Population Genetics of Paramecium Mitochondrial Genomes: Recombination, Mutation Spectrum, and Efficacy of Selection

Abstract

The evolution of mitochondrial genomes and their population-genetic environment among unicellular eukaryotes are understudied. Ciliate mitochondrial genomes exhibit a unique combination of characteristics, including a linear organization and the presence of multiple genes with no known function or detectable homologs in other eukaryotes. Here we study the variation of ciliate mitochondrial genomes both within and across 13 highly diverged Paramecium species, including multiple species from the P. aurelia species complex, with four outgroup species: P. caudatum, P. multimicronucleatum, and two strains that may represent novel related species. We observe extraordinary conservation of gene order and protein-coding content in Paramecium mitochondria across species. In contrast, significant differences are observed in tRNA content and copy number, which is highly conserved in species belonging to the P. aurelia complex but variable among and even within the other Paramecium species. There is an increase in GC content from ∼20% to ∼40% on the branch leading to the P. aurelia complex. Patterns of polymorphism in population-genomic data and mutation-accumulation experiments suggest that the increase in GC content is primarily due to changes in the mutation spectra in the P. aurelia species. Finally, we find no evidence of recombination in Paramecium mitochondria and find that the mitochondrial genome appears to experience either similar or stronger efficacy of purifying selection than the nucleus.

Introduction

Mitochondrial genomes have played integral roles in furthering our understanding of relationships among species as well as revealing population structure and demographic history. As a consequence, we have obtained insights into the unique population-genetic properties of mitochondrial genomes. In most species, mitochondria are inherited uniparentally (but see Barr et al. 2005) and mitochondrial genomes are known to frequently undergo recombination in plants (Stadler and Delph 2002; Mackenzie 2007), fungi (Fritsch et al. 2014), as well as in animals (Piganeau et al. 2004; Tsaousis et al. 2005; Ladoukakis et al. 2011). Because of the unique mode of transmission, mitochondria have been suggested to have lower effective population sizes than their nuclear counterparts and therefore to accumulate more deleterious mutations (Lynch and Blanchard 1998; Neiman and Taylor 2009). In addition, mitochondrial genomes experience much higher spontaneous rates of mutation than their corresponding nuclear genomes in animals, but exhibit the opposite trend in plants (Lynch et al. 2006).

Unlike the relatively uniform and conserved properties of bilaterian metazoan mitochondrial genomes, mitochondrial genomes in other eukaryotes exhibit remarkable variation in genome structure and GC content, particularly among unicellular lineages. Mitochondrial genome structures range from hundreds of short linear segments (0.3–8.3 kb) in the ichtysporean Amoebidium parasiticum (Burger et al. 2003), an opisthokont, to many small (<10 kb) circular genomes in the diplonemid Diplonema papillatum (Vlcek et al. 2011), an excavate, to a single larger linear or circular chromosome, and a variety of other states (reviewed by Smith and Keeling 2015). In addition to variation in organization, mitochondria from unicellular lineages display widely diverse GC compositions, ranging from as low as 10% in some yeast (Smith 2012) to as high as 60% in Lobochlamys culleus (Borza et al. 2009), although most species are AT rich, with an average GC content of 35% (Smith 2012).

Much has been learned about the structure, evolution, and population-genetic environment of mitochondria in the main model systems, especially in plants and metazoans (Lynch 2007; Smith 2016). However, we lack such understanding of mitochondria of the majority of unicellular eukaryotes, where the bulk of eukaryotic phylogenetic diversity lies. We address this gap by surveying both within and between-species variation in mitochondrial genomes among multiple ciliate species belonging to the genus Paramecium.

In the large and morphologically and ecologically diverse ciliate lineage, mitochondrial genomes sampled so far are organized into large linear chromosomes, several tens of kilobases in length, with telomeres at the ends (Goddard and Cummings 1975; Morin and Cech 1988; Swart et al. 2012). However, few mitochondrial genomes have until now been fully sequenced among the ciliates, with two in the Paramecium genus (P. tetraurelia and P. caudatum; Barth and Berendonk 2011), and only a few others: Tetrahymena pyriformis (Burger et al. 2000), Euplotes minuta and Euplotes crassus (de Graaf et al. 2009), Oxytricha trifallax (Swart et al. 2012), Stentor coeruleus (Slabodnick et al. 2017), Ichthyophthirius multifiliis (Coyne et al. 2011), and the anaerobic ciliate Nyctotherus ovalis (de Graaf et al. 2011).

Although the Paramecium genus contains a number of distinct morphospecies, it is especially known for including a species complex consisting of multiple morphologically identical but sexually isolated species—the P. aurelia complex (Sonneborn 1975). Species in the P. aurelia complex are ancient, with the estimated time of divergence for the complex as a whole being on the order of 300 Myr (McGrath, Gout, Johri, et al. 2014), implying that the genus Paramecium itself is even more ancient. Interestingly, among Paramecium species, there is an increase of GC content in mtDNA in the branch leading to the P. aurelia complex (Burger et al. 2000; Barth and Berendonk 2011), allowing us to study the evolution of nucleotide composition across mitochondrial genomes that are structurally very similar.

The Paramecium species offer a particularly interesting system in which to study the evolution of mitochondrial genomes because of the unique population-genetic environment experienced by their cellular organelles. Paramecium cells are mitochondria-rich: Each individual cell in P. aurelia species is estimated to contain about 5,000 mitochondria, with about 8–10 genomes per mitochondrion (Beale and Tait 1981), which is much larger than in mammalian cells with 1,000–2,000 mitochondria (Kukat et al. 2011) and yeast cells with 20–30 mitochondria per cell (Visser et al. 1995). In addition, Paramecium lineages, like other ciliate species, possess two nuclei: the polyploid somatic nucleus (called the macronucleus), which divides amitotically where the bulk of transcriptional activity occurs, and the diploid germline nucleus (known as the micronucleus), which is transcriptionally silent, and which undergoes sexual reproduction. All Paramecium species can proliferate asexually for a limited number of generations, after which they senesce unless they undergo sexual reproduction, that is, conjugation. During asexual proliferation, Paramecium undergoes binary fission; in this process, mitochondria appear to double in length, replicate their genomes (Perasso and Beisson 1978), and are randomly distributed between the two daughter cells (Adoutte and Beisson 1972; Adoutte and Doussiere 1978). Mitochondria are therefore thought not to experience any bottlenecks during binary fission division.

During conjugation, Paramecium cells exhibit cytoplasmic inheritance (Koizumi and Kobayashi 1989), that is, despite the exchange of micronuclei between the two conjugants there is almost no exchange of cytoplasm and other organelles (reviewed by Meyer and Garnier 2002). Thus, mitochondria are uniparentally inherited. A distinct aspect of Paramecium mitochondrial biology is that Paramecium mitochondria appear to exist as independent structural units and do not undergo fusion, unlike the constant flux of organelle fusion and fission in other metazoan mitochondrial populations (Kiefel et al. 2006). Both uniparental inheritance of mitochondria and the absence of fusion in the cytoplasm suggest a lack of recombination among mitochondria genomes.

