between dates established by fossils. But fossil dates must be paramount. Extrapolating beyond them is always risky and sometimes grotesquely wrong.

Figure 4 Flagellar origin unambiguously polarizes evolution from negibacteria to posibacteria. The TonB importer and its force-providing proton channel ExbB are homologues of the motor proteins MotB and MotA of eubacterial flagella and probably their evolutionary (more ...)

A second argument polarizing the transition from negibacteria to posibacteria is the mechanistic difficulty of adding an OM in one step to a flagellate posibacterium; nobody has explained how this could be done while plausibly allowing OM biogenesis. The difficulty of evolution is that hypothetical direction is compounded by the closest negibacterial relatives of posibacteria being Eurybacteria, which are glycobacteria with a much more complex OM than eobacteria, which are arguably more primitive (Cavalier-Smith 2006a). If evolution were in that direction, the OM must immediately have evolved hugely complex LPS and other complexities shared by glycobacteria, e.g. hopanoids, porins, TolC. If instead the first negibacteria were eobacteria, especially chlorobacteria with much the simplest OM, their origin is much simpler; I explained how the OM could have evolved simply and gradually to generate the first eobacterium (Cavalier-Smith 2001).

I argued that Omp85 was never lost by any organism that retained the OM; its apparent absence from chlorobacteria, alone among negibacteria, means that chlorobacteria are the most primitive negibacteria, originating before the Omp85-mediated mechanism for inserting OM β-barrel proteins (Cavalier-Smith 2006a). Thus, chlorobacteria are the most ancient negibacteria. Combining this argument with the polarization from negibacteria to posibacteria, chlorobacteria must also be the first eubacteria. Chlorobacteria are also simpler than Hadobacteria and glycobacteria in lacking 10 other widespread bacterial features (Cavalier-Smith 2006a).

Ribosomal RNA trees and all well-resolved and taxonomically well-sampled sequence trees for single proteins, and for combined data from many proteins show eukaryotes as sisters to archaebacteria, not nested within archaebacteria (Clarke et al. 2002). Holophyly of archaebacteria was strongly favoured by their unique isoprenoid ether lipids (Cavalier-Smith 1987b). If the common ancestor of archaebacteria and eukaryotes had such lipids, but no acyl ester lipids (present throughout eubacteria), the ancestral eukaryote must have reacquired acyl ester lipids and totally lost isoprenoid ethers, a very non-parsimonious assumption. As the eukaryote cenancestor already had a mitochondrion (Cavalier-Smith 2002b, 2003b, 2004b, 2006b; Stechmann & Cavalier Smith 2003; Embley 2006), a protoeukaryote with only isoprenoid ether lipids could theoretically have replaced them entirely by acyl ester lipids from the enslaved proteobacterium (Martin & Müller 1998), but this is mechanistically and evolutionarily extremely improbable. A further flaw in that improbable hypothesis is that proteobacteria lack phosphatidylinositol, one of the most important eukaryote phospholipids, required for eukaryote-specific cell signalling. The only eubacteria having phosphatidylinositol are Actinobacteria, which are ancestral or sister to neomura according to much other evidence (Cavalier-Smith 2002a, 2006a). Thus, eukaryote membrane lipids probably came vertically from an actinobacterial ancestor, archaebacterial lipids originating in their cenancestor after it diverged from eukaryotes (Cavalier-Smith 1987b). Archaebacteria, eukaryotes and neomura are all holophyletic.

Three proteins found only in archaebacteria, plus archaebacterial flagella, are synapomorphies for archaebacteria and evidence for their holophyly (Cavalier-Smith 2002a). Splitting the genes for RNA polymerase A (=β′ in eubacteria) and glutamate synthetase (GS) to code two and three proteins, respectively, is also a shared derived character for archaebacteria (Nesbo et al. 2001; Cavalier-Smith 2002a; two halophilic archaebacteria, Natronomonas and Haloarcula reacquired an unsplit GS gene by lateral transfer from a eubacterium; their GS is far more similar to that of the halophilic Salinibacter ruber (57% identity) than to archaebacterial ones—my Blast results). My database Blast analysis reveals no exceptions yet to the rule that neomuran polymerase A is split into separate A′ and A″ proteins in archaebacteria (Bartlett et al. 2004); this would have to have been reversed by gene fusion if, contrary to my arguments, archaebacteria were ancestral to eukaryotes, as no eubacterium could have provided this characteristically neomuran molecule to the cenancestral eukaryote by lateral gene transfer (LGT; nor could the mitochondrial ancestor). Another argument for archaebacterial holophyly comes jointly from MreB and the largest subunit of RNA polymerase. MreB, the probable eubacterial ancestor of actin and actin-related proteins (Cavalier-Smith 2002a), is found in most eubacteria (all phyla) but is rare in archaebacteria, found only in Methanopyrus and Methanothermobacter; probably other archaebacteria lost it—four times to generate the major archaebacterial clades, none of which could have been ancestral to eukaryotes as they lack MreB, which initiated eukaryogenesis by becoming actin for phagotrophy. RNA polymerase B is split in most archaebacteria including these two; such archaebacteria could not be ancestral to eukaryotes unless the split were reversed. The polymerase B split and MreB losses together exclude all archaebacterial branches as eukaryotes ancestors; the only remaining possibility (a stem lineage near the crenarchaeote/euryarchaeote divergence and first divergence within euryarchaeotes) would be disallowed if MreB entered Methanopyrus/Methanothermobacter by lateral transfer, not vertical inheritance. Scores of universal proteins and 310 protein domain superfamilies shared by eukaryotes and eubacteria are never found in archaebacteria (Cavalier-Smith 2002a; Yang et al. 2005); none could have entered eukaryotes vertically if archaebacteria were our ancestors. Probably most were vertically inherited by eukaryotes from a eubacterial ancestor prior to separation of pre-eukaryote and pre-archaebacterial lineages.

Of key importance is the proteasome, a tiny cylindrical protein-digestion chamber shared by all neomura and many actinobacteria only. Other eubacteria have no homologue or else a simpler one, HslV. HslV protease is a squat hollow cylinder of 12 identical proteins comprising two laterally attached hexameric rings. The proteasome core has two different proteins, related to HslV and each other, arranged as four heptameric rings making a longer and wider digestion chamber than HslV (Gille et al. 2003). I argued that core proteasomes evolved by duplicating HslV and its divergence into functionally different subunits (α, β) in the common ancestor of neomura and actinobacteria with proteasomes (Cavalier-Smith 2006a). The inner rings (β-subunits) retained protease activity but lost the ATPase-binding activity of HslV, whereas α-subunits lost protease activity and became able to bind a novel regulatory ATPase (Cavalier-Smith 2006a). Reverse evolution from this complex differentiated proteasome core to HslV by losing α-subunits and de novo acquisition by β-subunits of ATPase-binding capacity is mechanistically and selectively most unlikely; hypothetical reversal also leaves unexplained the origin of the proteasome. Thus, neomura plus actinobacteria are a clade; the root of the tree of life is therefore among other eubacteria (figure 5; Cavalier-Smith 2006a).

Figure 5 The tree of life. All 10 bacterial phyla that I currently recognize are shown, with Posibacteria split into subphyla Endobacteria and Actinobacteria (for more detailed classification see Cavalier-Smith 2002a, 2006a). It is uncertain whether Actinobacteria (more ...)

Figure 6 The phagotrophic origin of eukaryotes. (a) Transformation of growing trophic cells by phagotrophy and consequential endomembrane evolution. (i) A potentially flexible surface coat of N-linked glycoproteins facilitated the origin of phagocytosis by their (more ...)

As the eukaryotic morphological fossil record is incomparably better than that of bacteria, eukaryote phylogeny, the position of the root of the eukaryote subtree, and the nature of the first eukaryote are of utmost importance for properly interpreting it and dating the history of life. Recently, misunderstandings on these scores fostered by biased single-gene trees have been corrected by the use of exceptionally rare, practically irreversible shared-derived genetic characters like gene fusions and gene replacements and by multi-gene trees. That the root of the eukaryote tree is between animals and fungi on the one hand and the plant and chromist kingdoms on the other (figure 5) was proposed on grounds of ciliary and cytoskeletal evolution (Cavalier-Smith 2002b), and is now strongly supported by such molecular cladistic evidence (Stechmann & Cavalier Smith 2002, 2003; Richards & Cavalier-Smith 2005). Plants, chromists and three protozoan infrakingdoms (Alveolata, Excavata, Rhizaria) were grouped together as bikonts—a clade defined by evolution in their common ancestor of centriolar and ciliary transformation, in which at cell division the new centriole of each daughter cell bears the anterior cilium, but this reorients at the next cell division to become the older posterior centriole and cilium, often with different ultrastructure and beat pattern (Cavalier-Smith 2002b). This unique pattern of ciliary differentiation was known in all bikont groups, except Rhizaria. Now it has been found also in the rhizarian phylum Cercozoa (Karpov et al. 2006), so it was already present in the cenancestor of bikonts and helps define them.

The best available multi-gene tree based on 149 proteins strongly corroborates the bipartition between unikonts and bikonts and the monophyly of major groups (figures 5 and 7), including opisthokonts, unikonts, Plantae, Excavata and chromalveolates (Rodriguez-Ezpeleta et al. 2005). For parsimony in evolution of protein-targeting machinery in secondarily enslaved chloroplasts, Cavalier-Smith (1999) proposed that all chromophyte (chlorophyll c-containing) algae arose by a single enslavement of a red alga, and aplastidic chromalveolates evolved by chloroplast loss. This major phylogenetic simplification was rapidly compellingly supported by the serendipitous discovery of two gene replacements that must have occurred in the common ancestor of all chromalveolates. The original red algal plastid versions of the enzymes glyceraldehyde-phosphate dehydrogenase and fructose-bisphosphate aldolase were replaced by duplicates of the host cytosolic version of these enzymes (Patron et al. 2004). Chromalveolate monophyly means that the whole chromalveolate clade must be younger than red algae. There are many abundantly fossiliferous chromalveolates: haptophytes, diatoms, silicoflagellates, chrysomonads, dinoflagellates, and others that rarely fossilize—ciliates, in amber only—and, more dubiously, brown algae—I am sceptical of almost all fossil identifications of brown algae. Multi-gene trees show chromalveolates as younger than the primary divergence within reds: between (thermophilic) Cyanidiophyceae and the rest (Yoon et al. 2002). Not only symbiogenesis, but also LGT, can give relative (not absolute) dates independently of the fossil record, because of their rarity and the certainty that the donor cannot have evolved after the last common ancestor of all descendants of the recipient (disallowing time reversal). Thus, shared lateral transfers of bacterial genes by the metamonad superclass Eopharyngia (e.g. Giardia and Trichomonas) prove that the primary divergence within this group was after the evolution of all the bacterial donors. They also prove that Eopharyngia are a derived secondarily anaerobic clade within eukaryotes and excavates (Cavalier-Smith 2003a), and therefore that the root of the eukaryote tree cannot be between Giardia and Trichomonas, as 18S rRNA trees once suggested, corroborating evidence from many protein trees that rRNA trees were seriously misleading, with overall topology dominated by long-branch artefacts when both bacteria and eukaryotes are included.

