Strong Purifying Selection Is Associated with Genome Streamlining in Epipelagic Marinimicrobia

Abstract

Marine microorganisms inhabiting nutrient-depleted waters play critical roles in global biogeochemical cycles due to their abundance and broad distribution. Many of these microbes share similar genomic features including small genome size, low % G + C content, short intergenic regions, and low nitrogen content in encoded amino acid residue side chains (N-ARSC), but the evolutionary drivers of these characteristics are unclear. Here, we compared the strength of purifying selection across the Marinimicrobia, a candidate phylum which encompasses a broad range of phylogenetic groups with disparate genomic features, by estimating the ratio of nonsynonymous and synonymous substitutions (dN/dS) in conserved marker genes. Our analysis reveals that epipelagic Marinimicrobia that exhibit features consistent with genome streamlining have significantly lower dN/dS values when compared with their mesopelagic counterparts. We also found a significant positive correlation between median dN/dS values and % G + C content, N-ARSC, and intergenic region length. We did not identify a significant correlation between dN/dS ratios and estimated genome size, suggesting the strength of selection is not a primary factor shaping genome size in this group. Our findings are generally consistent with genome streamlining theory, which postulates that many genomic features of abundant epipelagic bacteria are the result of adaptation to oligotrophic nutrient conditions. Our results are also in agreement with previous findings that genome streamlining is common in epipelagic waters, suggesting that microbes inhabiting this region of the ocean have been shaped by strong selection together with prevalent nutritional constraints characteristic of this environment.

Bacteria and Archaea play key roles in marine biogeochemical cycles and are a dominant force that drives global nutrient transformations (Azam et al. 1983; Falkowski et al. 2008). Our understanding of microbial diversity in the ocean has been transformed in the last few decades due to the discovery of several marine microbial lineages that are among the most numerically abundant life forms on Earth (Giovannoni and Stingl 2005). Work on some of these abundant lineages succeeded in culturing representatives that could then be studied extensively in the laboratory, such as Prochlorococcus marinus (Chisholm et al. 1992) and heterotrophic bacterioplankton belonging to the Pelagibacteriales (Rappé et al. 2002), and Roseobacter groups (Luo and Moran 2014), but many other dominant microbial lineages have not been brought into pure culture and require cultivation-independent methods for analysis (DeLong and Karl 2005).

Previous research of Prochlorococcus marinus and Pelagibacter ubique genomes provided some of the earliest insights into the ecology and evolution of these dominant planktonic microbial lineages (Rocap et al. 2003; Giovannoni et al. 2005). It was quickly noted that both groups had small genomes that contained short intergenic regions and encoded among the fewest genes of any free-living organism (Giovannoni et al. 2005). These characteristics were explained through the proposed theory of genome streamlining, which states that genome simplification is an adaptation to consistently oligotrophic conditions, and that the loss of unnecessary genes and their corresponding transcriptional, translational, and regulatory burdens is advantageous (Giovannoni et al. 2014). Genome streamlining theory is supported by the observation that many streamlined genomes also have lower % GC content and subsequently contain fewer codons encoding nitrogen-rich amino acids (Grzymski and Dussaq 2012; Mende et al. 2017), which is expected to be advantageous in nutrient-depleted conditions found in the open ocean (Giovannoni et al. 2014). Genome streamlining therefore corresponds to multiple characteristics, and recent cultivation-independent studies have confirmed that many of them are present in the genomes of a variety of marine lineages in addition to Prochlorococcus and Pelagibacter (Dupont et al. 2012; Ghai et al. 2013; Swan et al. 2013; Luo et al. 2014; Getz et al. 2018), suggesting that common evolutionary drivers shape diverse bacterioplankton groups in the ocean.

Although the term “genome streamlining” implies adaptation under oligotrophic nutrient conditions, it remains a possibility that these genomic signatures are nonadaptive or potentially the result of genetic drift (Batut et al. 2014). For example, it has long been known that many endosymbiotic bacteria contain small genomes with short intergenic regions and low % GC content, but in these cases a small effective population size (Ne) and correspondingly high genetic drift are likely responsible for these features (Charlesworth 2009; Kuo et al. 2009). Although it remains unlikely that marine free-living bacteria have small effective population sizes comparable to those of endosymbiotic bacteria, it has been argued that population bottlenecks in the distant evolutionary past of some marine lineages may be responsible for aspects of their present genomic architecture (Luo et al. 2017). Moreover, recent work has also shown that weakly deleterious mutations and low recombination rates can substantially lower the efficacy of purifying selection in bacterial genomes (Price and Arkin 2015), implying that the large abundances of marine bacteria may not translate directly into high selection.

