Introduction

Recent genome sequencing suggest that bacteria and archaea with streamlined genomes are often numerically dominant in nature, with marine and freshwater bacteria as prominent examples (Hahn et al., 2012; Garcia et al., 2013; Swan et al., 2013; Eiler et al., 2014; Giovannoni et al., 2005a, 2005b). Streamlined genomes are characterized by small size (<1.5 Mb), low guanine cytosine (GC) content (<40%) and short intergenic regions framing a restricted set of genes (Giovannoni et al., 2014). The few and typically short intergenic regions, combined with a reduction in complexity of regulatory circuits, has posed the question about the capability of such streamlined cells and their populations to respond and adapt to environmental cues. Such genome minimalism has also been linked to unusual nutritional requirements, including auxotrophy for some vitamins and essential amino acids (Giovannoni et al., 2005b; Tripp et al., 2008, 2009; Schwalbach et al., 2010; Carini et al., 2013, 2014). Other apparent features are scarcity or absence of gene duplications, extrachromosomal elements and remnants of genes and phages (Giovannoni et al., 2014). The proposed selective drive for this minimalism is efficient use of limiting resources (for example, C, N, P) in nutrient-poor systems (Giovannoni et al., 2014), which can cause genome reduction when combined with large effective population sizes (Lynch, 2007).

These features, particularly the reduced genetic repertoire of individual cells, raises the question of how such organisms can maintain sufficient metabolic flexibility to radiate or possibly cross biome boundaries. A potential explanation is provided by recent genomic analyses of Prochlorococcus where bacterial populations with streamlined genomes were shown to contain considerable sequence variation in shared genes and to harbor a high level of heterogeneity in gene content (Kashtan et al., 2014). A combination of point mutations and homologous recombination were suggested to be responsible for most of the allelic diversity, whereas homologous and non-homologous recombination were implicated in the loss and gain of genes. The resulting combined diversity of streamlined genomes in these large populations thus could provide the means to cross biome boundaries, such as moving from oceans to freshwaters, with dramatic differences in solutes, salinity and resources, thereby forming a strong ecological and evolutionary barrier.

Similar to larger organisms, there is little overlap in the taxonomic composition of marine and freshwater microbial assemblages (Suzuki and DeLong, 2002; Logares et al., 2009; Newton et al., 2011). Nevertheless, phylogenetic analyses have uncovered some surprisingly close relatives among marine and freshwater microbiota (Logares et al., 2009; Newton et al., 2011). Both marine and freshwater bacterioplankton communities are frequently dominated by alphaproteobacteria of the SAR11 lineage (Morris et al., 2002; Rusch et al., 2007; Eiler et al., 2009; Newton et al., 2011; Salcher et al., 2011; Heinrich et al., 2013). Members of the marine SAR11 clades are characterized as slow growing oligotrophs (Giovannoni et al., 2014), and they are believed to have a central role in the global cycling of both carbon, sulfur and associated elements (Malmstrom et al., 2004). Among their adaptations to oligotrophic marine environments are proteorhodopsin-mediated phototrophic potential (Giovannoni et al., 2005a), reliance on externally supplied reduced sulfur (Tripp et al., 2008), auxotrophy for essential amino acids (Tripp et al., 2009) and lineage-defined metabolic restrictions for a wide range of labile low molecular weight organic metabolites, including C1 compounds (Schwalbach et al., 2010; Sun et al., 2011; Carini et al., 2013). Freshwater SAR11 are phylogenetically distinct from their marine relatives and group into the uncultivated clade named ‘LD12’ (Zwart et al., 1998). LD12 is recognized as a globally distributed freshwater group (Newton et al., 2011) that can make up 20–40% of the total bacterioplankton in lakes (Salcher et al., 2011; Heinrich et al., 2013), and isotope tracer experiments have verified that they can grow heterotrophically (Salcher et al., 2011).