Figure 7 Distance tree for 58 eukaryotes based on 2994 nucleotides of 28S rRNA. This weighted least squares power 2 (GTR+Γ+I model: a=0.59231, i=0.24389) tree is from von der Heyden (2004). Numbers at major nodes only from 500 wl2 bootstrap samplings (limited (more ...)

An important consequence of this root of the eukaryote tree and strong evidence that extant anaerobic eukaryotes probably all have organelles (hydrogenosomes or mitosomes) evolutionarily derived from aerobic mitochondria (Embley 2006) is that the last common ancestor of all eukaryotes was an aerobic amoeboflagellate with mitochondria. As this cenancestor arose by enslaving an α-proteobacterium, all eukaryotes must be younger than the oldest α-proteobacteria. Their origin must postdate the primary diversification of eubacteria into phyla and later diversification of proteobacteria into classes. Eukaryotes are younger than eubacteria, as morphological fossils showed long ago. This illustrates how phylogenetic analysis can order evolutionary events logically in time, independently of fossils. But only fossils give absolute dates.

Table 1 Earliest reasonably confident fossil dates for the major eukaryote groups.

The marked discrepancy between these late morphological records of all fossils unambiguously attributable to specific eukaryotic phyla and the very early occurrence of steranes often interpreted as eukaryotic derivatives (Brocks et al. 1999) is explained if the 2.7Gyr ago steranes actually came from eubacteria, at least four groups of which can make sterols (Cavalier-Smith 2002a; Pearson et al. 2003). Of these, Mycobacteria make cholesterol like eukaryotes, and belong to actinobacteria, the probable ancestors of eukaryotes. Sterol biosynthesis evolved polyphyletically by modifying universal eubacterial isoprenoid metabolism after atmospheric oxygenation made it mechanistically possible, long before eukaryotes. It was probably vertically inherited by the first eukaryote from an actinobacterium. Claims that mycobacterial sterol synthesis enzymes were laterally transferred from eukaryote DNA have not been substantiated by phylogenetic analysis (Cavalier-Smith & Chao 2003b), and reflect the false assumption that the ancestors of eukaryotes were archaebacteria, not modified derivatives of actinobacteria that generated the eukaryote/archaebacterial cenancestor. Fossil steranes are not sound evidence for eukaryotes (Cavalier-Smith 2002b); even were they specific for eukaryotes, the potential for downward mobility of petroleum fractions containing them is a perpetual worry for age authenticity—this problem is absent for body fossils, though their identification as eukaryotic or bacterial is also sometimes practically impossible, sometimes overoptimistic.

Some palaeontologists, rightly I think, do not accept Grypania (1.8Gyr old; Han & Runnegar 1992; redated by Knoll et al. 2006) as eukaryotic rather than a giant cyanobacterial sheath. Identity of the largest, most complex fossils ca 1.5Gyr ago often called eukaryotic (Javaux et al. 2001, 2003) is problematic. While some might be stem eukaryotes (Javaux et al. 2003), this is not compellingly demonstrated. I consider it more likely that most, probably all, are simply unusually large and complex prokaryotes. Unlike Javaux et al. (2001), I do not think their morphology demands the presence of an internal cytoskeleton or endomembranes; it is within the morphogenetic capabilities of bacteria. I agree with Butterfield (2005) that the morphologically most complex of these fossils is probably not an alga as originally assumed, but a sporangial entity broken from a branching trophic hyphal network, as are his beautiful 0.8–0.9Gyr fossils (mostly Tappania). In 2001, I microscopically examined the Harvard Tappania plana specimen shown in fig. 4 of Javaux et al. (2003); I told Javaux that I thought it was not a single algal cell and that the structure labelled there by an arrow was the broken base of a branched hypha passing through the outer ‘wall’ towards the dense central mass. However, I disagree with Butterfield's suggestion that these fossils are probably fungi. They could instead be actinobacteria similar to modern Amycolatopsis decaplanina and Kibdelosporangium that generate large pseudosporangia from mycelial networks (Wink et al. 2004); their pseudosporangia are very variable in size, but usually somewhat smaller than in the fossils, but the thickness of their hyphae is indistinguishable. Confusion of actinomycetes and fungi was long-standing even for extant cultivated species; actinomycetes were treated as fungi in my undergraduate mycology textbook (Alexopoulos 1952); it is markedly more difficult to differentiate between them in fossils.

I also do not accept Bangiomorpha (Butterfield et al. 1990; Butterfield 2000) as a red alga or eukaryote; it could be a mixture of two cyanobacterial species, possibly stigonematalean; I was mistaken in earlier calling it Oscillatoria-like (Cavalier-Smith 2002a). It is certainly not like the red alga Bangia, as it lacks the characteristic elongated projections of cells in its holdfast that result from intrusive growth, which had they been present would have supported a eukaryote nature. As some filamentous red algae lack such intrusive cells and are effectively indistinguishable at low resolution from Bangiomorpha, we cannot be sure it is not a red alga. However, the holdfast and all other morphological features are no more complex than in some cyanobacteria. Unfortunately, it lacks characters that require presence of a eukaryotic cytoskeleton or endomembrane system that would firmly make it eukaryotic or that prove that it cannot simply be a complex cyanobacterium. Assuming Bangiomorpha to be correctly dated at 1.2Gyr (currently unclear; perhaps younger), Bangiomorpha is 600Myr older than any fossil reasonably confidently a red alga (Xiao & Knoll 1999) and ca 400Myr older than unambiguously eukaryotic fossils.

Another Late Proterozoic fossil type sometimes called eukaryotic are broad filaments, e.g. Palaeovaucheria, conventionally assigned to the heterokont Xanthophyceae and dating to ca 1Gyr ago (Butterfield 2004). These lack features unambiguously identifying them as xanthophytes, heterokonts, or even eukaryotes; they could be large filamentous cyanobacteria. Although we cannot rule out their being a stem eukaryotic alga unattributable to phylum or kingdom, their early date makes it virtually impossible that they are xanthophytes, given the well-established heterokont molecular trees and recency of all well-fossilized heterokont groups (diatoms, haptophytes, chrysomonads, silicoflagellates); none go even into the Palaeozoic (table 1), and trees show xanthophytes as no older (figure 7; Cavalier-Smith & Chao 2006).

There are many technically inadequate attempts to date eukaryotes assuming a universal molecular clock (Graur & Martin 2004), even though this assumption is false and grossly misleading, as evolutionary rates of all molecules can change idiosyncratically across a molecular tree, sometimes by many orders of magnitude (Cavalier-Smith 2002a). Yet some stability in evolutionary rates of nucleotide or amino acid substitution exists in local parts of the tree for a given molecule, and assuming a local ‘clock’ is sometimes useful for interpolating between known fossil dates. Extrapolation backwards earlier than fossil dates is considerably more hazardous; nonetheless, with a statistical model that allows rates to change across the tree, reasonably realistic models of substitution, and calibration of many points on the tree by known fossil dates (not one as in many studies), it is worth attempting to try to assess congruence between the fossil record and molecular phylogenetic trees, and to improve interpretation of both if they mismatch; Roger & Hug (2006) discuss some of the problems.

Only two such studies, using a Bayesian ‘relaxed clock’, attain the highest current standards. One based on 129 proteins, 36 eukaryotes and six Phanerozoic fossil calibration points based on macro-organisms (three animals, two land plants, one fungal) deduced 950–1259Myr ago (mean 1085) for the divergence of Amoebozoa from other eukaryotes (Douzery et al. 2004); given the difficulty of resolving the branching order of Amoebozoa and bikonts on trees (Cavalier-Smith 2004b), this probably approximates to an estimate for the cenancestral eukaryote. Berney (2005) and Berney & Pawlowski (2006), using 18S rRNA (1465 nucleotides) from 83 eukaryotes and 26 Phanerozoic microfossil time constraints, similarly dated the cenancestral eukaryote at 948–1357Myr ago (mean 1126). Using microfossils to constrain trees has the advantage that such organisms as diatoms, radiolaria and foraminifera are superabundant, with billions of clearly identifiable fossils, so a group's first appearance is more accurately dated than for sparsely recorded macro-organisms and unlikely to be underestimated. However, Bayesian algorithms are unlikely to model episodic evolution (e.g. dramatic short-term changes in evolutionary rate) well. For rRNA, the evolutionary rate probably massively increased in the stem eukaryote lineage (possibly by approx. four orders of magnitude; Cavalier-Smith 2002a) then subsequently greatly slowed again. If slowing finished before the initial radiations, the earlier fast rates should not distort Berney's conclusion (though probably fatal for similar studies of the whole tree). If higher rates lasted after the primary eukaryotic divergence but declined before fossil calibration dates, this gene tree would overestimate the age of the cenancestral eukaryote. As many proteins used by Douzery et al. (2004) were ribosomal (coevolving with rRNA), their dates also could be overestimates. Many other proteins used in their dataset were novel eukaryotic proteins, expected to have evolved faster early on when function was freshly established than later when stabilizing and purifying selection would predominate; thus, the protein dates may be overestimates. This is supported by reanalysis of these data by a technically superior method; calculating trees separately for each gene and then combining likelihoods (Roger & Hug 2006) gave a younger estimate of ca 900Myr ago. Based on numerous proteins, this should be more reliable than the 200Myr earlier date from 18S rRNA (Berney 2005; Berney & Pawlowski 2006), especially if hyper-fast rRNA evolution in the pre-eukaryote stem (Cavalier-Smith 2002a) persisted somewhat after the earliest divergences.