In this study, we focused our analyses on the candidate phylum Marinimicrobia, a predominantly marine group that comprises diverse globally abundant lineages involved in distinct biogeochemical processes (Hawley et al. 2017; Getz et al. 2018). Formerly referred to as clade SAR406 or Marine Group A, the Marinimicrobia span a broad range of distinct marine lineages that are poorly understood, in part due to difficulties in cultivating representatives of this phylum. Advances in metagenomics and single-cell sequencing have yielded a large number of draft genomes from this group, however, and several recent studies have shed light on the important role of different Marinimicrobia lineages to carbon and nitrogen cycling in the ocean (Wright et al. 2014; Aylward et al. 2015; Zhang et al. 2016; Bertagnolli et al. 2017; Thrash et al. 2017; Plominsky et al. 2018). In a recent study, we compiled a set of draft Marinimicrobia genomes that have been sequenced using cultivation-independent methods (Getz et al. 2018), and we leverage this set here to analyze the evolutionary genomics of this group. The 211 genomes we used here belong to Marinimicrobia that inhabit both epipelagic and mesopelagic waters across the global ocean and comprise a broad range of phylogenetic diversity (38% average amino acid identity among marker genes in the CheckM marker set; see Materials and Methods).

The Marinimicrobia are an ideal group to test genome streamlining theory because streamlined genomic traits have evolved multiple times independently in this phylum (Getz et al. 2018). Moreover, streamlined genomic characteristics are linked to the environment in which Marinimicrobia are found; epipelagic Marinimicrobia tend to have genomes with low % GC content, short intergenic spacers, and relatively low nitrogen and high carbon-encoded amino acids, while mesopelagic Marinimicrobia generally lack these features (fig. 1). Higher levels of purifying selection in epipelagic Marinimicrobia would therefore be consistent with genome streamlining theory, because it would indicate that these genomic features are not due to genetic drift. Conversely, lower levels of purifying selection in epipelagic Marinimicrobia would suggest that their genomes are shaped by a process analogous to that experienced by endosymbiotic bacteria.

Fig. 1.

—Representation of phylogeny, habitat classification, and genomic features of Marinimicrobia. (a) Maximum likelihood phylogenetic tree of the 211 genomes constructed using amino acid sequences of 120 highly conserved marker genes. (b) Habitat classification based on Getz et al. (2018) and genomic features of Marinimicrobia genomes. Abbreviations: H, Habitat; GC, % GC content (range, 27–55%); IGR, median intergenic region length (range, 7–78 nucleotides); EGS, estimated genome size (range 1–4.4 Mb); N-ARSC (range, 0.3–0.34); C-ARSC (range, 3–3.2). Black points on branches represent support values >0.95.

In order to test our hypothesis, we estimated the ratio of nonsynonymous and synonymous substitutions (dN/dS) of conserved marker genes in Marinimicrobia. In general, dN/dS values <1 are indicative of purifying selection, and the relative strength of selection can be compared across groups using this metric, with lower values implying higher levels of purifying selection (Kryazhimskiy and Plotkin 2008; Kuo et al. 2009). To ensure that our results could be accurately compared across divergent clades, we used two sets of marker genes that are broadly shared among Bacteria, which we refer to here as the EMBL (Sunagawa et al. 2013) and CheckM (Parks et al. 2015) gene sets. We observed a general trend in which epipelagic genomes exhibited lower median dN/dS values (fig. 2 and supplementary data set S1, Supplementary Material online). Although less pronounced, we also observed higher median dS values in epipelagic Marinimicrobia (supplementary fig. S1, Supplementary Material online). The dN/dS values we obtained are far lower than one, which is consistent with the expectation that conserved phylogenetic marker genes experience purifying selection in order to maintain protein function. Our observation of lower median dN/dS values in epipelagic Marinimicrobia was strongly supported by statistical analyses of both the CheckM and the EMBL marker gene sets (Mann–Whitney U test, P < 0.005 in both cases; fig. 2). Our findings suggest that Marinimicrobia found in epipelagic waters experience higher levels of purifying selection than those inhabiting mesopelagic waters.