Here we compared the metabolic potential of 10 sequenced single amplified LD12 genomes obtained from freshwater lakes (Zaremba-Niedzwiedzka et al., 2013) with 18 marine SAR11 genomes, 4 single amplified genomes and 14 from isolates and originating from different coastal and oceanic habitats (Giovannoni et al., 2005b; Tripp et al., 2009; Grote et al., 2012; Viklund et al., 2013). Previous comparisons of these genomes have revealed lower recombination frequencies in the freshwater LD12 genomes (Zaremba-Niedzwiedzka et al., 2013) but not detailed the metabolic changes associated with the transition from marine to freshwater environments. Our analyses reveal a reduction of metabolic diversity and lower pathway redundancy in freshwater LD12 compared with marine SAR11, combined with a rewiring of the central metabolism. We propose that the process leading up to the marine–freshwater transition in SAR11 was more likely facilitated by recombination in the large and diverse populations of the marine ancestors than by gene family expansion or horizontal acquisition of metabolic modules from non-SAR11 lineages. Thus our results unveil an unknown chapter in the evolutionary history of these ubiquitous and successful bacterial groups in aquatic environments.

Methods

Genome data sets

The draft assemblies of the single amplified genomes (SAGs) and genome-sequenced SAR11 isolates were obtained from intergrated microbial genome (IMG)/ER (Markowitz et al., 2012). Protein sequences for the recently described pathway of dimethylsulfonipropionate (DMSP) demethylation were retrieved for Ruegeria pomeroyi and HTCC1062 as identified by Reisch et al. (2011). These were used as queries in BLASTP searches against the 18 marine and 10 freshwater SAR11 genomes.

Genome completeness

Genome size and completeness were estimated using a conserved single-copy gene (CSCG) set of SAR11 isolates (Markowitz et al., 2012). The set consists of 357 CSCGs that were found to occur only once in the genomes by analysis of an abundance matrix based on hits to the protein family (Pfam) database (Punta et al., 2012). Hidden Markov models of the identified Pfams were used to search all single amplified genome assemblies by means of the HMMER3 software (Eddy, 2011). Resulting best hits above the trusted cutoff (TC field as provided in the HMM files from Pfam) were counted, and the completeness was estimated as the ratio of found CSCG to total CSCGs in the set. The estimated complete genome size was then calculated by dividing the estimated genome coverage by the total assembly size.

Phylogenetic reconstructions

Orthologous genes were identified by running a BLASTP similarity search using the set of all marine and freshwater genes as the query and the database. Subsequently, we applied the Markov cluster algorithm (Enright et al., 2002) to the pairwise matrix of bit scores. Out of the 3598, we selected the clusters that had at least one gene pertaining to each of the seven families (‘Ia.1’, ‘II’, ‘Ia.2’, ‘V’, ‘IIIa’, ‘Ic’, ‘IIIb/LD12’) and no more than 30 genes in them, leaving us with 518 clusters. Genes were then aligned with muscle (Edgar, 2004), masked with gblocks (Castresana, 2000) and trees were built with RaxML (Stamatakis, 2006) using the ‘PROTGAMMAJTT’ model with an automatically determined number of quick bootstraps, thus resulting in one tree per cluster. At this point, we removed one cluster as it contained only gaps.

For the master ribosomal protein tree, we extracted 50 ribosomal proteins and subsequently selected only those that had at least one representative in each SAR11 subgroup (N=40). After alignment of randomly selected representatives from each SAR11 subgroup, we concatenated the alignments to form the so-called ‘master ribosomal alignment’. A RAxML tree was constructed using the ‘PROTGAMMAJTT’ model with an automatically determined number of quick bootstraps. The code produced for this analysis is available under an MIT license at http://github.com/limno/ld12.

Closest hits and gene family expansions

The first hit for every freshwater gene is obviously itself (though some genes have no other hits). Searching for the first hit that is not from any of the freshwater genomes, we extracted its taxonomy. If this top-hit-non-fresh is a marine SAR11 genome, we resolve it to one of the six marine clades; if outside, we resolve it to broad taxonomic groups. Duplication events (gene family expansions) were defined by two hits associated with the same freshwater genome and ranked prior to the first non-freshwater SAR11 hit.

Results

General features and SAR11 phylogeny

The freshwater and marine SAR11 genomes were extracted from public databases (Table 1). The genomes are in varying stages of completion, with some corresponding to a single circular chromosome whereas others are incomplete and consist of 50–150 contigs. The completeness of the genomes amplified from single cells was estimated to 24–86% using a conserved single-copy gene set of SAR11 genomes (Markowitz et al., 2012), with estimated genome size of the freshwater genomes ranging from 1.1 to 1.4 Mb compared with marine genomes ranging from 1.1 to 1.6 Mb. In addition to these small differences, the sampling sizes for subgroups II and IIIa are too small to determine with certainty whether differences in genome size are associated with subgroups and habitats. Unifying features are low GC (29–30%), a shared amino-acid usage and high percentage of predicted coding nucleotides (Table 1).