Berney used a clever method to test identifications of contentious Proterozoic fossils like Palaeovaucheria, Tappania, Pterocladus (claimed to be a cladophoran green alga) and Bangiomorpha. He assumed palaeontologists correctly identified them and used their dates as constraints for Bayesian calculations, deducing a hypothetical origin time for groups with easily identified fossils, e.g. diatoms, dinoflagellates. Invariably, this dated these groups with a continuous unambiguous fossil record grossly earlier than their actual fossil dates (typically overestimating ages of rhizosolenid diatoms, pennate diatoms and coccolithophorid haptophytes by 4–5 times). This strongly supports the view that these fossils were misidentified as crown eukaryotes (Cavalier-Smith 2002b). Although some might be stem eukaryotes, all are likely simply complex bacteria. Likewise, assuming that some vase-like fossils are euglyphids (Porter & Knoll 2000; Porter et al. 2003) overestimates these dates 4–8 times (because euglyphids nest relatively shallowly in the cercozoan tree; Cavalier-Smith & Chao 2003a); by contrast arcellinids nest relatively deeply in the amoebozoan tree (Nikolaev et al. 2005); their date of more than 760Myr ago (Porter & Knoll 2000; Porter et al. 2003) is consistent within the error range with estimates of ca 950Myr ago for Amoebozoa (Berney 2005; Berney & Pawlowski 2006) and ca 0.9Gyr ago for eukaryotes (Roger & Hug 2006).

My objection to identifying these fossils as eukaryotes is primarily on morphological grounds; morphology inadequately supports it. Their conflict with the phylogenetic evidence for approximately equal ages for bikonts and unikonts, and thus with plants and chromists being not dramatically older than animals, made me examine them critically, but I should be sceptical even were they much younger. The relaxed Bayesian estimates for red algae (ca 730Myr ago from rRNA; less than 928Myr ago from proteins) and Bangiophyceae (ca 680Myr ago from rRNA) are consistent with ca 570Myr old fossils being genuine florideophytes (Xiao et al. 2004). These estimates should not be called ‘molecular’. They synthesize fossil and molecular data with a model for rate change across the tree, i.e. fossil dates modulated and integrated by extensive comparative molecular evidence. Molecular trees and substitution models help integrate the partial fossil evidence with more representative molecular data and apply dates to groups lacking fossils. Molecular sequences alone give no dates.

Inferred cenancestral dates for chromalveolate groups (haptophytes 560Myr ago; heterokonts 580Myr ago; alveolates 600Myr ago) are consistent with the requirement that chomalveolates postdate red algae (having originated by one red algal enslavement), but the mean unseparated protein date for chromalveolates (919Myr ago, Douzery et al. 2004) is only just consistent with the red algal protein date and markedly earlier than the rRNA red algal date, suggesting an overestimate. Estimates for the red/green algal divergence of ca 930 (rRNA) and 1010Myr ago (proteins, unseparated; Douzery et al. 2004) should be slightly younger than for the origin of Plantae as glaucophytes diverge earlier. However, as rRNA gives 812Myr ago for the primary animal radiation and proteins 695Myr ago, early backwardly extrapolated dates of these studies may be 100–300Myr too old, it being unlikely that animals have been missed in the fossil record (Peterson & Butterfield 2005; Conway Morris 2006)—their true age is probably ca 555Myr ago (if Vendobiota are not animals) or ca 570Myr ago if Vendobiota are animals.

Taking fossils and molecular evidence together, 900±100Myr ago is the most likely origin date for eukaryotes. Dating the origin of chloroplasts and Plantae is harder, but most likely 570–850Myr ago, whereas ca 570Myr ago is reasonable for the origin of chromalveolates, opisthokonts, Rhizaria and excavates, when snowball earth melted.

It is unlikely that neomura are much older than either eukaryotes or archaebacteria. Probably eukaryotes and archaebacteria each originated almost immediately after neomura (Cavalier-Smith 2002a). It is especially unlikely that neomura can be as old as eubacteria. If they were, bacteria with a neomuran glycoprotein envelope, DNA-handling enzymes and ribosomes, but lacking specific archaebacterial properties (e.g. isoprenoid ether lipids, archaeosine, flagella), must have existed for 2.5Gyr before eukaryotes without leaving any descendants not converted to eukaryotes or archaebacteria; it is incredible that a major niche for such an intermediate bacterium persisted for more than 2Gyr, but was suddenly wiped out coincidentally with the origin of eukaryotes.

There is no morphological fossil record for archaebacteria and scant prospect of ever finding any. One potentially important biomarker unique to archaebacteria is the double-length C40 isoprenoid lipids that span the entire CM of hyperthermophilic (not other) archaebacteria. As hyperthermophily is probably the ancestral state (Cavalier-Smith 2002b), these probably first evolved in the cenancestral archaebacterium and would mark their origin. Unfortunately, they are unknown earlier than the Mesozoic, suggesting relative instability (they ought to date from the Early Neoproterozoic (ca 0.9Gyr), like eukaryotes) that underestimates the age of archaebacteria (Cavalier-Smith 2002a). Linear C20–C30 isoprenoids ca 1.6Gyr are proposed as biomarkers for halophilic archaebacteria, which typically make membrane lipid isoprenoids of C20 and C25 (Summons et al. 1998). However, all eubacteria have long-chain isoprenols of C50 or C55 lengths as membrane carriers, and one wonders whether partial degradation products of these or other eubacterial isoprenoids might persist and be hard to distinguish from halobacterial isoprenoids. R. E. Summons (2005, personal communication) noted that polyunsaturation of these eubacterial isoprenols makes them highly susceptible to degradation by light and oxidation, unlike saturated archaebacterial lipids. Given that atmospheric oxygen was substantially sparser ca 1.6Gyr ago (Holland 2006) and the deep ocean might have been anoxic, possibly anoxic burial of eubacteria occurred rapidly enough for degradation to yield the observed spectrum of chain lengths.

Organic carbon samples 2.8Gyr ago (and somewhat later) are markedly lighter isotopically than achievable by one-step CO₂ fixation by Rubisco (Δ¹³C −38‰). Previously, assuming that archaebacterial methanogenesis was ancient, this was interpreted as the result of recycling of biogenic methane made by archaebacteria (archaebacterial methanogenesis has exceptionally low ¹³C:¹²C ratios: Δ¹³C −30–50‰) by methanotrophic bacteria into organic carbon (Hayes 1983, 1994). Hayes suggested that the rise of oxygen restricted methanogen habitats sufficiently to make their hypothetical contribution quantitatively undetectable less than 2.2Gyr ago. As phylogenetic and morphological fossil evidence in combination strongly indicate that archaebacteria are much younger than 2.0–2.8Gyr, explanations for these light carbon deposits cannot involve archaebacteria. Straus et al. (1992) suggested that chemosynthetic bacteria or anoxygenic photosynthesis might participate in two-step fractionation that could produce lighter organic carbon than Rubisco alone. A specific suggestion was a chemotroph (presumably using Rubisco) taking as substrate CO₂ already made light by photosynthetic fixation and released by respiration. The lightness of the final product would depend on local recycling efficiency compared with mixing in the atmospheric pool. Section 5d develops an analogous explanation involving greater phototroph diversity generated by the glycobacterial revolution 2.8Gyr ago.

Ultralight reduced carbon is also made by autotrophic acetogenesis; some Clostridiales thus make acetate from CO₂ and hydrogen, with similar ¹³C depletion to methanogenesis (Δ¹³C −59‰; Gelwicks et al. 1989). As they use the same input materials as methanogens but are undoubtedly much older, they are a potentially more plausible explanation for the isotopic data. As anaerobes, their relative contribution to fixed carbon compared with aerobic phototrophs would have become insignificant just when the ultralight signal went. However, Clostridiales are Endobacteria, possibly not present as early as 2.7Gyr ago: §4d suggests they arose 2.3–1.5Gyr ago, but does not firmly exclude that they evolved earlier and contributed to the ultralight signal.

Another potential explanation is abiotic sources of methane plus aerobic biological methanotrophy. Some abiotic methane is not ultralight. Iron- and nickel-rich alloys, common in some minerals, catalyse production from bicarbonate at 200–300°C of abiotic methane indistinguishably ultralight (Δ¹³C−35–50‰) from archaebacterial methane. However, as §5d argues, biotic depletion is more likely. But, given the potential of three very different kinds of explanation not involving archaebacteria for the 2.8–2.1Gyr ago carbon isotopic data, they are not specific evidence for Archaean or Palaeoproterozoic archaebacteria, especially as all other evidence strongly contradicts that. Hayes (1994) recognized that ‘the antiquity of methanogens… is more speculative than proven’. I consider it now disproved (Cavalier-Smith 2006a).