Fig. 2.

—Violin plot representing median dN/dS values of epipelagic and mesopelagic Marinimicrobia. Statistical significance of differences between dN/dS values of the compared groups according to a nonpaired, one-sided Mann–Whitney–Wilcoxon test is denoted by: (***) for P < 0.005.

It has been shown that dN/dS values are dependent on the time scale in which comparisons are performed (Rocha et al. 2006; Balbi et al. 2009). To test if our results were consistent across different time scales, we created two sets of dN/dS values based on their corresponding dS values, which is a reflection of sequence divergence; one set corresponded to dS values greater than the mean (more divergent comparisons) while the other set corresponded to those lower than the mean (less divergent comparisons). We compared epipelagic versus mesopelagic dN/dS values for both marker sets, and found that epipelagic Marinimicrobia had lower median dN/dS values in all cases (P < 0.005), indicating that the time dependence of dN/dS values is not responsible for our findings.

We also explored the correlation between the strength of selection and several genomic features associated with streamlining. For this purpose, we generated several habitat-specific clusters of closely related Marinimicrobia and plotted their median dN/dS ratio against average genomic characteristics within that cluster (see Materials and Methods). In general, we found that features consistent with genome streamlining were correlated with low dN/dS values (fig. 3 and supplementary fig. S2, Supplementary Material online), with the strongest correlations observed for % GC content (ρ = 0.71, fig. 3a), nitrogen content in amino acid residue side chains (N-ARSC; ρ = 0.54, fig. 3b), and median intergenic region length (ρ = 0.68, fig. 3e). The low N-ARSC values in epipelagic Marinimicrobia are consistent with previous findings, and are likely a product of nutrient limitation in surface waters (Getz et al. 2018). Similarly, we identified a negative correlation between dN/dS values and the carbon content of amino acid residue side chains (C-ARSC; ρ = −0.59, fig. 3d), which is also consistent with higher carbon availability in epipelagic waters. The weakest correlation we observed was between dN/dS values and estimated genome size (ρ = 0.24, fig. 3e). To confirm these trends, we also performed a multivariate analysis of the dN/dS values and genomic features of the Marinimicrobia genome clusters, and the results of this analysis confirmed the tendency of streamlined epipelagic genomes to have lower dN/dS values (fig. 4 and supplementary fig. S3, Supplementary Material online).

Fig. 3.

—Scatter plots showing the relationship between median dN/dS values and streamlined genomic features of the Marinimicrobia genome clusters. Median dN/dS values were calculated using the CheckM marker gene set. (a) GC content versus dN/dS; (b) N-ARSC versus dN/dS; (c) Median intergenic regions length (bp) versus dN/dS; (d) C-ARSC versus dN/dS; (e) estimated genome size (log bp) versus dN/dS. Spearman correlations were performed for each variables pair and details can be found on the main text. Details for the genome clusters can be found in supplementary data set S2, Supplementary Material online.

Fig. 4.

—PCA analysis displaying the Euclidean distance among Marinimicrobia genomes. Abbreviations: gc, %GC content; narsc: N-ARSC; igr, intergenic regions length; dNdS, dN/dS ratio; egs, estimated genome size; CARSC, C-ARSC.

Several previous studies compared a broad array of bacterial lineages and found that dN/dS ratios are generally higher in smaller genomes, suggesting that genetic drift is a prominent evolutionary force in genome reduction (Kuo et al. 2009; Novichkov, et al. 2009b; Sela et al. 2016). We did not identify a strong relationship between estimated genome size and dN/dS ratios in the Marinimicrobia, suggesting that this trend may not hold for this group. Many abundant marine lineages remain poorly represented in sequenced genome repositories due to difficulties in cultivation, and the evolutionary factors shaping their genomes are therefore relatively unexplored. There is evidence of widespread genome streamlining in abundant marine lineages in the ocean (Swan et al. 2013), and as more genomes become available it will be possible to rigorously evaluate the strength of purifying selection on these groups and its possible impact on genome size. The theory of genome streamlining predicts that streamlining may occur across a range of genome sizes due to different genetic repertoires that are necessary to thrive in different environments or ecological niches (Giovannoni et al. 2014), and it is therefore unclear if we would expect streamlined epipelagic Marinimicrobia to have substantially smaller genomes overall. An additional complication of the present study is that the genomes analyzed are incomplete owing to their sequencing via metagenomic or single-cell sequencing efforts, and we extrapolated genome sizes from completeness estimates. The lack of a significant correlation between dN/dS values and genome size must therefore be interpreted with caution, and further studies will be needed to examine this in more detail.