In this study, we further the understanding of the biology and the population-genetic environment of ciliate mitochondria by presenting the complete mitochondrial genomes of nine species belonging to the P. aurelia complex, four outgroup (relative to P. aurelia) species, and additional 5–10 isolates for each of four Paramecium species. Using phylogenetic and population-genetic analyses, we investigate variation in protein-coding genes, tRNA content, and the forces governing the evolution of nucleotide composition of mitochondrial genomes across the phylogeny. Finally, we estimate the recombination rate across the genome and address the controversy of whether mitochondrial genomes experience reduced efficacy of purifying selection in comparison to their nuclear counterparts.

Results

Sequencing and Assembly of Mitochondrial Genomes and Detection of Single Nucleotide Polymorphisms

We assembled complete mitochondrial genomes of seven species belonging to the P. aurelia complex: P. biaurelia, P. tetraurelia, P. sexaurelia, P. octaurelia, P. novaurelia, P. decaurelia, P. dodecaurelia, P. quadecaurelia, and P. jenningsi. In addition, we analyzed previously reported complete mitochondrial genomes of P. tetraurelia and P. sexaurelia belonging to the P. aurelia complex, as well as four outgroup species: P. caudatum, P. caudatum-C026, P. multimicronucleatum, and P. multimicronucleatum-Peniche3I (fig. 1; isolates sequenced by Johri et al. [2017]). We note that P. caudatum-C026 and P. multimicronucleatum-Peniche3I were initially sampled as individual isolates belonging to the P. caudatum and P. multimicronucleatum species (based on morphological criteria), respectively. However, the analysis of the mitochondrial sequences revealed that they are highly diverged from the reference strains (see below) and are therefore almost certainly separate species and were treated as such in subsequent analyses. For all seven new mitochondrial genomes sequenced and assembled in this study, Illumina reads were assembled using SPAdes (Bankevich et al. 2012), and mitochondrial contigs were identified by BLAST searches against the publicly available P. caudatum and P. tetraurelia mitochondrial sequences (see the Materials and Methods section for more details).

Fig. 1.

—Mitochondrial phylogeny and change in nucleotide composition across Paramecium species. The species in green belong to the Paramecium aurelia complex. Numbers on branches indicate total number of substitutions per site and bold numbers on nodes show bootstrap values. Black solid circles show the inferred event of loss of the ribosomal protein rpS19. The numbers on the right show the total assembly size (in bp) and mean GC content of the whole mitochondrial genome (Total), at 4-fold degenerate sites, at intergenic regions, and of tRNAs, respectively. Starred (*) numbers indicate incomplete assembly of mitochondrial genomes.

In addition, we examined sequenced mitochondrial genomes of ten isolates of both P. tetraurelia and P. sexaurelia, and five isolates of each P. caudatum and P. multimicronucleatum (supplementary fig. 1, Supplementary Material online) sampled worldwide (Johri et al. 2017). Illumina paired-end reads from these isolates were mapped to the assembled reference genomes, and single nucleotide polymorphisms (SNPs) were called as in Johri et al. (2017). In this study, all mitochondrial genomes of individual isolates were also assembled de novo (supplementary figs. 2–4, Supplementary Material online) in order to examine large-scale genome organization mapping.

Genome Structure, Organization, and Telomeric Repeats

All Paramecium mitochondrial genomes are linear (Goddard and Cummings 1975; Morin and Cech 1988; Swart et al. 2012). Our assemblies show that they are ∼40 kb in the P. aurelia species and ∼44 kb in P. caudatum and P. caudatum-C026. Telomeric repeats and gene content observed at ends of the assembled genomes imply that the full lengths of the linear contigs have been assembled in most species with the exception of the P. multimicronucleatum and P. multimicronucleatum-Peniche3I mitocontigs, which appear to be missing small portions of the 3′-end of the chromosome (fig. 2). The length of the assembled contigs in P. multimicronucleatum and P. multimicronucleatum-Peniche3I suggests an overall size closer to that observed in P. aurelia than to the larger mitochondrial genomes in the P. caudatum lineage. We also note that the raw assemblies for two of the P. aurelia species contain extensions (supplementary figs. 3, 5, and 6, Supplementary Material online). In P. novaurelia, an additional ∼18 kb is present at the 5′-end of the mitocontig, whereas in P. quadecaurelia, a small, ∼1-kb extension is seen at the 3′-end. However, the read coverage over these regions is very different from the rest of the mitocontigs, suggesting either misassembly or heterogeneity within cell populations. We therefore ignored these extensions in subsequent analysis. The organization of the genomes is very similar, with gene order preserved almost perfectly across all species (fig. 2).

Fig. 2.

—Structure of mitochondrial genomes in Paramecium. All Paramecium mitochondrial genomes are linear with 24 protein-coding genes (shown in reddish orange), 2 rRNA genes (yellow), lineage-specific genes referred to as Ymf genes (in green) and varying numbers and content of tRNA genes (shown in light blue).

We also identified telomeric repeats based on the sequences at the end of assembled mitocontigs (see the Materials and Methods section for more details). In most P. aurelia species, we identified almost identical repeats, with a 23-bp consensus sequence GCCCTGGTGGCCCTAGAAGCTCC (fig. 3). However, in P. jenningsi and P. sexaurelia the telomere repeat motif is the same length, but differs from the consensus sequence in other species at one nucleotide position. Of note, these two species are the earliest diverging ones within the P. aurelia complex species included in our analysis. We observed even more divergent telomeric sequences in P. caudatum (GCCCTGGTAACGCTGGTCGCCCTTTTAAAATA) and P. multimicronucleatum (GCCCTTGTTACACTTGGTGGCTCTTTAAGCTCT). In these species, the core telomeric repeat sequence has been expanded by an additional 10 bp of sequence not present in P. aurelia. The Paramecium core telomeric repeat is broadly similar to that of Tetrahymena (ACCCTCGTGTCTCTTTA; fig. 3), the other Oligohymenophorean genus for which mitochondrial genomes are available, but distinct from what is observed in distantly branching ciliates such as the spirotrichean O. trifallax (fig. 3).

Fig. 3.

—Telomeric repeat sequences in the Paramecium genus. Nucleotides in red show the single base pair differences among the Paramecium aurelia species. The phylogenetic tree on the left is built using the telomeric repeat sequences.

Fig. 4.

—tRNA content variation in Paramecium caudatum and Paramecium multimicronucleatum strains. tRNAs shaded in gray are present in all Paramecium and Tetrahymena species. tRNAs whose anticodons were ambiguous are displayed as “Undet.”

Examination of existing RNA-seq data sets for several of the species revealed no evidence for RNA editing in Paramecium mitochondria (data not shown), in concordance with previous reports (Orr et al. 1997).