Biomarkers for hopanoids suggest that glycobacteria, cyanobacteria in particular, existed more than 2.7Gyr ago (Summons et al. 1999, 2006), assuming that these potentially mobile organics are truly endogenous to these rocks, which is uncertain (Schopf 2001). Because of this worry, the extensive evidence that oxygenic photosynthesis had arisen by the Late Archaean (Holland 2006) is probably stronger evidence for negibacteria and cyanobacteria. If 2.7Gyr hopanes are endogenous, this is a minimum age, earlier deposits being unstudied. I suggest that the glycobacterial revolution and origin of cyanobacteria occurred 2.8–2.9Gyr ago and that all indications of life from 3.5 to 2.9Gyr ago can be attributed to Eobacteria, with ecosystems entirely dependent on photosynthetic Chlorobacteria for primary production, or to stem bacteria. I suggest ca 2.9Gyr ago for the origin of oxygenic photosynthesis because the limited and temporary oxidative removal of abiogenic atmospheric methane by biogenic oxygen seems a simpler explanation of the 2.9Gyr ago glaciation than indirect depletion of hydrogen and methane by biological sulphate reduction as Kasting & Ono (2006) suggest. If oxygenic photosynthesis originated in the common ancestor of Hadobacteria and cyanobacteria (Cavalier-Smith 2006a), biogenic oxygen first came not from cyanobacteria but from an immediately preceding stem lineage. However, there is no reason to postulate significant delay after photosystem duplication before the origin of phycobilisomes. A 100Myr delay could have occurred only if a well-adapted oxygenic bacterium existed that long in a well-defined adaptive zone. No such organisms exist, consistent with quantum evolutionary principles that they were evolutionarily unstable, short-lived transitional intermediates in perfecting oxygenic photosynthesis. It is unlikely that cyanobacteria postdated oxygenic photosynthesis by even 1Myr, so if the 2.9Gyr glaciation was caused by methane oxidation glycobacteria and cyanobacteria probably originated ca 2.9Gyr ago. This increased biotic complexity of microbial mats, with at least one new group that used only Rubisco for carbon fixation, would have made carbon fixed within mats still lighter—directly by increasing the Rubisco:hydroxypropionate ratio and indirectly by increasing the scope for repeated recycling of light CO₂ within the mat, and might therefore have caused the ca 2.8Gyr ago ultralight carbon isotopic spike (Hayes 1983, 1994; Hayes & Waldbauer 2006). Carotenoid biomarkers demonstrate proteobacteria 1.6Gyr ago (Brocks et al. 2005), and R. E. Summons (personal communication) found them in 2.7Gyr old samples. If confirmed, the radiation into all four photosynthetic glycobacterial phyla plus the heterotrophic spirochaetes was over by 2.7Gyr ago.

If Chlorobacteria are the earliest diverging phylum, as I argue, they must be older than the divergence of Hadobacteria (i.e. 2.9Gyr ago, if oxygenic photosynthesis is that age, as just suggested, and did evolve immediately prior to that node on the tree; figure 5). The marked separation of chlorobacteria on rRNA and some other trees from other photosynthetic groups suggests that they may have arisen at least 10% earlier than that (i.e. ca 3.2Gyr ago); an origin 20% earlier (ca 3.5Gyr ago) would be not unreasonable, assuming some substitutional saturation reducing the good resolution on sequence trees that such a difference in age should otherwise cause. Thus, chlorobacteria probably originated ca 3.2±0.3Gyr ago. Although Chlorobacteria are probably the most ancient phylum, the first life would have been a stem lineage preceding the divergence of chlorobacteria and other organisms. Even though the bacterial cenancestor was probably already complex, with probably at least 2000 genes and a complex envelope with peptidoglycan and lipoprotein and fully complete metabolic and bioenergetic (probably both photosynthetic and anaerobic respiratory) repertoires, bacterial generations can be so rapid, selective forces so strong, and bacterial populations so huge, that bacterial evolution can be extremely fast on a geological time-scale. The time from the first life to the cenancestral negibacterium need not have exceeded 1–10Myr. As Eobacteria include photoautotrophs, photoheterotrophs, non-photosynthetic anaerobic respirers and fermenters, thermophiles and mesophiles, they could have formed a diversified photosynthesis-based ecosystem for hundreds of millions of years before oxygenic photosynthesis and glycobacteria. As photosynthetic chlorobacteria are typically filamentous gliders, like cyanobacteria, they could have helped generate all genuinely biogenic Archaean stromatolites. Contributions from cyanobacteria could have been largely, perhaps entirely, Proterozoic.

Flagella probably evolved in the common ancestor of Gracilicutes (Proteobacteria, Sphingobacteria, Planctobacteria, Spirochaetes) and Eurybacteria. As this ancestrally photosynthetic assemblage is probably sister to cyanobacteria, it should be approximately equal in age. As multigene trees do not clearly resolve the branching order of these phyla, deduced mainly by molecular cladistics, they probably diverged in a rapid early radiation of glycobacteria. The often facultatively aerobic proteobacteria, which include purple photosynthetic bacteria and an immense variety of chemotrophs and heterotrophs, e.g. sulphate reducers and H₂S- and iron-oxidizers, and the Sphingobacteria that comprise anaerobic green-sulphur bacteria (e.g. Chlorobium) and aerobic Flavobacteria (including predatory cytophagas) are probably almost as old as cyanobacteria. Thus, rapid diversification of photosynthetic machineries and antenna pigments followed the glycobacterial revolution. Photoreceptors allowing physiological chromatic adaptation and light-oriented movements probably also stem from the glycobacterial ancestor; sensory rhodopsins occur in cyanobacteria and proteobacteria (Jung et al. 2003; Ruiz-Gonzalez & Marin 2004; Venter et al. 2004; Vogeley et al. 2004). One green non-S bacterium photosynthesizes using geothermal radiation in deep-sea vents (Beatty et al. 2005); like sulphur-oxidizers there, they are glycobacteria and so younger than chlorobacteria. If deep ocean chlorobacteria do not include autotrophs, vent autotrophy probably postdates the glycobacterial revolution and is irrelevant to early life.

Although some suggest that bacterial sulphate reduction (BSR) is 3.5Gyr old, sulphate deposits more than 3.2Gyr ago, but not for the next 900Myr, imply that it evolved less than 3.2Gyr ago (Kasting & Ono 2006). Phylogenetic evidence fits the widespread consensus that isotopic sulphur fractionation began only ca 2.7Gyr ago and sharply increased quantitatively ca 2.3Gyr ago when atmospheric oxygenation would have increased sulphate supplies (Holland 2006; Kasting & Ono 2006). Heterotrophic anaerobic iron and sulphate reducers probably evolved early in diversification of Proteobacteria (sensu Cavalier-Smith 2002a) in lineages that lost photosynthesis. Probably BSR was first exclusively by Proteobacteria, making its onset a marker for their origin; possibly all Archaean BSR was proteobacterial. However, metabolism of most Chlorobacteria is unknown; if their huge uncharted diversity (Morris et al. 2004) included sulphate reducers, BSR could be 3.2Gyr old. Carotenoid biomarker and phylogenetic evidence fit an origin of Proteobacteria and Sphingobacteria ca 2.8Gyr ago. As the distinctively motile heterotrophic spirochaetes adapted for corkscrewing through soft media like microbial mats or sloppy organics-rich mud, probably diverged slightly before, they are marginally older. Thus, by ca 2.8Gyr ago seven of the eight negibacterial phyla had already evolved. Only Planctobacteria, apparently sisters of (or derived from) Proteobacteria, which mostly lost murein and are often aerobic flagellates or parasites, may be somewhat younger. Possibly Planctobacteria evolved 2.1–2.5Gyr ago during the great oxygenation event, becoming significant heterotrophic bacterioplankton; a substantially later origin is unlikely, or else they would nest more definitely and more shallowly within Proteobacteria in sequence trees.

Eurybacteria are probably at least as old as Proteobacteria; although including photosynthetic Heliobacteria, which do not fix CO₂, and the thermophilic or hyperthermophilic Thermotogales, they contribute little to biogeochemical cycles, having low diversity and narrow metabolic capabilities. Antiquity of Thermotogales is hard to estimate as hyperthermophily of Thermotoga probably makes it branch artefactually early on many trees; nonetheless, Thermotogales probably did diverge relatively early and may be nearly as old as Proteobacteria. As hyperthermophilic Thermotoga species nest relatively shallowly among thermophilic species, transfer of hyperthermophilic genes (notably reverse gyrase) from archaebacteria (Nesbo et al. 2001; Nesbo & Doolittle 2003) is much more recent, not contradicting the late origin of archaebacteria. The evolutionary importance of eurybacteria is the evidence they give of early diversification of flagellate glycobacteria and the likelihood that they were ancestral to Posibacteria and thus ultimately to neomura and eukaryotes (Cavalier-Smith 2006a).

Although Actinobacteria are related to Endobacteria, it is uncertain if they are sisters to Endobacteria, and roughly equally old (previously assumed; Cavalier-Smith 2002), or derived from them, and thus younger. Many actinobacteria use the thick posibacterial walls to become morphologically highly complex; actinomycetes can be macroscopic and visible to the naked eye, like some cyanobacteria that they rival and sometimes exceed in morphological complexity. Actinomycetes are throughout the world in soils and sediments, even in deepest oceanic trenches. Surprisingly, palaeontologists almost never consider them possible candidates for more complex fossils, especially those conventionally seen as candidate stem eukaryotes. 1.5Gyr ago is when distinctly larger than previously fossil thick-walled cysts begin to occur at low frequency (Schopf & Klein 1992; Knoll et al. 2006). Butterfield (2005) broke with the tradition that most if not all of these are algal, for which there was never compelling evidence; in his words, micropalaeontologists adopted ‘a search image weighted excessively in favour of unicellular plant protists’. He reasonably suggests that Tappania and possibly also Germinosphaera, Foliomorpha, Trachyhystrichosphaera, Shuiyousphaeridium and Dictyosphaera are not single cells but multi-genome marine benthic mycelial, osmotrophic, heterotrophs. He thought they were fungi, but such as Tappania could be actinomycetes similar to Amycolatopsis (Wink et al. 2004) and Kibdelosporangium. If they are, as I suspect, actinomycetes are probably at least 1.5Gyr old if any Roper fossils (Javaux et al. 2001, 2003; Knoll et al. 2006) are actinomycetes (I agree with Knoll et al. (2006) that Roper T. plana may differ from Butterfield's Neoproterozoic fossils).

Since actinomycetes are just one derived subclade within the actinobacterial tree, unless Tappania is an extinct derivative of an earlier branching clade now represented only by morphologically simpler species, then actinobacteria are probably distinctly older. I suggest that Actinobacteria are no younger than ca 1.8Gyr. Thus, we have a rough estimate of the maximum age of Endobacteria as 2.3Gyr (2Gyr more likely) and a minimum age for Actinobacteria of ca 1.8Gyr. These estimates are compatible with their being sisters and both 2–1.8Gyr old or with Actinobacteria being younger than Endobacteria and derived from them. The closeness of these estimates implies that even if Actinobacteria evolved from Endobacteria, they should nest deeply within Endobacteria, less than or equal to 10% of the distance from their base with perfect phylogenetic reconstruction. In the real world of imperfect phylogenetic algorithms and pervasive systematic biases, e.g. from the exceptionally high-GC nucleotide composition of actinobacterial DNA and unusually low GC in Endobacteria, we should not expect Actinobacteria and Endobacteria to group together on most single-gene or multigene trees, even if Posibacteria are monophyletic. If they are sisters, mild systematic biases—or sampling error on single-gene trees—could falsely separate them.