It is important to note that the results we present here do not imply that strong selection directly leads to low % GC content, low N-ARSC, or other streamlined features, since the genomic changes that result from strong selection depend on prevailing environmental factors. For example, strong selection on genomes in mesopelagic waters would not be predicted to lead to a decrease in the N-ARSC of encoded proteins, since nitrogen is more abundant in deeper waters and this evolutionary transition would not be advantageous. The disparate genomic features of epipelagic and mesopelagic Marinimicrobia are therefore likely the result of differential nutrient availability and environmental factors along the water column; surface waters are relatively depleted in nitrogen and phosphorus, while mesopelagic waters contain more of these nutrients but less photosynthetically derived carbon (Karl 2002; Moore et al. 2013). Other factors in addition to selection under different environmental conditions may also play a role in genome streamlining; for example, it has been hypothesized that an increase in mutation rate may lead to some of the genomic features of both endosymbiotic bacteria and abundant marine bacteria (Marais et al. 2008). In our phylogeny of the Marinimicrobia, genomes that contain streamlined genomic features in Clades 2 and 3 (red and light green in fig. 1, respectively) are associated with long branches that may be indicative of increased mutation rates. This link between long branches and genome streamlining must be made with caution, however, because long branches do not definitively demonstrate increased mutation rates, and even so it is unclear if this would lead to other genomic features of streamlining. Nevertheless, given the complexity of these genome evolutionary processes, it is likely that multiple factors are responsible for the trends we observe here.

An important caveat of the dN/dS ratio is that it only provides insight into the strength of recent selective pressure and therefore cannot be used to infer the selective strength experienced by lineages in the past. Other streamlined lineages such as the Pelagibacterales and Prochlorococcus are thought to have underwent genome reduction in the distant past, and it is therefore difficult to assess the strength of selection on these ancestral genomes during these transitions. Some studies have suggested that genetic drift due to possible population bottlenecks drove these genomic changes (Luo et al. 2017), while other studies have argued that strong purifying selection was the primary driver (Sun and Blanchard 2014). In contrast to these other streamlined groups, the Marinimicrobia appear to have experienced multiple independent genome transition events relatively recently in their evolutionary history (Getz et al. 2018), and comparison of the selective pressures across disparate clades with similar genomic features therefore provides insight into more recent selective regimes that led to current genomic architectures. Overall our results suggest that although parasitic and endosymbiotic bacteria share some genomic features with streamlined bacteria, these features are the product of distinct evolutionary paths.

Materials and Methods

Marinimicrobia Genomes Used

We analyzed a set of 211 Marinimicrobia genomes derived from a previous study (Getz et al. 2018). This data set included genomes from GenBank (Sayers et al. 2019), the Integrated Microbial Genomes database (IMG; Markowitz and Kyrpides 2007), and from two different studies in which Metagenome-Assembled Genomes (MAGs) were generated (Delmont et al. 2018; Tully et al. 2018). The data set employed by Getz et al. was complemented with the genomes SCGC_AD-604-D17, SCGC_AD-606-A07, SCGC_AD-615_E22 from another recent study (Plominsky et al. 2018). Methods for quality filtering, estimation of genome completeness and contamination, and the calculation of genomic features have been described previously (Getz et al. 2018).