Variation in Protein-Coding Gene Content between and within Species

All Paramecium mitochondrial genomes contain a core set of 15 genes (fig. 2) involved in electron transport and ATP synthesis (atp9, cob, cox1, cox2, nadh1, nadh2, nadh3, nadh4, nadh4L, nadh5, nadh6, nadh7, nadh9, and nadh10), the heme maturase YejR, and 8 ribosomal protein genes (rpL6, rpL12, rpL14, rpL16, rpS3, rpS12, rpS13, and rpS14). One ribosomal protein subunit rpS19 was found to be absent from P. tetraurelia, P. octaurelia, P. novaurelia, P. dodecaurelia, and P. quadecaurelia and therefore appears to have been independently lost from mitochondrial genomes at least three times along the P. aurelia phylogeny (fig. 1). Remarkably, we also find a presence–absence polymorphism of rpS19 within P. sexaurelia isolates, with the open-reading frame (ORF) not being identifiable in 2 out of 11 isolates studied (supplementary fig. 10, Supplementary Material online). Although this might suggest a mitochondrial gene in the process of transfer to the nucleus, the closest hit in the nuclear genome has an E-value of 0.27, implying either a complete loss or perhaps transfer to a mitochondrial plasmid, which has not been captured in existing assemblies.

Paramecium genomes also contain 16 ciliate-specific genes of unknown function, called Ymf genes (Burger et al. 2000), named according to the Commission on Plant Gene Nomenclature (Price and Reardon 2001). Each sequenced strain contains all of these genes, with the exception of P. multimicronucleatum strain M13 in which Ymf80 and Ymf83 appear to have fused with nadh4 and Ymf64, respectively. In addition, several ORFs originally identified in P. tetraurelia are observed in only one or a few additional P. aurelia mitochondrial genomes, typically overlapping longer ORFs: orf189, orf285, and orf314 restricted to the three P. tetraurelia isolates; orf221 present in P. tetraurelia, P. octaurelia, and P. biaurelia; orf265 in P. biaurelia; and orf78 found only in P. tetraurelia isolate A.

In addition, we found a previously unidentified ORF larger than 100 amino acids, which appears to be present in all Paramecium mitochondrial genomes and is located at the very 5′-end of the chromosome, immediately before the large ribosomal RNA. This ORF has homology to the clustered regularly interspaced short palindromic repeats (CRISPR)-associated endoribonuclease Cas6; its functional significance is unclear at present. Interestingly, an ORF homologous to CRISPR-associated helicase (Csf4) also has been identified in mitochondrial genomes of multiple bivalve species (Milani et al. 2013) and has been shown to produce a functional protein product (Milani et al. 2014). Thus, the ORF found in Paramecium mitochondrial genomes could potentially be functional and a case of horizontal gene transfer from some of the endosymbiotic bacteria (Preer 1969; Fokin and Gortz 2009) often associated with Paramecium. Sequencing of more ciliate mitochondrial genomes should provide deeper insights into its origin.

Paramecium Species Are Highly Diverged and Contain Cryptic Species Complexes

We found most species to be fairly evolutionarily distant from each other, as measured by the average number of nonsynonymous (dN) and synonymous (d_s) site substitutions per site (supplementary fig. 11, Supplementary Material online) between all pairs of species, using yn00, PAML (Yang 2007; version 4.9a). Parameciumdecaurelia and P. dodecaurelia (average d_s: 0.85; average dN: 0.05; average total divergence: 0.23) are the closest species pair, followed by P. tetraurelia and P. octaurelia (average d_s: 0.93; average dN: 0.09; average total divergence: 0.27). All other species pairs exhibit on average d_s > 1.0, that is, synonymous sites in protein-coding genes have on average undergone more than 1 substitution each since the time of divergence. We found that mean d_s between P. caudatum and P. caudatum-C026 is ∼2.0 (average dN: 0.07; average total divergence: 0.28), implying that these two isolates possibly represent separate species. A similar observation confirms that P. multimicronucleatum-Peniche3I and P. multimicronucleatum are probably separate species (average d_s = 1.97; average dN: 0.10; average total divergence: 0.26). Our observations suggest the possibility of cryptic species complexes in both P. caudatum and P. multimicronucleatum, as previously suspected (Hori et al. 2006; Tarcz et al. 2012). We also note that several additional P. multimicronucleatum isolates most closely related to P. multimicronucleatum-Peniche3I might also be separate species as they are fairly divergent from the reference strain as well as P. multimicronucleatum-Peniche3I (supplementary fig. 1, Supplementary Material online). The phenomenon of species complexes containing numerous morphologically identical species, already known for P. aurelia (Sonneborn 1937) and T. pyriformis (Gruchy 1955), is therefore probably much more widespread among ciliates than previously appreciated.

Selective Pressures on Ciliate-Specific Mitochondrial Genes

The 16 ciliate-specific mitochondrial genes found in all Paramecium species have orthologs in Tetrahymena species as well as in Oxytricha. Thus, these genes have been preserved for a long evolutionary time, and yet have diverged sufficiently that no known homologs exist in well-studied species in other eukaryotic kingdoms. Across the Paramecium species, they appear to be on average faster evolving (supplementary fig. 12 and table 1, Supplementary Material online) with average dN/d_s ≅ 0.11, in comparison to genes that encode enzyme components of the respiratory chains (average dN/dS ≅ 0.04) and ribosomal proteins (average dN/d_s ≅ 0.06). The dN/dS values here were obtained for each gene under the assumption that dN/dS remains constant across the phylogeny using CODEML, PAML (Yang 2007; version 4.9a). Ciliate-specific Ymf genes therefore exhibit a relatively higher rate of evolution than other genes, with as little as 30% sequence identity between P. aurelia, P. caudatum, and P. multimicronucleatum, similar to observations in Tetrahymena (Moradian et al. 2007), as well as with higher values of π_n/π_s in P. tetraurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum (supplementary fig. 13, Supplementary Material online).

The higher rate of evolution of Ymf genes can either be explained by relaxed purifying selection at the sequence level, or recurrent positive selection over long periods of time, or a combination of both. To distinguish between these possibilities, we performed a McDonald–Kreitman test (MK test) in the species P. tetraurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum. As most of these species are highly diverged from each other, we used ancestral reconstruction over the set of all 13 taxa to first infer ancestral nucleotides for each internal node. The numbers of synonymous (D_S) and nonsynonymous (D_N) changes were then inferred along each terminal branch leading to each of the four species mentioned above. We tested for positive selection by determining whether the proportion of nonsynonymous fixed differences relative to synonymous ones was significantly larger than the proportion of nonsynonymous relative to synonymous polymorphic differences. None of the genes in P. tetraurelia was found to have undergone positive selection by this criterion (Holm–Bonferroni corrected P ≥ 0.05 in each case). In P. sexaurelia, cox1, Ymf76, and Ymf63 are the only genes that showed significant positive selection, whereas in P. caudatum, only Ymf64 exhibits positive selection (supplementary table 2, Supplementary Material online). With very few SNPs available in P. multimicronucleatum, the MK test was not significant for any gene. Nevertheless, we find no statistical enrichment for number of genes evolving under positive selection among the Ymf relative to other genes in P. sexaurelia (P = 0.56; Fisher’s exact test) and P. caudatum (P = 0.40; Fisher’s exact test), suggesting that the majority of them are likely evolving fast due to relaxed sequence constraints.