Table 1 Description of the genome data
Full size table

Phylogenetic analysis of the selected genomes using concatenated alignment of 40 ribosomal proteins revealed 7 clades (Figure 1), consistent with previous analyses (Grote et al., 2012; Viklund et al., 2013; Zaremba-Niedzwiedzka et al., 2013; Thrash et al., 2014). In this study, we have not included any other alphaproteobacterial species, which prevents a discussion about the position of subgroup V, represented by genome HIMB59, in the alphaproteobacterial tree. The disagreement about its phylogenetic placement was discussed in quite some detail in a previous publication (Viklund et al., 2013), but this does not have any implications for our conclusions. Suffice it to conclude that HIMB59 is used as an outgroup to the other SAR11 genomes examined here. By using HIMB59 as an outgroup, we inferred phylogenetic trees from 518 orthologous protein clusters that had at least one representative in each of the 7 subclades. Out of these, only a minor proportion (N=88) of the protein trees are strictly coherent with regards to collapsing the 7 main clades (that is, all genes from a given branch are monophyletic). For a third of these (N=33), the branching patterns agree with the ribosomal protein tree, and the branching pattern between the freshwater LD12 (SAR11 clade IIIb) and clade IIIa is observed in 388 out of the 518 trees, albeit with different statistical significance (representing 75% of all trees as indicated in Figure 1).

Figure 1
figure 1

Evolutionary relationships between marine and freshwater SAR11 genomes. Unrooted maximum likelihood phylogenetic tree of concatenated ribosomal protein sequences from single-cell genomes and isolate genomes of the SAR11 clade. In addition to bootstrap values as inferred by maximum likelihood, the second (large and bold) number at the branching points show proportion of protein trees with corresponding branching pattern. These proportions were generated by using the Robinson–Foulds metric summarizing 518 orthologous protein trees that had at least one representative in each of the seven subclades (for details, see Supplementary Table S1). Freshwater SAR11 (subgroup IIIb/LD12) are indicated in red, whereas marine SAR11 subgroups in other colors.

Full size image

Comparing this with the number of cases where the bootstrap support was >80% at this particular branch, freshwater LD12 branched closest with IIIb in 109 trees, whereas in 9 trees LD12 was closest with another SAR11 subgroup. These nine trees were inferred from individual protein alignments involved in amino-acid biosynthesis and conversion (Figures 2a–c), chemoorganotrophic energy acquisition (Figures 2d–f), carboxylic acid membrane transport (Figures 2g and h) and DNA methylation (Figure 2i). This proportion of non-matching tree topology, either 9% or 25% depending on bootstrap cutoffs (Figure 1; see a report detailing the mismatches and the Robinson–Foulds metric in Supplementary Table S1), implies alternative evolutionary histories of various gene families in SAR11.

Figure 2
figure 2

Nine protein trees with evolutionary inconsistency when compared with the ribosomal tree. (a–i) Unrooted phylogenetic trees of orthologous protein clusters with alternative tree topologies (that is, sister clade to IIIb/LD12 is not IIIa) when compared with the ribosomal tree (Figure 1).

Full size image

Two trees are largely consistent with the ribosomal protein tree with the exception of the branching orders of subgroups II and III (Figures 2b and d). Another three trees have more serious inconsistencies, such that subgroups Ia and Ic are not sister clades and that the two members of subgroup IIIa do not cluster (Figures 2a, c, e, g and i). The most dramatic inconsistency is observed for the tree inferred from the tripartite ATP-independent periplasmic transporters (TRAP transporters) (Figures 2h and i). In these trees, all members of subgroup LD12 and two genomes of subgroup Ia are separated from all other genomes by long branches. This could be indicative of paralogous gene families or increased evolutionary rates for this gene in LD12. This is potentially interesting as TRAP transporters may be involved in the metabolite flux into the tricarboxylic acid (TCA) cycle through the import of malate (Adnan et al., 2015).