Similar considerations and more accurately dating Actinobacteria are important for interpreting sequence trees in relation to neomuran origins. If eukaryotes and neomura are only 0.9Gyr old and Actinobacteria are 1.8Gyr old, perfect trees ought to nest neomura within Actinobacteria. Previously, as some eukaryote-like properties are phylogenetically restricted, I assumed that actinobacteria are paraphyletic ancestors of neomura, and thus older (Cavalier-Smith 2002a). If instead neomura and Actinobacteria are sisters (not disproved) and ca 1.5Gyr old, as in some fossil interpretations, neomura should NOT nest within Actinobacteria on perfect trees but should be their sisters (as they are weakly on some trees, especially if the often closer, probably artefactual, position near neomura of Thermotoga/Aquifex is set aside). Systematic bias from accelerated evolution of most neomuran genes may be sufficient to exclude neomura from Actinobacteria artefactually on most sequence trees, even if they differ in age by 800Myr; thus we cannot use their normal exclusion to argue that their ages are markedly closer than that, or for their being sisters.

Figure 8 Key events in global history, integrating evidence from palaeontology and cell phylogeny. The Late Archaean and Late Proterozoic saw momentous global changes caused fundamentally by two biological revolutions—the glycobacterial and neomuran. Most (more ...)

During this quasi-equilibrium posibacteria evolved, paving the way for its disruption by the neomuran revolution near the Meso/Neoproterozoic boundary and the immediately ensuing eukaryote and archaebacterial revolutions that generated the modern world. This third phase of Earth history I call ‘the age of eukaryotes’ as it has been dominated by greater morphological complexity made possible by the eukaryote cell and phagotrophy. As for the age of cyanobacteria, there was early instability for at least 100Myr and global glaciation while new groups diversified to generate the major novel types. Invention of archaebacterial methanogenesis and methanotrophs, origin of the chloroplast and secondary symbiogeneses to form more complex algae, and the origin of fungal hyphal heterotrophy were probably all between 850 and 570Myr ago, amidst the Neoproterozoic snowball Earth episodes. All major forms of neomuran matter and energy processing evolved by 570–545Gyr, when the origin of epithelial and mesenchymal connective tissue generated the animal kingdom and immediately following Cambrian explosion of animals.

Methanotrophy and biological methanogenesis are both done by a suite of 15–16 different enzymes related to those that mediate other C1-compound conversions, some using unusual cofactors made by enzymes encoded by ca 10 other genes (Chistoserdova et al. 2004). Belief that these 25–26 genes were restricted to methanogenic archaebacteria and proteobacteria of subphylum Rhodobacteria (purple bacteria plus non-photosynthetic descendants) fostered hypotheses of LGT from archaebacteria to proteobacteria (Chistoserdova et al. 1998; Boucher et al. 2003; Martin & Russell 2003) or vice versa (Cavalier-Smith 2002a). Lateral transfer from archaebacteria, the more popular, was temporally impossible because methanogens are probably about three times younger than Rhodobacteria (ca 2.7Gyr old), and practically highly improbable because their C1-related genes are scattered all over the chromosome and could hardly all be cotransferred. Reverse transfer from Proteobacteria to Archaebacteria is mechanistically more plausible as many are closely clustered in operons in eubacteria and could theoretically be cotransferred. However, discovery of related genes in Planctobacteria, probably sisters of Proteobacteria, and in non-methanogenic archaebacteria and one even in the actinobacterium Streptomyces, plus recent phylogenetic analysis argues strongly that almost all these genes have simply been inherited vertically, being repeatedly lost by lineages without them (Chistoserdova et al. 2004). Lateral transfer need be invoked only for one Fae homologue of unknown function from a proteobacterium to the planctomycete Pirellula. Phylogeny of six genes with straightforward history (three C1 pathway and three cofactor genes) is entirely congruent, if rooted between Proteobacteria/Planctobacteria (collectively Exoflagellata; Cavalier-Smith 2002a) and Archaebacteria, with my rooted universal tree (figure 5), favouring vertical inheritance. Gene phylogenies contradict suggestion D of Chistoserdova et al. (2004) of lateral transfer from Planctobacteria to Proteobacteria and Archaebacteria.

The simplest interpretation of evolution of methylotrophy and methanogenesis is not scenario E of Chistoserdova et al. (2004), which makes the usual incorrect assumption that the tree's root is between neomura and eubacteria, but that these C1 enzymes first evolved in the common ancestor of Gracilicutes and Eurybacteria, were inherited vertically by Planctobacteria, Rhodobacteria and Archaebacteria, and lost by eurybacteria, eukaryotes, most Posibacteria, Sphingobacteria and Spirochaetes. In Planctobacteria they do not mediate methylotrophy or methanogenesis; they originally probably oxidized C1 compounds, e.g. formaldehyde. Aerobic methylotrophy and methanotrophy occur only in α- and γ-proteobacteria, probably originating after they diverged from planctobacteria. If vertically inherited, these enzymes must date approximately to the ancestral rhodobacterium (2.7Gyr; see above); C1 enzymes generally must be as old. Thus, methylotrophy extends back to the 2.78 ultralight C-isotope spike discussed above. Methanotrophy need not be that old; adding only one enzyme, methane monooxygenase, more easily transferred by LGT than the whole pathway, would convert a methylotroph to a methanotroph. However, the closest relative of proteobacterial methane monooxygenase is ammonium monooxygenase of β-proteobacteria, somewhat more divergent from methane oxygenases of α- and γ-proteobacteria than they are from each other, suggesting that divergence in function of the two paralogues occurred in the ancestral rhodobacterium and methanooxygenase and methanotrophy also originated ca 2.7Gyr ago, as suggested for methylotrophy. Some aerobic methanotrophs can scavenge methane even from the atmosphere (Henckel et al. 2000). Aerobic methanotrophy needs more than 20p.p.m. methane; atmospheric abiotic methane levels needed to solve the faint sun problem would have provided enough. The initial limiting factor was oxygen (Hayes 1994).

It is reasonable that C1 oxidation originated at virtually the same time as oxygenic photosynthesis providing the oxygen (2.8–2.9Gyr ago). The initial impetus could have been protection against harmful molecules like formaldehyde by converting them to CO₂. Abiotic methane destruction was probably markedly accelerated ca 2.7Gyr ago by aerobic methanotrophy. Lovelock (1988) and Pavlov et al. (2000) suggested that the origin of oxygenic photosynthesis would have accelerated methane removal by oxygenating the atmosphere and could thus have caused the 2.45–2.25Gyr ago global snowball Earth episodes (probably two). Section 5d discusses this further.

Methanogenesis is probably younger than archaebacteria, as methanogens nest within one archaebacterial subphylum, Euryarchaeota (Gribaldo & Brochier 2006). As archaebacterial sequence trees are substantially non-clock like (Gribaldo & Brochier 2006), estimating the age of methanogens is somewhat hazardous. In the absence of a Bayesian relaxed clock analysis using an arbitrary age for the archaebacterial cenancestor to estimate how much younger methanogens may be, a rough estimate can be made: inspection of the concatenated ribosomal protein tree (Gribaldo & Brochier 2006) and a crude clock suggest that the cenancestral methanogen was 13–32% younger than the cenancestral archaebacterium. A rounded middle 20% and date of 0.9Gyr ago for archaebacteria (as for eukaryotes; see above) gives ca 720Myr ago for the origin of archaebacterial methanogenesis. As this coincides with onset of Neoproterozoic near-global freezing, I suggest this atmospheric infusion of biogenic methane caused snowball Earth indirectly by the mechanism of Schrag et al. (2002). Eventually stabilization occurred as the biosphere adapted to the innovation. One important adaptation would have been the origin of anaerobic methanotrophy, done only by still-uncultivated archaebacteria (Orphan et al. 2002). Methanotrophic archaebacteria all nest within the methanogens (Gribaldo & Brochier 2006), so must be younger. They probably evolved from methanogens by reversing the metabolic pathway and adding methane oxidase; small innovations, given prior evolution of methanogenesis (Chistoserdova et al. 2005). Anaerobic methanotrophs are most related to Methanosarcinales, which from the same concatenated tree (Gribaldo & Brochier 2006) are 24–57% younger; if methanogens are 720Myr old and Methanosarcinales ca 30% younger, Methanosarcinales would be ca 500Myr old. If anaerobic methanogens are their sisters, they are similarly aged or a little older. I suggest they evolved less than 570Gyr ago, when snowball Earth episodes ceased, and by depleting methane close to its source (their methanogenic congeners) they have helped prevent excessive global warming by biogenic methane ever since.