It is possible that most of the Ymf genes are simply highly diverged genes that perform the same functions as genes found in other mitochondrial genomes, but are not identifiable by sequence. In order to gain some insight into their possible functions, we used HHPred (Soding et al. 2005) to predict their homologs and found that parts of Ymf genes display homology to standard mitochondrial genes, encoding ribosomal subunits and proteins involved in electron transport. About 37% of sequence from Ymf59 was found to be homologous to rpS10, ∼22% sequence of Ymf61 to a 30S ribosomal protein S24E, ∼28% of Ymf64 sequence to a 30S rpS3, ∼79% of Ymf65 sequence to the NADH subunit, and ∼11% of Ymf67 to cytochrome c oxidase, subunit P. Ymf59 and Ymf64 in Oxytricha were also previously found to be homologous to rpsS10 and rpS3, respectively (Swart et al. 2012).

tRNA Content Variation across and within Species

Although the number and identity of protein-coding genes are generally the same across the genus, tRNAs display significant variation. All species in the P. aurelia complex have only three tRNA loci, at highly conserved locations (fig. 2), most of the time consisting of a Y, F, and W tRNAs (the latter recognizing the UGA stop codon that has been reassigned to tryptophan in ciliate mitochondria). In stark contrast, large and highly variable sets of tRNA are predicted in mitochondrial genomes in the P. caudatum and P. multimicronucleatum lineages (fig. 3), with as many as 38 tRNA predictions in P. multimicronucleatum M05, and with individual P. caudatum and P. multimicronucleatum isolates having different tRNAs at multiple loci in their mitochondrial genomes. These tRNAs often overlap protein-coding genes (in contrast to what is observed in P. aurelia), in particular Ymf genes. Of note, the three tRNAs found in the P. aurelia species are also present in all outgroup species. The absolute numbers of the tRNAs found should be taken with caution, as some of the predicted tRNAs overlap each other as well as Ymf64 in the P. multimicronucleatum strains. However, even if these are excluded, the number of tRNAs in both P. caudatum and P. multimicronucleatum strains is still large and variable and consists of tRNAs not found in the P. aurelia species. This is a unique finding among mitochondrial genomes and should be investigated more thoroughly in future studies to better understand the origin of the tRNAs.

Mutation Spectrum and GC-Content Variation within the Paramecium Genus

Perhaps, the most remarkable discontinuity in Paramecium mitochondrial genomes is the difference between the high GC content observed in species of the P. aurelia complex—∼40%—and the low GC content in the outgroup species—∼20% (supplementary figs. 7–9, Supplementary Material online). Very low GC content is also observed in mitochondrial genomes of other ciliates, for instance, in the five Tetrahymena species (∼18–21%), in Icthyophthirius multifiliis (∼16%) (Coyne et al. 2011), and in O. trifallax (∼24%) suggesting a lineage-specific increase in GC content along the branch leading to P. aurelia species. In contrast, nuclear genomes have relatively similar levels of GC content across all Paramecium species (McGrath, Gout, Doak, et al. 2014). The low GC content of P. caudatum and P. multimicronucleatum mtDNA is marked by a highly biased codon usage (Barth and Berendonk 2011), with most synonymous positions exhibiting a strong bias for A or T nucleotides (supplementary fig. 14, Supplementary Material online). Indeed, the difference in GC content is most pronounced at 4-fold degenerate sites, with the GC content at such sites being as low as 3.5% in P. multimicronucleatum and as high as 54.5% in P. decaurelia (fig. 1). The 0-fold redundant sites in the outgroup species have much higher GC content (∼26–37%), and the GC content (31–36%) of the rRNAs is similar across the entire genus suggesting strong selection on rRNA nucleotide content.

At least three forces may be responsible for determining the GC content at 4-fold degenerate sites: 1) mutational processes, 2) codon usage bias due to selection, and/or 3) genome-wide selection for higher/lower GC content. We therefore examined GC composition in other regions of the genomes, such as tRNAs and intergenic regions, which do not experience selection for codon usage. Both tRNAs and intergenic regions have elevated GC content in the P. aurelia species (∼40–45% and ∼34–37%) relative to P. caudatum and P. multimicronucleatum (∼14–22% and ∼16–20%, respectively; fig. 1), suggesting that either differences in mutational bias or selection for genome-wide GC content is primarily responsible for changes in nucleotide composition across the genome.

To distinguish between mutation and selection as the possible explanations for these observations, we evaluated the mutational pressures acting on Paramecium species, using two approaches. First, we used the mitochondrial mutation spectrum obtained from mutation-accumulation (MA) studies of P. tetraurelia (Sung et al. 2012) to calculate the AT mutation bias, m = v/u, where v is the mutation rate from G/C to A/T and u is the rate of mutation from A/T to G/C (table 1). We found m ≅ 1.0, showing little mutation bias and corresponding to expected equilibrium GC content of 51.3% (table 1). This is very close to the GC content of 4-fold redundant sites (54%) in P. tetraurelia. We also analyzed existing sequencing data from MA experiments in P. biaurelia and P. sexaurelia (Long, Doak, et al. 2018; supplementary table 3, Supplementary Material online). These data imply slight mutation bias toward GC in P. biaurelia and P. sexaurelia, corresponding to expected equilibrium GC percentages (Sueoka 1993) of 55.8% and 67.0%, respectively. Given the relatively small number of mutations (a few tens) identified in each set of MA experiments, there is considerable uncertainty in these estimates. Nonetheless, these data do not present evidence that 4-fold degenerate sites in P. aurelia mitochondria are evolving away from mutation equilibrium via selection.

Table 1.

Mutation Bias toward AT (m) in the Mitochondrial Genomes of Paramecium and Other Species, Where v and u Are Mutation Rates from G/C to A/T, and A/T to G/C, respectively.

Species	Type of Study	Total Number of Base-Substitution Mutations or SNPs	AT Mutation Bias (m = v/u)	Expected Equilibrium GC Content [1/(1 + m)]	Whole-Genome GC Content	4-Fold Site GC Content	S (avg)	S (4-fold)
Parameciumtetraurelia1	MA	Not known	0.88	0.53	0.42	0.54	0.197	−0.013
Parameciumtetraurelia	Population	37	0.95	0.51	0.42	0.54	0.163	−0.047
Parameciumbiaurelia2	MA	55	0.79	0.56	0.40	0.51	0.278	0.085
Parameciumsexaurelia2	MA	87	0.49	0.67	0.39	0.47	0.504	0.368
Parameciumsexaurelia	Population	112	1.56	0.39	0.39	0.47	0.000	−0.137
Parameciumcaudatum	Population	96	5.02	0.17	0.22	0.07	−0.151	0.397
Other species from previous studies
Saccharomyces cerevisiae3	MA	13	0.00	1.00	0.17	0.08	>>0	>>0
Caenorhabditis elegans4,5	MA	25	3.79	0.21	0.24	0.14	−0.073	0.224
Daphnia pulex6	MA	6	2.48	0.29	0.38	0.30	−0.182	−0.026
Caenorhabditis briggsae7	MA	15	18.10	0.05	0.25	0.14	−0.780	−0.451
Drosophila melanogaster8	MA	28	381.25	0.00	0.18	0.06	−1.923	−1.386

note.—Values of m <1 indicate that mutation spectrum is biased toward G/C. S = 4N_es (or 2N_es) for diploids (or haploids) represents the strength of selection favoring A/T composition, where negative values of S represent selection favoring G/C. S is calculated using the equation, P_AT = 1/(1 + m⁻¹e^−S), where P_AT is the fraction of AT nucleotides. 1: Sung et al. (2012); 2: Long et al. (2018); 3: Lynch et al. (2008); 4: Denver et al. (2000); 5: Konrad et al. (2017); 6: Xu et al. (2012); 7: Howe et al. (2010); 8: Haag-Liautard et al. (2008).