Apart from the incongruence of protein trees, patterns in gene synteny (Supplementary Figure S1) and functional annotations (Supplementary Figure S2) suggest that the analyzed SAR11s group into six (plus outgroup V) distinct clades, basically confirming shared gene phylogenies (Grote et al., 2012; Viklund et al., 2013; Zaremba-Niedzwiedzka et al., 2013; Thrash et al., 2014). The distinction in gene sets between marine and freshwater genomes was shown by permutational multivariate analysis of variance (P<0.001) for four annotation subsystems (pfam, tigrfam, COG, KO) and also by dendrograms based on bit scores retrieved from pairwise blastn searches (Supplementary Figure S3). In addition, each of these subgroups is preferentially retrieved in a distinct habitat, indicating that they correspond to ecological coherent units (Rusch et al., 2007; Andersson et al., 2009; Carlsson et al., 2009; Newton et al., 2011; Heinrich et al., 2013; Thrash et al., 2014).

Energy and carbon metabolism

We identified genes encoding a complete TCA cycle (Figure 3) and a proteorhodopsin when combining the 10 partial freshwater SAR11 genomes. This led us to hypothesize that, analogously to their marine siblings, freshwater SAR11 are heterotrophs capable of supplementing heterotrophic growth on organic carbon with light-mediated ATP production (Martinez-Garcia et al., 2012) when carbon starvation limits respiration (Steindler et al., 2011). Still, there are conspicuous and consistent variations in gene content related to aerobic chemoorganoheterotrophy as already highlighted by previous metabolic reconstruction restricted to marine SAR11 (Tripp et al., 2008, 2009; Schwalbach et al., 2010; Sun et al., 2011; Grote et al., 2012; Carini et al., 2013, 2014). Here we expand on these previous observations by showing that freshwater genomes encode all genes of the Embden–Meyerhof–Parnas (EMP) pathway, whereas most marine SAR11 representatives encode all genes of the Entner–Doudoroff (ED) pathway (Figure 3).

Figure 3
figure 3

Central carbon metabolism and other relevant metabolic pathways identified in SAR11 genomes. Glycogenesis, TCA cycle and glyoxylate shunt are shown in the center of the plot with adjunction pathways to either side such as the pentose phosphate and ED pathway. The color of the arrows indicates genes encoded in at least one genome of each SAR11 subgroup (see legend for details). The presence and absence of genes was determined within the IMG system based on automated and manual annotations. Detailed results are in Supplementary Tables S6–S9.

Full size image

Another striking difference is the absence of the glyoxylate shunt (isocitrate lyase (aceA), malate synthase (malG)) and anapleurotic enzymes (phosphoenol pyruvate carboxylase (ppc); see Figure 3) in subgroups IIIa and IIIb, pointing to major differences in carbon and energy-acquisition strategies for these SAR11 clades. The glyoxylate shunt bypasses two steps in the TCA cycle where carbon dioxide is released concomitant with a production of ATP and NADH but instead enable economic use of carbon atoms at the expense of energy acquisition. It essentially allows microorganisms to use simple carbon compounds as a carbon source when complex nutrients such as glucose are not available. In the absence of available carbohydrates, the glyoxylate cycle would, for example, enable synthesis of carbohydrates needed for cell-wall assembly from lipids via acetate supplemented with inorganic carbon fixed via anapleurotic reactions (Moran and Miller, 2007).

There is evidence that marine SAR11 can oxidize several small organic carbon compounds that are potentially very abundant in marine systems, in order to satisfy cellular energy demands (Tripp et al., 2009; Schwalbach et al., 2010; Sun et al., 2011). Several genes and pathways involved in the oxidation of C1 and methylated compounds to CO2, previously identified in marine SAR11 genomes (Sun et al., 2011), appear to be missing in the freshwater single amplified genomes (see Supplementary Material for details). At the same time, the ability to use other common low molecular weight carbon substrates are maintained throughout the entire SAR11 lineage, one example being glycolate oxidase (glcDEF) which enable metabolic use of the abundant exudate glycolate, which is released in large amounts by nutrient-limited phytoplankton as an energy overflow mechanism (Bertilsson and Jones, 2003).

While inspecting the completeness of vitamin biosynthesis pathways among the freshwater genomes, we found no major differences when comparing our results with marine genomes (for details on the presence of vitamin biosynthesis homologs, see Supplementary Material). Confirming previous findings (Tripp et al., 2008), we could not identify any sulfite reductase in marine SAR11 genomes nor could this gene be detected in any of the freshwater representatives, implying that freshwater SAR11 also rely on external supply of reduced sulfur similar to their marine siblings. This auxotrophy for reduced sulfur seems to be a common feature in abundant freshwater bacteria as revealed by genome reconstructions (Garcia et al., 2015). To access reduced sulfur, marine SAR11 are capable of demethylating the abundant marine molecule DMSP. However, homologs of the DMSP pathway genes could not be identified in the 10 freshwater (LD12) or clade IIIa genomes. In addition, we could not identify any homologs to adenylylsulfate reductase, sulfate permease and sulfate exporter in the freshwater single amplified genomes, making sulfur-containing amino acids the most likely reduced sulfur source.