Kopp et al. (2005) argue that cyanobacterial expansion and oxygen rise were so fast that cyanobacteria must have evolved only ca 2.4Gyr ago. But if the 2.7Gyr bitumen biomarkers are truly endogenous (uncertain) that cannot be true. Apart from possible past downward mobility, Kopp et al. (2005) give three reasons for discounting that 2.7Gyr hopanoid evidence for cyanobacteria: (i) some methylotrophs (e.g. Methylobacterium) make low levels of 2-methyl-bacteriohopanepolyol; (ii) other hopanols are not restricted to aerobes but present in Geobacter; (iii) methyl-bacteriohopanepolyol might have been made by ancestral cyanobacteria before oxygenic photosynthesis arose. All are refuted by phylogenetic arguments: Methylobacterium and other aerobic methylotrophs, and Geobacter are all proteobacteria, thus part of the Gracilicute clade, whose common ancestor must have had two photosystems like cyanobacteria (Cavalier-Smith 2006a). They are also part of a larger ancestrally flagellate eubacterial clade that is sister to the non-flagellate cyanobacteria. They cannot therefore be older than cyanobacteria and are probably somewhat younger. If any of these bacteria were present 2.7Gyr ago, cyanobacteria must have been also. Hopanols evolved no later than the common ancestor of cyanobacteria and proteobacteria; as cyanobacteria are holophyletic (Gupta et al. 2003), their common ancestor with Gracilicutes had both hopanols and two contrasting photosystems. Unless a reason other than the origin of oxygenic photosynthesis can be found for divergence of two photosystems in one cell, then this glycobacterial common ancestor already had at least primitive oxygenic photosynthesis. As there is no evidence for hopanoids or two photosystems in Eobacteria, they probably evolved in this common ancestor immediately before it diverged to cyanobacteria and gracilicutes/eurybacteria. Thus, hopanols in general, not just 2-methyl-bacteriohopanepolyol, are probably good proxies for the age of cyanobacteria and (insignificantly earlier) glycobacteria as a whole. If my argument (based on catalase evolution) that oxygenic photosynthesis evolved in the common ancestor of glycobacteria and Hadobacteria is correct (Cavalier-Smith 2006a), it probably even slightly preceded the origin of glycobacteria. Even though cyanobacteria probably do not make sterols (Summons et al. 2006), their biosynthesis requires oxygen, suggesting that biotic oxygen was made in microbial mats as early as 2.715Gyr ago. It is unwise to seek escape from this refutation in an unknown anaerobic mechanism for adding oxygen (Kopp et al. 2005).

Hayes & Waldbauer (2006) point out that the reservoir of reduced iron and sulphur able to mop up early biogenic oxygen was so huge that one expects a substantial lag after oxygenic photosynthesis before atmospheric oxygen levels rose. If oxygen was not generated faster than these minerals could remove it, a 400Myr lag can easily be explained by that reservoir size. Kopp et al. (2005) calculated biogenic oxygen fluxes compared with removal by Fe, assuming that cyanobacterial population expansion was limited only by total oceanic P or N nutrients, and concluded that atmospheric oxygen should rise enough to deplete methane within a few million years. But this has two problems: it is the available stock of reduced Fe and S that matters initially not its regeneration rate, which Kopp et al. assumed limited O₂ sequestration; secondly, it is unrealistic that cyanobacteria were limited only by nutrients. Before an ozone layer they would have been seriously restricted by harmful UV radiation (Cockell 2000; Cockell & Horneck 2001).

Consider the isotopic evidence for the state of the carbon cycle at the putative time (2.8Gyr) of the glycobacterial revolution. Rothman et al. (2003) and Hayes & Waldbauer (2006) showed that traditional steady-state carbon cycle models of the past 30 years have been oversimplified in two respects, making predictions disagree with the data. First, it is unrealistic to treat all oceanic/atmospheric carbon as one pool; one must differentiate between dissolved organic carbon and CO₂ (Rothman et al. 2003). Secondly, the crust/oceans/atmosphere is not self-contained; as geothermal activity continually injects CO₂ from the mantle, one must also weigh export of oxidizing power to the mantle in the balance (Hayes & Waldbauer 2006). Hayes & Waldbauer (2006) also note that as crustal burial processes important for redox mass balance calculations operate on a much slower time-scale than biological processes that directly generate or consume oxygen, isotopic fractionation of buried carbon need not correlate with atmospheric changes; large negative excursions of Δ¹³C in organic carbon are not mirrored by complementary changes in inorganic carbonate from 2.8 to 2.0Gyr ago, so they argue that changes in burial rates were not a major factor—in marked contrast to Neoproterozoic perturbations to the isotope record affecting both organic and inorganic deposits.

A further complication not considered previously is heterogeneity of the environment where primary production occurs. For Phanerozoic oceans dominated by plankton a homogeneous well-mixed model is reasonable. But in the Archaean and Early Proterozoic more than 2.3Gyr ago with no ozone layer, UV irradiation probably largely excluded photosynthetic bacterioplankton from the upper photic zone. Phototrophs were probably concentrated in shallow water where iron-impregnated mineral grains or snow overlying thin ice offered enough protection from UV radiation. Global productivity was much less than today; life concentrated in stratified microbial mats (fossilizable as stromatolites) rather than as well-mixed phytoplankton freely communicating with the atmospheric CO₂ pool. Mat complexity increased greatly with the glycobacterial revolution (figure 9) offering increased opportunity for internally recycling CO₂. Figure 9b shows a simplified model with only two strata, which allows two-stage fractionation by Rubisco. First, by cyanobacteria in the upper layer. Then by anaerobic bacteria, e.g. chlorobacteria or purple bacteria (proteobacteria) such as Chromatium in the lower layer. Fermentative and respiring heterotrophs would have been in both layers using organics generated by the phototrophs. CO₂ generated by these heterotrophs or phototrophs at night would be already depleted of ¹³C by Rubisco carbon fixation. Recycling it within the mat could deplete it again—and again, producing increasingly light organic carbon for burial. For a strictly two stage recycling (figure 9b) the depletion in buried organic carbon would be: ΔC+f_orrΔC, where ΔC is the depletion by a single stage, r is the fraction of CO₂ fixed by the second stage (assumed for simplicity to be solely purple bacteria, but in practice it could be any Rubisco-using phototroph in the mat) and f_or is the fraction of buried carbon that comes from the purple bacteria after recycling (1−f_or would come from cyanobacteria).

Figure 9 Ecosystems and the carbon cycle before and after the glycobacterial revolution. (a) In the age of chlorobacteria, life was mainly restricted to low diversity microbial mats protected from UV radiation by mineral particles. (b) After cyanobacteria evolved, (more ...)

If cyanobacterial photosynthesis and CO₂ trapping were very efficient, the upper layer could absorb most unfractionated CO₂ diffusing from the atmosphere, leaving only a small proportion for the lower layer. Thus, values for r could probably easily exceed 0.5 and could be as high as 1. Assuming a value of 0.9 and a value of 0.5 for f_or gives a total fractionation 1.45 times that of one stage. Note that f_or is determined by the relative biomass of the two layers, not by relative rates of synthesis. This is important because if photosynthesis in the lower layer were CO₂-limited, it would occur at a lower rate. But if the lower layer were thicker it could have a higher biomass, even if turnover was slower. If it were three times as thick and as dense, f_or would be 0.75 and total fractionation 1.65 times that of a single stage. In fact, mats can have much greater anaerobic than aerobic biomass (Sorensen et al. 2005). If one Rubisco stage can achieve −30‰, two could yield −45‰ and three could yield Δ¹³C −63‰. I propose that a sudden increase in biological complexity of microbial mats, favouring multistage recycling, caused the negative Δ¹³C spike 2.77Gyr ago and that this negative spike is the best way of dating the glycobacterial revolution. There is no need to invoke methanogenesis or acetogenesis. Logan et al. (1999) found that in terminal Proterozoic samples those from microbial mats were more depleted in ¹³C than planktonic samples, fitting the idea that mats can favour recycling already lighter CO₂ more than possible in suspension and thus serial light-biased fractionations by Rubisco. Mats often have three layers, with green-sulphur bacteria at the bottom, but in some habitats these largely replace the purple bacterial layer of figure 9c (Sorensen et al. 2005).

This stratified mat model explains timing of the onset and end of this unusual depletion. The negative spike would disappear completely when the photic zone was fully colonized by glycobacterial plankton (mainly cyanobacteria and proteobacteria) after the ozone layer was formed. If it is correct, the isotope record suggests that this expansion into the plankton was complete by 2.1Gyr in the later stages of the great oxidation event. However, the spike is greatly reduced by 2.55Gyr, well before even the slight oxidation that ended mass-independent oxygen 2.45Gyr ago and the likely onset of the ozone layer (2.3Gyr). This early decline of the negative spike clearly contradicts the methanogenesis explanation, which suggested that termination was caused by atmospheric oxidation of methane, which could not have occurred till after 2.45Gyr. It is no problem for the stratification theory if a partial shift into the plankton occurred prior to the ozone layer. A sound phylogenetic reason expects such a partial shift: the origin of flagella. Flagella arose after the glycobacterial revolution, but before the origin of proteobacteria (Cavalier-Smith 2006a). I suggest that the spike peak represents the brief period when cyanobacteria had evolved, but proteobacteria had not. At that time the only phototrophs were chlorobacteria and cyanobacteria, both with gliding motility and no flagella. If that is correct, the second anaerobic stage of CO₂ recycling (figure 9b) must then have been by Rubisco-using chlorobacteria, not proteobacteria. If flagella evolved ca 2.75Gyr ago, they would have enabled facultatively aerobic photosynthetic purple bacteria to swim above the mat and photosynthesize as plankton, reducing the recycling fraction r. If they kept in the lowest photic zone region they would suffer little UV damage.

This scenario is evolutionarily sounder than the original methanogenesis/methylotrophy as there is direct evidence from biomarkers that cyanobacteria and purple bacteria had both evolved by that date. If the overlying water column were also stratified, impeding downward flux of CO₂ from the atmosphere, this would have favoured preferential reuse of respired CO₂ compared with that from the atmospheric pool, increasing fractionation. However, this scenario does not require two different types of photosynthesis. It could occur by repeatedly recycling previously fixed and respired CO₂ by one kind of photosynthesizer alone, e.g. purple bacteria in an entirely anaerobic mat or as suggested by Straus et al. (1992) by a chemotroph within the mat recycling previously phototrophically fixed CO₂. The key thing is restriction of free mixing with the atmospheric CO₂ pool, i.e. the physical conditions not the precise organisms involved. Thus, extra-light carbon cannot be used to infer the presence of a particular kind of bacterium; rather it tells us either that special physical conditions to allow local recycling of gaseous products without global mixing were present in the particular habitats where the ultralight carbon was laid down or that little-known abiotic processes contributed to such a signature. However, although §4b cited some able to do so, there is no obvious rationale why abiotic causes should have been globally effective just then, neither before nor later, so they are probably irrelevant. Stratified mats, composed of the same globally distributed bacteria could give a consistent global isotopic signal without mat CO₂ having to mix freely globally. As Hayes & Waldbauer (2006) stress, kinetic effects can dominate the short term, and the spike was definitely short term compared with rock burial cycles. Microbial mats are not the only situation where isotopic fractionation can be higher than the standard model predicts. Biomass stemming from proteobacterial chemotrophs densely packed within animals is markedly more depleted (Δ¹³C −30–35‰; much more than in phytoplankton) than the 24‰ caused in one step by the corresponding Rubisco (Scott et al. 2004b). This is partly because the symbiotic system recycles already depleted inorganic carbon from the local environment; but CO₂ recycling within the animal's chemotroph mass could also occur.