Next, derived singleton alleles at 4-fold degenerate sites were used to quantify the number of G/C to A/T mutations relative to A/T to G/C mutations and thus infer an estimate of AT mutation bias from population data. Again, we find that m is close to 1.0 (i.e., no mutation bias toward A/T) in both P. tetraurelia and P. sexaurelia, with the predicted GC content under mutation equilibrium being remarkably close to that of their 4-fold degenerate sites (table 1). These results suggest that the composition of 4-fold degenerate sites is mostly determined by mutation. The same can be seen by calculating the strength of selection (S = 4N_es) favoring A/T at 4-fold degenerate sites. As S usually takes values between 0.1 and 4.0 (Lynch 2007; Long, Sung, et al. 2018) and we obtain much lower magnitudes (0.0–0.5) in the P. aurelia species (table 1), we infer negligible or weak selection at 4-fold degenerate sites, consistent with absence of codon bias at the third position.

In contrast to the P. aurelia species, we infer a significant AT mutation bias in P. caudatum (table 1), yielding a predicted GC content at equilibrium of ∼17%, with the observed GC content in intergenic and tRNA regions (16% and 20%) being remarkably close to this value. These calculations suggest that genome-wide GC content in P. caudatum mitochondria is mostly governed by mutation bias, and that there has been a major shift in mutation bias along the branch leading to the P. aurelia species.

A possible confounding factor affecting these analyses is the presence of chunks of mitochondrial sequence in the nuclear genome, also known as NUMTs (nuclear mitochondrial DNA sequences). In general, the nuclear dualism of ciliates would be expected to result in few to no NUMTs being present in significant quantities as the highly polyploid macronucleus contributes nothing to future generations whereas the germline micronucleus is transcriptionally silent and heterochromatinized (thus less susceptible to insertions of NUMTs). Nevertheless, the possibility that NUMTs are present in Paramecium genomes cannot be dismissed, especially given that no micronucleus assemblies are available and even the macronucleus ones are not entirely complete. To account for this possibility, we carried out a parallel reanalysis of our data using a more conservative set of SNPs for all species, derived by mapping all sequenced reads to the mitochondrial reference genome and the macronuclear reference genome simultaneously, and subsequently excluding all sites that exhibit heterozygous genotypes (if NUMTs are the source of these variants, their genotypes would be heterozygous). We found that most variant positions and their allele frequencies do not change (supplementary table 4, Supplementary Material online) and that π_n/π_s values for all genes are highly correlated (R² of 0.88, 0.99, and 0.67 in P. tetraurelia, P. sexaurelia, P. caudatum, respectively; supplementary fig. 15, Supplementary Material online). Most importantly, only a very slight change in the inferred mutation spectrum from population data is observed between the two sets of analyses, with the AT mutation bias in the conservative SNP set being 1.102 in P. tetraurelia (using 41 SNPs), 2.058 in P. sexaurelia (using 99 SNPs), and 6.479 in P. caudatum (using 30 SNPs).

Recombination in Paramecium Mitochondrial Genomes

Next, we sought to evaluate evidence for the occurrence of recombination in Paramecium mitogenomes by performing multiple tests for its presence. First, we evaluated the relationship between linkage disequilibrium (LD), calculated by r² (Hill and Robertson 1968) and distance between sites. We find that r²does not decrease with distance in P. tetraurelia, P. sexaurelia, and P. caudatum (P. multimicronucleatum isolates lacked sufficient number of polymorphic sites; supplementary fig. 16, Supplementary Material online), consistent with the expectation under no recombination. Next, we conducted the four-gamete test (FGT) (Hudson and Kaplan 1985), which detects recombination by searching for pairs of polymorphic sites with all four segregating haplotypes and assumes that such pairs of sites must have arisen via recombination (under the infinite-sites model). We found all four gametes at 2 pairs of sites (1.45 × 10⁻⁵ of all pairs) in P. tetraurelia, 312,781 pairs (2.14 × 10⁻² of all pairs) in P. sexaurelia, 42,647 pairs (1.21 × 10⁻² of all pairs) in P. caudatum, and 0 pairs of sites in P. multimicronucleatum. The observed variation between species correlates well with levels of total sequence diversity and indicates the bias in power to detect recombination toward species with more sequence variation. Although results from the FGT suggest the presence of some recombination, the probability of finding all four gametes does not increase with the distance between the pairs of sites (slope= −7.64 × 10⁻⁹, P = 0.765 for P. sexaurelia; slope = −1.39 × 10⁻⁸, P = 0.792 for P. caudatum), contrary to the expectation under recombination, and is inconsistent with other distance-based analyses. Violation of the infinite-sites model (i.e., recurrent mutation at the same site) can result in the presence of pairs of sites with all four alleles. The FGT works best for species and genomes whose recombination rate is much larger than the mutation rate, which may not be the case in Paramecium mitochondria.

In order to test whether the presence of four gametes was more likely to be caused by recombination or mutation, we used LDhat (McVean et al. 2002) to test for recombination and estimate 2N_er, using values of 2N_eμ estimated by nucleotide diversity as calculated by Johri et al. (2017) (supplementary table 5, Supplementary Material online). Here, r is the recombination rate, μ is the mutation rate, and N_e is the effective population size of the species under consideration. 2N_er estimated under the finite-sites mutation model (Wakeley 1997; Hudson 2001) was found to be 0.0 for all four species, and is consistently 0.0 in all nonoverlapping windows (1,000 bp) spanning the genome. We also used permutation tests (McVean et al. 2002) to detect recombination. The idea behind these tests is that in the absence of recombination the relative position of SNPs would not change inferred values of statistics used to measure recombination. Thus, in the absence of recombination, statistics like sum-of-distance between sites with all four gametes (G4) (Meunier and Eyre-Walker 2001), and correlation between r² or D′ and physical distance, would not be significantly different from an expectation obtained by randomly shuffling SNPs and computing these statistics. We find no statistical significance for the presence of recombination using permutation tests (supplementary table 5, Supplementary Material online) and thus conclude that there is no recombination in Paramecium mitochondria. It should be noted that the final sets of SNPs used to infer the presence of recombination consist of 526 SNPs in P. tetraurelia, 5,401 SNPs in P. sexaurelia, 3,353 SNPs in P. caudatum, and 545 SNPs in P. multimicronucleatum. These sets of SNPs are not only large in absolute terms, but are also uniformly distributed across the genome (supplementary fig. 16, Supplementary Material online), providing a lot of power to detect any recombination.