Osmolytes, such as DMSP, glycine betaine, proline, mannitol, taurine and spermidine, are produced in high amounts by phytoplankton and other marine microplankton, which use these compounds to maintain their intracellular osmotic balance. These organic compounds can build up intracellular concentrations as high as 100–530 mM in some phytoplankton species (Keller et al., 1999; Sunda et al., 2007). Analyses of marine SAR11 genomes (Giovannoni et al., 2005b) have revealed numerous ABC transporters for organic osmolytes that are common in the marine environment (see Supplementary Material for detail on transporters). An ability to use glycine betaine and sulfur-containing osmolytes may also release organisms from an inability to synthesize certain amino acids and reduced (organic) sulfur compounds de novo (Giovannoni et al., 2014). Although glycine and serine biosynthesis is enabled by 3-phosphoglycerate dehydrogenase (serA) in freshwater SAR11 (for more detail on amino-acid biosynthesis, see Supplementary Material), organosulfur acquisition seems to be restricted to amino acids, such as methionine and cysteine. This is not surprising considering that osmolyte diversity is restricted to non-sulfur compounds such as sucrose and trehalose in freshwater phytoplankton (Batterton and van Baalen, 1971; Joset et al., 1996; Page-Sharp et al., 1999). Still, no known hydrolases to use sucrose and trehalose could be identified in any of the inspected freshwater SAR11 genomes. As indicated by the presence of TRAP system, a feature shared with subgroup IIIa, freshwater SAR11 most likely use low-molecular-weight carboxylic acids.

Interestingly, two TRAP system-related orthologous gene clusters showed phylogenetic patterns highly divergent from the ribosomal protein trees (Figures 2g and h), suggesting that this transporter system has been widely exchanged within the SAR11 clade.

The marine and freshwater SAR11 pangenome

In addition to phylogenies (see Figures 1 and 2), we explored gene content variations among the SAR11 genomes. Orthologous protein clusters were formed by Markov cluster algorithm (Enright et al., 2002) from pairwise BLASTP searches using 23 545 predicted SAR11 proteins. Of the 3589 total orthologous clusters identified in the 28 SAR11 genomes, 1467 were present in at least one freshwater SAG, and 368 of these were exclusive to freshwater genomes. Most of the 8332 predicted genes from the freshwater SAGs were most closely related to marine subgroup IIIa (Figure 4a). Out of the 368 orthologous clusters (including 720 genes) exclusive to freshwater SAR11, 59 clusters (including 86 genes) had no hit against bacterial and archaeal subject sequences in refseq. Altogether, 94 of the freshwater-exclusive genes (26%) were most closely matching alphaproteobacteria, whereas only 7 (2%) were most closely related to genes from marine SAR11 (Figure 4b). Although around a sixth of the predicted genes from the freshwater SAGs were most closely related to non-marine SAR11, it cannot be concluded that these genes were acquired horizontally from outside the SAR11 clade. The reason is that the origin and flux of these genes are uncertain when considering that protein diversity was highly undersampled in both the marine and freshwater branch of SAR11, highlighted by the rarefaction analysis (Figure 4c). The large diversity in protein families is further imposed by 6273 protein families in the marine and 2320 in the freshwater SAR11 pangenome as estimated by non-parametric diversity approximations (Chao, 1987).

Figure 4
figure 4

Closest blast hits against refseq and rarefaction curves of orthologous SAR11 protein clusters. (a) Bar charts revealing top hits from BLASTP against all available genomes in refseq. Hits are classified based on their taxonomic affiliation with higher taxonomic levels removed from lower taxonomic levels (that is, SAR11 hits are removed from the alphaproteobacterial bin, with alphaproteobacterial hits removed from the proteobacterial bin and so on). Black bars indicate hits of genes specific to freshwaters as inferred by orthologous clustering. (b) Detailed representation of freshwater-specific orthologous protein clusters and their closest hits classified on their taxonomic affiliation with higher taxonomic levels removed from lower taxonomic levels. (c) Results from a rarefaction analysis of orthologous protein clusters after normalization to account for the partial single-cell genomes.