The above is oversimplified. Like earlier models (Hayes 1983, 1994; Hayes & Waldbauer 2006) it ignores the fact that Δ¹³C by Rubisco varies evolutionarily, being typically lower in bacteria than eukaryotes, usually ca −20‰ not −30‰ as in spinach (Guy et al. 1993). This is a problem for the classical assumption that pre-2.77Gyr ago Archaean Δ¹³C of 35‰ reflects a single-stage Rubisco fractionation. There was no spinach in the Archaean. I suggest that two-stage recycling in mats is also needed to explain the more than 2.77Gyr ago data—unless they were produced purely abiotically and the 2.77Gyr spike is a marker not for glycobacteria but for the origin of life and photosynthesis, which cannot be conclusively rejected, but I think is unlikely. The problem for the chlorobacterial Early Archaean is even greater, because most chlorobacteria, except Oscillochloris that uses Rubisco with Δ¹³C of ca 20‰ (Ivanovsky et al. 1999), do not use Rubisco, but the 3-hydroxypropionate pathway, showing less favouritism for light carbon (e.g. Chloroflexus). We currently cannot infer which carbon fixation pathway was ancestral for chlorobacteria and whether both or only one were present in the Archaean. Assuming an equal mixture gives Δ¹³C of only −16‰, so much recycling would be needed to produce the observed values; some is unavoidable even if they only used Rubisco like that of Oscillochloris (Ivanovsky et al. 1999). However, there is a vast unexplored diversity of chlorobacteria, the most neglected bacterial phylum; other mechanisms may exist with higher fractionation. Immensely more research is needed on Chlorobacteria if we are to understand Archaean ecosytems.

Another complication is the unpublished evidence for a temporary pulse of oxygen sufficient to abolish mass-independent sulphur isotope fractionation around 2.9Gyr ago (Kasting & Ono 2006). Was the glycobacterial revolution, therefore, earlier than suggested? Not necessarily. One possibility is that this pulse came from pre-cyanobacterial oxygenic photosynthesizers, which probably evolved immediately before Cyanobacteria and Hadobacteria diverged (Cavalier-Smith 2006a). This could be a signal from that node in the tree, and could have caused the glaciation 2.9Gyr ago, as suggested above. If it were, why did oxygen not remain high enough to prevent mass-independent fractionation? One possibility is that aerobic respiration evolved and reduced oxygen concentrations at source sufficiently for abiotic hydrogen and methane to accumulate enough to scavenge atmospheric oxygen, keeping it below the critical level more than 2.45Gyr ago. Thereafter continued oxygenic photosynthesis produced the ozone layer, allowing cyanobacterial populations to expand rapidly and oxidize the earth. Once oxygen reached a certain threshold and the ozone layer began, positive feedback through population expansion would make oxygen levels rise explosively. Snowball Earth itself would have favoured phototroph expansion into the plankton through UV protection from snow on sea ice (Cockell et al. 2002; Cockell & Cordoba-Jabonero 2004), which would probably not have been too thick for light penetration (McKay 2000) and thus synergistic with ozone rise. As cytochrome oxidase probably evolved from an oxygen-independent oxidase previously used in anaerobic respiration, and the earlier part of the respiratory chain was just taken over from anaerobic respiration, aerobic respiration could have evolved rapidly after oxygen levels in mats became appreciable. This is important, as it must have happened fast enough to prevent oxygen rising above the threshold for oxidizing minerals like uraninite (only 10 times that which abolishes mass-independent fractionation: Kasting & Ono 2006) which remained reduced until 2.4Gyr ago.

Finally, why did global glaciations cease for 1.6Gyr? This is unexplained. Had the sun perhaps warmed enough to avoid global freezing altogether? Perhaps biology helped by invasion of land by cyanobacteria that the ozone layer allowed after 2.25Gyr. Vast desert areas and parts of the Arctic are now covered by a thin dark cryptogamic crust of cyanobacteria, lichens and fungi (Johansen 1993). Prominent therein is the black cyanobacterium Scytosiphon, but actinobacteria and proteobacteria also abound (Nagy et al. 2005). They reduce albedo and in the Arctic can warm surfaces by 8–12°C and soil by 4–5°C (Gold 1998). Cyanobacterial blackening of Early-Mid-Proterozoic continents, albeit initially perhaps only half their present extent, possibly saved the day, until quantum evolution again intervened, starting the Neoproterozoic snowball.

In discussing carbonate isotopic levels 2.0–2.3Gyr ago, Hayes & Waldbauer (2006) wrote ‘In all likelihood, the diagenetic alternative has failed to win popularity because an alternative does not appear to be needed.’ Likewise the biogenic methane explanation for the organic carbon negative spike and the increased ¹³C in 2.0–2.3Gyr carbonates remains popular because the necessity for an alternative is widely unrecognized. If the present explanation is found wanting, another must be found, not invoking methanogens, which were almost certainly absent more than 1.5Gyr ago, and probably also more than 0.75Gyr.

The age estimated above of the plant kingdom and the cyanobacterial enslavement that first generated eukaryotic algae is 850Myr, when atmospheric oxygen is inferred to have risen to present levels (Holland 2006). Holland (2006) attributes this rise to burial of fixed carbon and the origins of new types of eukaryote algae and protozoa. I suggest that larger cell sizes and more recalcitrant cell walls (both vegetative and of cysts) of the newly evolved eukaryotic algae and of cysts of protozoa that fed on them were the key causes of greater burial. Thus, cyanobacteria probably first oxygenated the atmosphere to moderate levels. Then eukaryotic algae increased the concentration further soon after their origin. Survival of eukaryotic algae through global freezing is sometimes raised as an objection to or problem for snowball Earth (Hoffman & Schrag 2002). But it is not. Equatorial sea ice was probably thin enough for sub-ice photosynthesis and survival of some unicellular eukaryotic algae (McKay 2000).

Stanier (1970) remarked that chloroplasts ought to have arisen before mitochondria as cyanobacteria long preceded the rise in oxygen that would make mitochondria possible. His environmental determinism reflects widespread misconceptions that environment drives megaevolution and ignores endogenous evolutionary limitations that I think mainly govern its timing. Chloroplasts were not enslaved for ca 2Gyr after cyanobacteria arose for potential enslavement. Enslavement was not possible until after eukaryote cells evolved, which was itself late because of the lateness and improbability of the neomuran revolution and the billions of years that eubacteria were prevented by their murein corset from taking up other cells.

That glycoproteins are independent molecules free to diffuse in the fluid CM, and not covalently bonded into a three-dimensional rigid meshwork as in murein, preadapted the first neomuran to evolve phagocytosis. Substitution of murein by glycoproteins was a prerequisite for evolution of phagocytosis by folding the CM around the prey and forming the endomembrane system by evolving coated vesicles. Food vacuoles could not possibly have formed with a thick murein wall covalently attached to the plasma membrane as in Posibacteria. Having only one bounding membrane, not two as in Negibacteria, was also a prerequisite for phagotrophy and eukaryotes; the independent loss of murein by Planctobacteria could not have led to the origin of phagotrophy if Posibacteria had never evolved, though it seems convergently to have allowed a different type of membrane invagination from other Negibacteria that some misinterpret as related to those of eukaryotes (Fuerst & Webb 1991; Lindsay et al. 2001; Fuerst 2005). If Posibacteria evolved only ca 2Gyr ago, eukaryotes could not have evolved earlier, because of the structural limitations of the negibacterial envelope. As mycoplasmas probably evolved only after eukaryotes gave them a protective habitat (Cavalier-Smith 2002a), the successful transition from posibacterial to neomuran walls was uniquely successful in generating a novel kind of free-living bacterium from a posibacterial ancestor. Its delay of probably 1Gyr after the origin of Posibacteria is unlikely to have been through environmental limitations. It probably simply reflects the extreme difficulty of this change in wall organization. This explains the lack of innovation for the ‘boring billion’ years: all major metabolic innovations except methanogenesis had already evolved; the only possible major evolutionary advance was by fundamental change in the cell envelope, the mechanistically and selectively difficult replacement of murein by glycoprotein.

Seeking to explain why eukaryotes delayed for a billion years before becoming complex (Anbar & Knoll 2002) is pointless if they originated only ca 900Myr ago and there was no such delay. A delay of diversification by snowball Earth till ca 570Gyr is a comprehensible environmental brake. But thereafter protist diversification and the origins of macro-organisms probably took place without significant exogenous limitations, being limited merely by the inherent difficulties of the transitional stages. Three of these were evolutionarily unusually complex cases of quantum evolution: origins of chloroplasts, choanoflagellates, and sponges. Even though origin of plastids was very complex, probably involving tens of thousands of mutations (Cavalier-Smith 1982, 2000), it apparently happened early in bikont and eukaryote evolution and was thus probably not exceptionally difficult, given the prior occurrence of aerobic eukaryotes with good protection against O₂ toxicity, and probably not a major retarding factor.