Efficacy of Selection in Mitochondria versus the Nucleus

Finally, we asked whether mitochondrial genes experience weaker efficacy of purifying selection compared with nuclear genes, as would be expected due to smaller effective population size and the lack of recombination in the mitochondria. We did so by comparing multiple statistics in each of the two genomes. Because our statistics included divergence at synonymous sites which are saturated for most species pairs, we conducted these analyses primarily on P. tetraurelia where values of divergence were calculated with respect to the closest outgroup species (d_s), P. biaurelia, for which we had available sequences in both the nucleus and mitochondria. Average d_s values for the set of nuclear and mitochondrial genes were found to be 0.885 and 1.174, respectively (table 2), a small but significant difference. In order to control for differences in d_s driving the patterns, we also conducted all analyses restricted to genes with d_s < 1.0. For the set of genes with d_s < 1.0, the average value of d_s is not significantly different between the two genomes, 0.636 among the nuclear and 0.658 among the mitochondrial genes. A potential caveat of comparing the efficacy of purifying selection between all genes present in the mitochondria versus nucleus is that we might instead be measuring differences in strength of purifying selection. In order to correct for that, we also compared nuclear genes with similar functions to those in the mitochondria. We compared the 14 mitochondrial genes with ∼13–27 (of total 87) nuclear genes involved in the oxidative phosphorylation (OXPHOS) pathway, as well as the 8 ribosomal genes in the mitochondria with ∼77–430 (of total 585) nuclear genes that encode for structural components of the ribosomes.

Table 2.

NI of Mitochondrial and Nuclear Genes in Paramecium tetraurelia, Calculated Using Multiple Statistics

	All (excluding Ymf)	dS < 1.0	OXPHOS	Ribosomal
dS
Nuclear	0.885	0.636	0.774	0.404
Mitochondrial	1.174	0.658	1.324	1.581
P	2.74 × 10−6	0.807	0.035	4.22 × 10−5
dN/dS
Nuclear	0.079	0.086	0.029	0.088
Mitochondrial	0.082	0.142	0.052	0.044
P	0.307	0.210	0.208	0.208
πn/πs
Nuclear	0.235	0.235	0.200	0.092
Mitochondrial	0.163	0.180	0.127	0.062
P	0.183	0.618	0.223	0.387
NIπ [(πn/πs)/(dN/dS)]
Nuclear	4.018	3.436	6.712	1.653
Mitochondrial	2.294 (2.495)	1.218	2.384	2.248
P	0.124 (0.537)	0.024	0.422	0.262
NI [(Pn/Ps)/(Dn/Ds)]
Nuclear	2.145	2.271	6.087	2.873
Mitochondrial	1.528 (1.855)	1.294	1.696	2.195
P	0.272 (0.817)	0.208	0.018	0.072
NIDoS [Dn/(Dn+Ds)] – [Pn/(Pn+Ps)]
Nuclear	−0.041	−0.053	−0.227	−0.136
Mitochondrial	0.026 (0.024)	0.046	0.001	0.073
P	0.272 (0.738)	0.208	0.010	0.072

note.—All P values (corrected for multiple testing by Holm–Bonferroni method) represent comparisons between nuclear and mitochondrial statistics. Statistically significant scores (P < 0.05) are underlined. Values in round brackets refer to analyses done with the set of genes excluding Ymf genes.

Average d_N/d_s is similar or slightly lower for genes in the nucleus (mt = 0.082, nuc = 0.079, P = 0.307; table 2), with the exception of the well-conserved ribosomal genes, which have lower d_N/d_s in the mitochondria (mt = 0.044, nuc = 0.088, P = 0.208). None of these differences in d_N/d_s is significant, suggesting that the average amount of purifying selection experienced by genes in the nucleus is not significantly different from that experienced by those in the mitochondria.

The efficacy of purifying selection can be measured by estimating the fraction of segregating nonsynonymous polymorphisms that undergo fixation. Such a measure can be calculated using the neutrality index (NI = [P_n/P_s]/[D_n/D_s], where P_n and P_s are the number of nonsynonymous and synonymous polymorphisms, respectively; D_n and D_s are the number of nonsynonymous and synonymous fixed differences, respectively). The NI is usually found to be larger than 1.0 for genes experiencing purifying selection because mildly deleterious variants are allowed to segregate among individuals, but rarely fix in populations. Higher values of the NI suggest that smaller proportion of segregating nonsynonymous variants are allowed to fix in the population, indicating stronger efficacy of purifying selection. The number of branch-specific substitutions at synonymous (D_s) and nonsynonymous (D_n) sites was calculated by ancestral reconstruction, using P. tetraurelia, P. biaurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum, for both the nucleus and the mitochondria, and restricted to sites for which changes could be confidently inferred along specific branches. The advantage of using ancestral reconstruction to infer D_n and D_s is that particular sites that are very fast evolving can be excluded from the analysis, thus minimizing problems arising due to saturation of divergence at synonymous sites. Comparing the distributions of values of NI in the mitochondrion and nucleus shows that overall there is no significant difference between the two sets when all genes are included (whether we include or exclude Ymf genes) in P. tetraurelia (table 2). However, for genes involved in OXPHOS, nuclear genes appear to be experiencing significantly decreased efficacies of purifying selection than mitochondrial genes (mt NI = 1.696, nuc NI = 6.087, P = 0.02).

We also calculated the NI as (π_n/π_s)/d_N/d_s) (Betancourt et al. 2012; denoted by NI_π below) for each gene with d_N/d_s obtained from pairwise comparison with respect to P. biaurelia. In this case all sites contribute to the analysis, but the maximum likelihood estimate of pairwise d_N/d_s can minimize biases due to saturation of d_s. Again, we observe that NI of nuclear genes is either similar or higher than those of mitochondrial genes. Genes involved in OXPHOS in the nucleus have higher but not significantly different NI_π compared with those in the mitochondria (mt = 2.384, nuc = 6.712, P = 0.422).

In order to minimize statistical biases that can arise due to sampling, we also estimated a variation of the NI: NI_DoS = D_n/(D_n + D_s) − P_n/(P_n+P_s) (Stoletzki and Eyre-Walker 2011), where more negative values represent stronger efficacy of purifying selection. As previously observed, we find no significant difference between the NI of genes in the nucleus versus mitochondria except for OXPHOS genes, where nuclear genes have significantly lower values of NI_DoS than mitochondrial genes (mt NI_DoS = 0.001, nuc NI_DoS = −0.227, P = 0.01)

This suggests that there are more deleterious variants segregating in the nucleus than in the mitochondrion in P. tetraurelia. Overall, mitochondria appear to experience either similar or stronger efficacy of purifying selection than the nucleus.

Lastly, we used π_n/π_s as a proxy for the efficacy of recent purifying selection. This comparison can be extended to all four species (supplementary table 6, Supplementary Material online). We find that average π_n/π_s of mitochondrial (mt) and nuclear genes (nuc) is not significantly different in P. tetraurelia (mean mt = 0.163; mean nuc = 0.235; P = 0.18) and P. caudatum (mean mt = 0.135; mean nuc = 0.170; P = 0.22), respectively. However, mean π_n/π_s in the mitochondrial genes is significantly lower than that of nuclear genes in P. sexaurelia (mean mt = 0.051; mean nuc = 0.268; P < 2.2 × 10⁻¹⁶) and P. multimicronucleatum (mean mt = 0.099; mean nuc = 0.178; P = 1.87 × 10⁻³), respectively, suggesting that mitochondrial genes might be under stronger purifying selection than nuclear genes.