Full size image

BLASTP searches were also used to infer gene family expansions unique to the freshwater SAR11 population by applying the following criteria: two hits associated with the same genome were ranked prior to the first non-freshwater SAR11 protein. This identified at most four orthologous clusters per freshwater genome, thus likely originating from recent duplication events as they were also adjacent to each other in the genome. These orthologous clusters were exclusively annotated as hypotheticals or cell communication systems (for details, see Supplementary Table S5) and are likely located in hypervariable regions. Such hypervariable regions are apparent in a composite freshwater SAR11 genome and seem to host genes for the biosynthesis of lipopolysaccharides (Zaremba-Niedzwiedzka et al., 2013). The comparative features of these 28 genomes point to the existence of large gene frequency variations in SAR11 populations.

Discussion

As shown previously (Zaremba-Niedzwiedzka et al., 2013), phylogenetic reconstructions strongly suggest that the freshwater SAR11 clade originates from a population of marine ancestors with streamlined genomes considering the branching pattern leading to LD12. However, we cannot preclude with certainty that IIIa resembles a marine ancestor of LD12 or that IIIa originated from a freshwater ancestor. Nevertheless, we favor the former scenario as a freshwater origin would require two independent transitions, whereas one is sufficient for a marine SAR11 ancestor. In addition, the larger estimated marine gene set when compared with the freshwater equivalents would have required an acquisition of genes independent from the freshwater SAR11. The likelihood for such a vast gene acquisitions is challenged by large population sizes that provide low probabilities of new genes arising and being fixed in streamlined genomes. Likewise, a common streamlined ancestor is implied by freshwater and marine SAR11 clades sharing many typical features of streamlined cells, such as the small cell size (Giovannoni et al., 2005b; Salcher et al., 2011), auxotrophy, small genome size and low GC content. With no known exceptions, this argues against non-streamlined ancestoral populations at the main branching points.

Besides the shared reductionism, both marine and freshwater SAR11 are predicted to host a typical electron transport chain, a complete TCA cycle and a proteorhodopsin system for phototrophic energy acquisition (Giovannoni et al., 2005a; Martinez-Garcia et al., 2012). Still, there are conspicuous variations in gene content related to aerobic chemoorganoheterotrophy between marine and freshwater SAR11, emphasizing functional adaptations beyond osmoregulation as a necessity to overcome the marine–freshwater boundaries. This is in agreement with recent findings from shotgun metagenomic analyses, where quantitative differences related to respiratory and glycolytic pathways were observed in the gene pools of marine and freshwater communities (Oh et al., 2011; Dupont et al., 2014; Eiler et al., 2014). In particular, the changes in glycolysis combined with the apparent loss of the glyxoylate shunt and the capacity to oxidize a broad array of C1 compounds point to differentiation in heterotrophic substrate use between marine and freshwater SAR11, with the marine representatives more extensively relying on phytoplankton-derived osmolytes, such as proline, betaine glycine and DMSP. Instead, freshwater SAR11 feature a complete EMP, enabling efficient energy acquisition from C6 and C5 sugars combined with the TRAP system to use low-molecular-weight carboxylic acids known to be produced in abundance by photochemical degradation of humic substances in freshwaters (Bertilsson and Tranvik, 2000). Such tuning toward locally available organic substrates is likely linked to observed differences in abundance patterns in relation to environmental conditions in the respective biomes (Eiler et al., 2009; Salcher et al., 2011) and the formation of ecotypes within the SAR11 clade (Carlsson et al., 2009; Vergin et al., 2013).

The reliance on either EMP or ED can be predicted to change the energy status and the redox balance of the cell. The ED pathway (Entner and Doudoroff, 1952; Conway, 1992) has a net yield of 1 ATP, 1 NADH and 1 NADPH for each glucose molecule, whereas the EMP has a net yield of 2 ATP and 2 NADH for each processed glucose molecule. Besides the distinct yields of ATP and NAD(P)H, the EMP and ED pathways result in a range of different metabolic intermediates for each molecule of sugar consumed (Neidhardt et al., 1990). It is remarkable that, while some bacterial genomes possess complete and coexisting EMP and ED pathways (suggested for HIMB59), others have one or the other, and there are also SAR11 genomes that only seem to host incomplete pathways (Grote et al., 2012). A flux of genes involved in central metabolism is further implied by the incongruence in tree topology of the glyceraldehyde 3-phosphate dehydrogenase (Figure 2d), the enzyme that catalyzes the sixth step of glycolysis, a phosphorylation reaction coupled to oxidation.