Once sponges arose, and the spongocoel later became the coelenterate gut, and nematocysts and synaptic transmission evolved, the only remaining major innovation was the anus. Very likely this anal breakthrough ca 550Gyr ago stimulated the long-puzzling Cambrian explosion. Peterson & Butterfield (2005) imply that the anus evolved at the Early Ediacara (570Myr). However, like Conway Morris (2006), I am unconvinced that any Ediacara fauna (Xiao et al. 2005) are bilateria; nor can I place them in any protist or plant group. Most likely they are early animals related to sponges and cnidaria, but fossils have too little detail to show if they belong within either or are a distinct phylum; whichever makes little difference to the suddenness of the animal Cambrian radiation and its contemporaneity with the origin of most protist phyla. The Ediacaran/Cambrian explosion of protists and animals is a fundamental feature of eukaryote evolution, not a preservational artefact (Peterson & Butterfield 2005; Conway Morris 2006). Given the initial bilaterian pattern of a continuous-flow through gut and associated developmental mechanisms, all feasible animal bilateral body plans would inevitably be tried out without further delay by minor developmental modifications to basic animality, leading to an immediate explosive production of all bilateral animal phyla within a few tens of millions of years, each involving rapid quantum evolution and the entry to a major animal adaptive zone. The anus was a prerequisite for intelligence; without it, heads and brains would not have evolved. But after it evolved many primitively sessile groups remained headless. Contrary to Hyman's (1940) postulated multiple decephalizations, I consider that Brachiozoa, bivalves, Bryozoa, crinoids, pterobranchs, were not secondarily beheaded, but primitively acephalic tentaculate feeders, and that heads evolved polyphyletically (but using ancestral bilaterian antero-posterior polarity determinants) in the most cephalized groups (vertebrates, radulate molluscs, arthropods, polychaetes) in lineages that took to active burrowing, creeping or directional swimming rather than sedentary filter feeding or entrapment, the ancestral animal state (sponges) that required only limited neural development (cnidaria).

This burgeoning variation is fascinating from an anthropocentric perspective, but from that of biogeochemistry and global processes the most significant innovation was the later evolution of embryophytes that clothed the land with cooling greenness, enabling complex terrestrial life and greater stability to soil biomes (Wellman et al. 2003). This key innovation was probably delayed merely by the developmental and evolutionary difficulties of the transition from a charophyte alga to an embryophyte, not by environmental factors like the possibly insufficient ozone layer as sometimes speculated; oxygen had probably reached present levels 375Myr before land plants arose and 300Myr before animals (Holland 2006), so frequent suggestions that the origin of either was triggered by oxygen are erroneous. Both directly and by altering cloud cover, forest growth affected global processes but relatively far less than bacterial metabolic innovations.

I do not agree that animal origin was inevitable (Conway Morris 2006). Our presence owes much to rare and unique historical and evolutionary accidents. The neomuran revolution was delayed 2.6Gyr after the origin of life; why could it not have been delayed 4Gyr, or till after the sun runs out of fuel? Conversely, if on another planet cell walls were not first covalently bonded corsets, but made of modules like neomuran glycoproteins linked by weaker forces, then phagotrophy and its ability to generate internal cellular complexity would probably evolve much earlier. Since complex behaviour depends on internal cell complexity and on morphological combinations of basically similar neurons, morphological complexity, not basic chemical and genic complexity, makes intelligent life. Thus, if many different planets evolved life, wall chemistry of the first cells might be the primary determinant of when, if ever, intelligent life might evolve. Were they bonded covalently in three dimensions or by weaker forces allowing flexibility to phagocytose and make dendrites?

At the largest scales, the phrase punctuated equilibrium aptly describes evolution and Earth history, even though stasis between punctuating episodes of quantum evolution is neither a thermodynamic equilibrium nor a true steady state, since evolution of a more modest kind necessarily continues throughout, as do irreversible inorganic processes like continental drift, orogenesis and erosion, slowing the Earth's rotation, and warming of the Sun. Quantum evolution and punctuated equilibria conceptually correct the deep-seated idea of uniformism: the false assumption that rates and types of change are spread uniformly through time and across lineages. Concepts of quantum evolution and punctuated equilibria are essential for a realistic description of what happened in history. But Gould (1994) confused things by contrasting punctuated equilibria not with uniformism, but with gradualism: the view that evolution proceeds gradually by cumulative spreading within populations of numerous relatively small mutations, not by a sudden ‘macromutation’ making a new organ or species overnight. Stated thus, gradualism is generally correct, but it must be qualified, as hybridization and polyploidy together have created thousands of new species overnight (allopolyploidy), albeit only a minority of all speciations and mostly in hermaphrodites. More dramatically, symbiogenesis wrought far more dramatic sudden changes that contradict overstated versions of gradualism, but only ca 6–7 times in all history (notably origins of mitochondria, plants, chromalveolates). But even these do not undermine the philosophic core of gradualism; the initiating changes were not DNA macromutations, but trivial accidents in the now immensely common process of internalization of foreign cells by phagocytosis: rare failures of digestion. Next was a key mutation that made the symbiont useful to the host in one step. This key mutation—for mitochondria, arguably, inserting an inner membrane carrier—radically changed the selective forces on the host, so an almost inevitable cascade of thousands of mutations (many newly beneficial, but some initially harmful) then led to organelle formation (Cavalier-Smith in press). Key mutational innovations that radically change selective forces on an organism are the fundamental initiating factors in quantum evolution; many examples are given in my writings on cell evolution, and above. But they need not be classical macromutations (Goldschmidt 1940) that were rightly anathema to population geneticists because proponents seemed to ignore the need to explain how mutations drastically changing phenotypes could survive and spread within sexual populations.

Some key mutations, however, e.g. murein hypertrophy and loss of OM or loss of murein, are phenotypically macromutations, but could be micro at the genetic level, being in principle causable by a single protein-inactivating nucleotide substitution or gene deletion (and could spread in bacteria which are non-sexual, lacking syngamy). Overstatements of gradualism asserting that megaevolution is simply the accumulation of large numbers of small mutations of the same kind as predominate at lower evolutionary levels are simply wrong and must be abandoned. The clearest counter example involves new gene formation by chimaerization; the mutational fusion of two unrelated genes to code for one, more complex, protein of novel function; many involve novel combinations of protein domains, e.g. origins of archaebacterial reverse gyrase or solenoid/propeller proteins vital for eukaryogenesis (figure 6a legend), but novel domains must sometimes evolve thus themselves. Such statistically rare ‘macromutations’ were crucial for many megaevolutionary episodes, but could be totally absent in many cases of microevolution and speciation. Thus, one cannot simply extrapolate from microevolution to megaevolution. New genes are more commonly made by gene duplication and extreme divergence, also non-essential for microevolution but fundamental to megaevolution (e.g. MreB and FtsZ duplications generated the eukaryote cytoskeleton, figure 6). Ancestors of thousands of novel eukaryotic genes cannot be identified by sequence bioinformatics, which misses the most important part of the quantum evolutionary picture because of extreme episodic sequence divergence during eukaryogenesis. With less marked divergence sister protein paralogues can be put on the same sequence tree, but episodic basal divergences so violate substitution models of uniform or only gradually changing divergence that most paralogue trees are very confusing, and seriously misleading conclusions can be drawn (Cavalier-Smith 2002b, 2006a). Mutations of large phenotypic effect are not Goldschmidt's (1940) macromutations by wholesale genomic reorganization, but simple gene mutations, duplications, deletions and chimaerizations. Transposable elements can cause major genome reorganization, but it is typically dissociated from and irrelevant to major phenotypic change (Cavalier-Smith 1993). The largest genomic reorganization in history converted bacterial circular single replicons with numerous operons (cotranscribed gene clusters) and virtually no non-coding DNA to eukaryotic linear chromosomes with multiple replicons and separately transcribed genes during eukaryogenesis. This caused no dramatic phenotypic changes, but was an adaptively trivial consequence of two of them (origins of mitotic spindle and nuclear envelope; Cavalier-Smith 1993, 2005). Goldschmidt's view that big phenotypic changes need big genetic causes was wrong; but in focusing on how mutations effect radical phenotypic change he was far ahead of others.

Population genetic paradigms that focus on changing allele frequencies for existing genes, or phylogenetic trees that chart their divergence, sidestep many key evolutionary innovations and like sequence bioinformatics see only part of the evolutionary picture. These approaches, though vital, must be supplemented. Understanding organismal evolution requires focus on specific phenotypic effects of different types of mutation in specific kinds of organisms, and their ecological consequences, not just treating mutations as generalized abstractions for mathematical manipulation. To comprehend megaevolution of macro-organisms, we must understand developmental biology and how it makes new forms (here genetically small mutations can have huge phenotypic effects; they should not be unthinkingly dismissed as ‘hopeful monsters’—origins of phyla are exceedingly rare events that may involve inherently unlikely intermediates, and key mutations acting early in embryogenesis to reform body plans grossly). Understanding microbial megaevolution requires cell biology—not just biochemistry. To link either with palaeontology we must have the correct phylogenetic tree and map it properly onto the fossil record. Attempts to reconstruct Earth history in a phylogenetic vacuum ignore important constraints, causing mistakes.

Concluding remarks. Gould (1994) also confused us by linking punctuated equilibrium to a mistaken macroevolutionary theory grossly overemphasizing species selection and lineage sorting. Therefore, I must stress that I am neither attacking nor ignoring population genetics or sequence phylogenetics, or the basically sound Neodarwinian paradigm, in advocating the merits of concepts of punctuated equilibrium, quantum evolution, and—especially—key mutations that dramatically change selective forces (and sometimes unusually substantially change phenotypes). These and other special requirements for understanding unique historical events that cannot be treated statistically were part of the synthetic ‘theory’ as seen by Simpson (1953), one of its leading architects, but have been partially forgotten and need restating each generation. Quantum evolution and punctuated equilibria are not mechanisms to replace mutation and selection, but descriptive devices to characterize large-scale evolutionary patterns. Truly synthetic evolutionary biology requires a balanced synthesis of reconstructed historic patterns and mechanistic explanations of the changes. Its must draw on and attempt to integrate all relevant disciplines, none preeminent—not easy in an age of overspecialization. My efforts here to portray a three-stage view of life with only two really major punctuations in more than 3Gyr may stimulate others to do better. A possible simplification to just two stages, only hinted at above, arises if the origin of life was actually much closer to the glycobacterial revolution than widely assumed, e.g. ca 2.9–3.1Gyr, making the origins of Chlorobacteria and Hadobacteria only slightly before that of glycobacteria. If so, the glycobacterial revolution would simply be a later part of one explosive early negibacterial radiation, as argued previously (Cavalier-Smith 1987a, 1992, 2002b). In that case the neomuran revolution would be the only really major punctuation in all evolution. Better understanding of the Early Archaean is needed to decide.

I thank NERC for research grants, and the Canadian Institute for Advanced Research and NERC for fellowship support.