As a side note, values of NI for P. sexaurelia, P. caudatum, and P. multimicronucleatum were consistently found to be much less than 1.0, likely due to underestimation of changes at synonymous sites. We aimed to reduce the bias caused by saturation of D_S by recalculating NI only in the mitochondrion using all 13 taxa (in order to break up longer branches). We continue to recover extremely low values of NIs (supplementary table 7, Supplementary Material online) and do not obtain values close to those obtained using NI_π, suggesting that the absolute values of NI can be misleading and have to be interpreted with caution.

Discussion

In this study, we greatly expand the set of sequenced ciliate mitochondrial genomes by presenting a wider sampling of the mitogenome diversity within the Paramecium genus. Using this wealth of sequence data, we characterize the diversity and conservation of genome organization and gene content, and we study in depth the population genetic characteristics such as mutational and selection pressures acting on mtDNA within the genus.

Ciliate-Specific Mitochondrial Genes

Paramecium mitochondrial genomes possess 16 lineage-specific ORFs (referred to as Ymf genes) that have no known homologs in nonciliate species and lack assigned functions, but are nonetheless conserved across Paramecium and Tetrahymena species. Other ciliates like Oxytricha have also been found to have unidentified ORFs (Swart et al. 2012), but not all Ymf genes have homologs identified in ciliates outside of Oligohymenophora (which contains both Tetrahymena and Paramecium). We find that Ymf genes are evolving faster, mostly due to relaxed purifying selection, as Ymf genes are not significantly more likely to undergo positive selection than other genes. Shedding light on the identity of the Ymf genes could possibly indicate entirely new sets of genes present in the ancestor of Oligohymenophorean mitochondrial genome.

We found that parts of five of the 16 Ymf genes show homology to standard mitochondrial proteins, especially ribosomal proteins. As ciliates are evolutionarily distant from most model organisms in the eukaryotic tree, it is possible that some Ymf genes are ribosomal subunits or genes belonging to the NADH complex and are simply not identifiable because of being highly diverged. Similar findings have been reported for other protozoan mitochondrial genomes in the past (de Graaf et al. 2009; Pombert et al. 2013; Burger et al. 2016; Skippington et al. 2017) as well as some nonprotist species like bivalve molluscs and cnidarians (Shao et al. 2006; McFadden et al. 2010; Kayal et al. 2012; Milani et al. 2013) that have diverged substantially from the most well-studied mammalian mitochondrial genomes. In some species, RNA editing (including insertions and deletions) can be substantial (Gray 2003), which can mask the proteins encoded from identification from genomic sequence data. We, however, found no evidence of RNA editing, consistent with a previous report in P. tetraurelia mitochondria (Orr et al. 1997), thus it is unlikely to account for the observed divergence. An interesting possibility is that most of the Ymf genes are either part of or interact with the ATP synthase complex. In Tetrahymena thermophila, the ATP synthase has been reported to form an unusual structure possessing completely novel subunits whose orthologs are not identifiable in other organisms, some of which are Ymf genes (Balabaskaran Nina et al. 2010). It is therefore possible that ciliates possess structurally unique proteins that perform relatively conventional functions in the mitochondria, but are difficult to identify based on other known sequences. Although the origin of these genes is unclear, it is intriguing that a set of such fast evolving genes is preserved across highly diverged species.

Change in Mutation Spectrum and Nucleotide Composition

We used a combination of previous MA studies and our population-genomics data to determine that the change in nucleotide composition of P. aurelia mitochondrial genomes toward higher GC is most likely the result of changes in mutational biases. Unfortunately, due to our modest sample sizes, some of the SNPs observed as singletons in our data may in fact be fairly common in the population at large. Our estimates of mutation spectra might thus be biased by selection. MA lines in P. caudatum would help further disentangle these forces. Because ciliates have very low mutation rates, their MA study requires a large number of generations to produce only a handful of mutations, making it very difficult to obtain precise estimates of the mutation spectrum. The most feasible strategy to more precisely estimate these spectra in Paramecium would thus be to acquire larger population samples in order to observe lower-frequency variants. We also note that although an alternative explanation for higher GC content in the P. aurelia species could be biased gene conversion, we found no evidence of mitochondrial recombination (including noncrossovers) in the P. aurelia species.

An interesting question raised by our results is how fast the mutation spectrum in mitochondria evolves across species (Lynch et al. 2008; Montooth and Rand 2008). Reanalyzing data from previous work on MA lines in mitochondria of other model organisms presents a number of interesting observations (table 1). First, within opisthokonts, there is a huge variation in mutation bias (m) in the mitochondria, ranging from nearly 0 (strongly biased toward GC) to values much larger than 1 (biased toward AT; table 1). However, most species except Saccharomycescerevisiae have a mutation bias toward A/T, consistent with most mitochondrial genomes being AT-rich. Second, across opisthokonts, we find that selection can have a significant impact on mitochondrial GC content. The effect of selection (S) can be observed by comparing observed genome-wide nucleotide composition (or specifically at 4-fold degenerate sites) with the expected composition under mutation equilibrium (table 1). Again, in most species selection favors higher G/C, but in S. cerevisiae there appears to be strong selection favoring A/T genome-wide. Thus, there might not be a universal direction for mutation bias or selection for nucleotide composition in mitochondria. Finally, closely related species in other lineages have been shown to have significant differences in mutation bias in the mitochondria as observed for example between Caenorhabditis elegans and Caenorhabditis briggsae (mutation bias recalculated by combining observations from Howe et al. [2010] and Konrad et al. [2017]). All of these observations suggest that it may not be such an unusual event for the mutational biases to shift away from AT along the long branch leading to the P. aurelia species. However, the exact nature of the biochemical mechanisms responsible for remains an intriguing open question for future research.

Changes in mutation spectra could occur due to environmental variables resulting in differential mutagenic pressures, or due to the differences in DNA repair processes. However, we note that nuclear GC composition is very similar across all Paramecium species; the change in GC composition in the P. aurelia complex has only occurred in the mitochondrial genome. This quite strongly suggests that environmental conditions are unlikely to account for the observed differences, otherwise we should expect to find concordant differences between nuclear genomes. In addition, P. caudatum and the P. aurelia species very often co-occur in the same lakes, sometimes even from the same sampling site. Changes in mitochondrial DNA repair systems along the P. aurelia branch are therefore more likely to be the explanation. One possibility is that the nuclear whole-genome duplication events that happened prior to the P. aurelia radiation have resulted in an expansion in the number and variety of DNA repair enzymes, which has resulted in changes in the DNA repair systems operating in mitochondria specifically. Indeed, a number of genes functioning in DNA repair processes are retained in more copies in the P. aurelia than in P. caudatum nuclear genomes (supplementary table 8, Supplementary Material online). However, whether any of these proteins have been neo- or subfunctionalized to play a role in mitochondrial DNA repairs cannot be determined at present due to the very limited knowledge of DNA repair mechanisms in ciliates in general, and in their mitochondria more specifically; further experimental studies will be needed to answer these questions.