In addition to the metabolic distinction between marine and freshwater SAR11, our analyses revealed that the gene pool of marine and freshwater SAR11 are highly diverse and undersampled. The notion that marine SAR11 populations appear to have rather diverse genomes was previously observed in marine metagenomes (Wilhelm et al., 2007) and was also shown for Prochlorococcus as another abundant and widespread marine bacterium with small genomes (Kashtan et al., 2014). Overall, we show that SAR11 subgroups share a core set of genes comprising approximately half of the orthologous genes in each genome, independent of whether orthologous genes are defined based on annotations or clustering using sequence similarity. As expected, the core genes encode functions defining the SAR11 clade as an organo-heterotrophic group with varying potential for photo-autotrophy, whereas the flexible genome provides functions that confer selective advantages under very different conditions. This modular structure of genes that confer the ability to take up and metabolize specific organic molecules imply metabolic flexibility and an adaptive tuning to resources.

This functional tuning within SAR11 is facilitated by high rates of recombination (Vergin et al., 2007) among even distantly related marine groups fostering a global SAR11 population where multiple ecotypes with high genetic linkages coexist. A dynamic gene pool in SAR11 is implied by gene content variations, alternative tree topologies of the orthologous clusters shared among all SAR11 subgroups and, as shown previously, by high ratios of homologous recombination to mutation rate in marine SAR11 (Vergin et al., 2007). Even if extremely high recombination has been suggested for marine SAR11 (Vergin et al., 2007; Zaremba-Niedzwiedzka et al., 2013), we argue that the exchange of genetic material seems to be mainly restricted to members within the SAR11 clade, as suggested by a common SAR11 ancestor for the orthologous genes that were present in all marine and freshwater SAR11 genomes. As the marine gene pool is undersampled, we argue that one cannot make any inferences on horizontally acquired genes from outside the SAR11 clade to facilitate the marine–freshwater transition. A likely scenario is that, as SAR11 gradually became adapted to life at low salinities, recombination events with divergent SAR11 in the former marine environment became less likely and eventually mutations overwhelmed recombination, as indicated previously (Zaremba-Niedzwiedzka et al., 2013).

Resulting gene content differences that affect biosynthesis and transport functions are not restricted to specific regions but are instead embedded in different genomic contexts, as implied by the large-scale synteny variations in the SAR11 genomes. Besides these ecotype-defining and broad-scale genomic variations, hypervariable regions containing genes for the biosynthesis of lipopolysaccharides were previously identified in both marine and freshwater SAR11 clades (Grote et al., 2012; Zaremba-Niedzwiedzka et al., 2013). This fits the previously proposed concept of low frequency genes being implicated in evolutionary responses to local and variable biotic interactions, such as competition and phage predation (Cordero and Polz, 2014).

Adaptive selection that can lead to niche partitioning and the formation of novel ecotypes in bacterial populations is assumed to depend on a wide array of traits, including large population size (Levin and Bergstrom, 2000), a dynamic gene pool facilitated by genetic exchange (Cohan, 2001; Popa et al., 2011) and gene family expansion by duplications (Alm et al., 2006; Serres et al., 2009), the small size of recombined fragments (Zawadzki and Cohan, 1995) and simplicity and modularity of metabolic features (Doyle et al., 2007; Lawrence, 1999). In the long run, these mechanisms can allow lineages to take major leaps into novel environments or even new biomes, even when populations are made up of individuals with streamlined genomes. For streamlined bacteria in particular, we argue that reshuffling of metabolic modules related to energy and carbon metabolism seem to facilitate such adaptive radiation, as shown in the case of metabolic tuning during the marine–freshwater transition. Following such expansion, we imply from our and previous genome reconstructions (Zaremba-Niedzwiedzka et al., 2013) that gene exchange barriers arise, which fix only a small fraction of the original pangenome in the novel biome. As such, freshwater SAR11 seem to be less prone to adaptive radiation, at least if we assume that the genetic structure of global populations controls selection efficacy and ability to adapt in free-living streamlined bacteria.