Phylogenetic Reconstruction

To reconstruct the Marinimicrobia phylogeny, we predicted proteins from genomes using Prodigal v2.6.2 (Hyatt et al. 2010) and identified phylogenetic marker genes using HMMER3 (Eddy 2011). We constructed a phylogeny from an amino acid alignment created from the concatenation of 120 marker genes that have been previously used for phylogenetic reconstructions of Bacteria (Parks et al. 2015). The trusted cutoffs were used in all HMMER3 searches with the “cut_tc” option in hmmsearch. We used the standard_fasttree workflow included in the ETE Toolkit which includes ClustalOmega for alignment (Sievers and Higgins 2018), trimAl for alignment trimming (Capella-Gutierrez et al. 2009), and FastTree for phylogenetic estimation (Price et al. 2010). The different branches obtained were classified into clades based on previously published results (Getz et al. 2018). We visualized the resulting tree in the interactive Tree of Life (iTOL; Letunic and Bork 2016; https://itol.embl.de/tree/45379142397251562088683).

dN/dS Ratio Calculation and Filtering

To estimate the strength of purifying selection, we used the ratio of nonsynonymous and synonymous substitutions (dN/dS). When considering values <1, lower values are a sign of higher purifying selection while higher values are a sign of higher genetic drift (low purifying selection). To calculate genome-wide dN/dS ratios, we used two sets of conserved marker genes that would be expected to be found in most genomes. The first one consists of 120 phylogenetic marker genes that are highly conserved in Bacteria, which we also used for phylogenetic reconstruction (Parks et al. 2015). The second set consists of 40 phylogenetic marker genes used in phylogenetic reconstructions, which we refer to as the EMBL set due to its development in the European Molecular Biology Laboratory (Sunagawa et al. 2013).

For both marker gene sets, we predicted proteins from each genome using Prodigal and then annotated the marker genes of interest using the hmmsearch tool of HMMER3 with model-specific cutoffs. We aligned the amino acid sequences for each annotated gene coming from Marinimicrobia genomes separately using ClustalOmega, and the resulting alignments converted into codon alignments using PAL2NAL (Suyama et al. 2006). Maximum-likelihood approximation (codeML) within the PAML 4.9 h package (Yang 2007) was used through Biopython in order to perform dN/dS pairwise comparisons within the clades previously established (Getz et al. 2018). We removed dN/dS values with dS ≥1, which implies that synonymous substitutions are near saturation. Moreover, to avoid comparing sequences from genomes that may be part of the same population, we also excluded comparisons for which dN = 0 and dS ≤0.01. Additionally, we discarded all dN/dS values ≥10 on the grounds that these were largely artifactual. Lastly, because we wished to compare dN/dS values from Marinimicrobia that reside in different habitats, we only included dN/dS values where the pair of compared genomes were from the same habitat (epipelagic and mesopelagic). All values used can be found in supplementary data set S1, Supplementary Material online.

Genome Clustering

To compare dN/dS values with other genomic features, it was first necessary to generate clusters of closely related genomes. For this, Marinimicrobia genomes were compared using the MASH program (Ondov et al. 2016), which rapidly identifies similarities in the k-mer profiles of genomes and provides statistical measures of nucleotide similarity. Comparisons that yielded MASH e-values <1e-100 were retained and used to link closely related genomes, and final genome clusters were generated using a single-linkage clustering algorithm in R. Median dN/dS values for all clusters were calculated and then plotted against average genome features within that cluster (% GC content, estimated genome size, median intergenic region length, estimated genome size, N-ARSC, and C-ARSC; see figs. 3 and 4). Clusters that had fewer than 10 total dN/dS measurements were excluded from further analyses. Details for the genome clusters can be found in supplementary data set S2, Supplementary Material online. This approach of using clusters of related genomes to estimate group-specific dN/dS values is similar to previously used methods (Kuo et al. 2009; Novichkov et al. 2009a).

Statistical Analyses

In order to investigate the strength of selection acting on epipelagic and mesopelagic Marinimicrobia, genomes were classified into epipelagic and mesopelagic based on their biogeographic distribution (Getz et al. 2018). For statistical analysis, we loaded the filtered dN/dS values into R and performed comparisons using the Mann–Whitney U test (wilcox.test() function). Additionally, to investigate the relationship between median dN/dS and genomic features associated to each cluster, we applied the “cor.test” function using the Spearman method. Comparisons and correlation plots were visualized through the ggplot2 package (Wilkinson 2011). We also explored the distance between epipelagic and mesopelagic Marinimicrobia genomes employing the genomic features and median dN/dS values through a PCA analysis with the “prcomp” function available on R. Euclidean distance was visualized using the “ggbiplot” function within the ggplot2 package (Wilkinson 2011).

Supplementary Material

Supplementary data are available at Genome Biology and Evolution online.