Similar Efficacy of Purifying Selection Experienced by the Mitochondria and Nucleus

Mitochondrial genomes are often nonrecombining and are usually passed on via uniparental inheritance. Mitochondria are therefore expected to have lower effective population sizes than that of the nucleus within the same organism (Lynch and Blanchard 1998; Neiman and Taylor 2009). One consequence of reduced effective population sizes is an increased probability of segregation and fixation of slightly deleterious mutations. There have been multiple contradictory reports on whether mitochondrial genes experience stronger or weaker purifying selection than the nuclear genes. Studies examining a small number of protein-coding loci across a large number of species have concluded that mitochondria experience less effective purifying selection than the nucleus (Weinreich and Rand 2000; Betancourt et al. 2012; Popadin et al. 2013). However a recent study conducted on multiple individuals and whole genomes in Drosophila melanogaster and humans found no significant difference between the efficacy of purifying selection in the mitochondrial versus nuclear protein-coding genes (Cooper et al. 2015).

We confirm the absence of recombination in mitochondria of three species of Paramecium. Previous studies in P. tetraurelia (Adoutte et al. 1979; Barth et al. 2008) and P. primaurelia (Beale et al. 1972) had reached similar conclusions using a small number of markers. It should be noted that we do not detect recombination with extant sequence variation, which is consistent with the observed uniparental inheritance in Paramecium species. However, the lack of fusion of mitochondria presents a barrier to recombination within the cell. Thus, even if recombination was occurring at an extremely low rate, it might be difficult to detect it. Despite the lack of recombination in the mitochondria, the efficacy of selection in the mitochondria is similar to if not stronger than that of the nucleus.

Our results seem to be in discordance with theoretical predictions according to which one would expect the efficacy of selection in the mitochondria to be lower than in the nucleus. One possibility is that because there are multiple copies of mitochondrial genomes within a cell, recessive deleterious mutations might not produce a phenotypic effect unless they reach sufficiently high frequencies. A subsequent bottleneck could result in either transmission of highly fit mitochondrial mutations, or possibly the most deleterious. This increase in variance of fitness of individual cells would result in efficient selection between cells, eventually leading to stronger purifying selection (Stewart et al. 2008; Wai et al. 2008; Ghiselli et al. 2013; Stewart and Larsson 2014). We might thus expect mitochondrial genes to experience more efficient purifying selection than the nuclear genes. However, the severity of mitochondrial bottleneck in Paramecium is not yet clear (see below). Another possibility is that we observe stronger efficacy of selection in the mitochondria because the magnitude of negative selection itself is stronger in the mitochondrial genes (Popadin et al. 2013), which differ from the average nuclear gene in several key ways (Adrion et al. 2016). Most mitochondrial genes are expressed at higher levels compared with nuclear genes (Havird and Sloan 2016). Mitochondrial encoded proteins (cox1 and cox2) that are part of the OXPHOS pathway are core enzyme catalytic subunits (Tsukihara et al. 1996; Zhang and Broughton 2013; Havird and Sloan 2016). Finally, most genes retained in the mitochondria encode for highly hydrophobic proteins and have a high GC content (Johnston and Williams 2016). Thus, finding comparable sets of genes between these two genomes is admittedly difficult (but see Lynch 1996; Lynch 1997).

Lastly, our results need to be evaluated in the light of Paramecium-specific life cycle. In Paramecium, there is almost no exchange of cytoplasm during conjugation (Koizumi and Kobayashi 1989; Meyer and Garnier 2002). Thus, it appears that both parents pass on their mitochondria to their respective offspring, without exchange or degradation of mitochondrial genomes. In addition, mitochondria double before binary fission (Perasso and Beisson 1978) and appear to be distributed randomly across the daughter cells. Evidence of the latter comes from experiments where P.aurelia cells were injected with two different mitochondrial genotypes, and were proliferated in nonselective media. Paramecium cells were observed to maintain the two populations stably if the two mitochondrial genotypes were equally fit when present as pure populations in cells (Adoutte and Beisson 1972; Adoutte et al. 1979). On the other hand, if one of the genotypes was less fit than the other, it would be lost, with the time of loss being proportional to the fitness difference (Adoutte and Doussiere 1978). These studies provide indirect evidence for equal and random distribution of mitochondria in the two daughter cells. Thus, mitochondria do not appear to undergo bottlenecks during any stage of the life cycle in Paramecium, and experience no associated reduction in effective population size. On the other hand, Paramecium species frequently undergo asexual reproduction, thereby reducing the nuclear effective population size. Therefore, effective population sizes of mitochondrial and nuclear genomes may be more similar in Paramecium than in many other organisms. Due to the combination of sexual and asexual reproduction, under equilibrium conditions, expected heterozygosity in the nucleus would take values between 4N_eμ_n – 2N_eμ_n. Similarly, in the mitochondria, expected hetrozygosity would be ∼2N_eμ_m. Assuming that the ratio of divergence at silent sites can be used as a proxy for the ratio of mutation rate between the two compartments, it is possible to approximately estimate the ratio of effective population sizes of the two compartments as N_e(m)/N_e(n) = y × (π_m/π_n)/(d_m/d_n), where y would be some number between 1 and 2. For P. tetraurelia, for which we have relatively closer outgroup species and thus more reliable divergence estimates, our estimated range of N_e(m)/N_e(n) is 0.94–1.88. Although underestimation of neutral divergence in the mitochondria relative to the nucleus could skew our inference slightly, the above calculation suggests that effective population sizes of mitochondria may be similar or, larger than that of the nucleus in Paramecium.

We therefore conclude that our finding of similar or stronger efficacy of selection in the mitochondria relative to the nucleus in Paramecium may lie within theoretical expectations given Paramecium’s unique life cycle and mode of mitochondrial transmission. A better understanding of the Paramecium life cycle in the wild might help build more appropriate null expectations in the future. Our results suggest the possibility that unicellular eukaryotes in general may have larger mitochondrial than nuclear effective population sizes and more efficacious purifying selection in the mitochondria might be more common than believed.

Materials and Methods

Genome Sequencing and Assembly

Single isolates of P. jenningsi (strain: M), P. octaurelia, P. decaurelia (strain: 223), P. dodecaurelia (strain: 274), P. novaurelia (strain: TE), and P. quadecaurelia (strain: 328) were used to extract macronuclear DNA. DNA extraction, sequencing library preparation, and genome sequencing were previously described (Johri et al. 2017). Sequencing reads were assembled using SPAdes (Bankevich et al. 2012; version 3.5.0) after removing potential adapter sequence with Trimmomatic (Bolger et al. 2014; version 0.33). Mitochondrial contigs were identified from the resulting assemblies by BLAST (Altschul et al. 1997) searches against the published P. caudatum and P. tetraurelia mitochondrial genomes.

SNP Detection

Whole-genome sequencing raw reads from five to ten isolates of P. tetraurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum were downloaded from SRA (SRA accession: SRR8698631–SRR8698604; Bioproject: PRJNA525710; Biosample: SAMN11059622–SAMN11086832), and SNPs were called as described by Johri et al. (2017), using reference genomes of strain 99 for P. tetraurelia, strain 130 for P. sexaurelia, C104 for P. caudatum, and M04 for P. multimicronucleatum. Briefly, reads were trimmed using Trimmomatic (version 0.36) (Bolger et al. 2014) and mapped to reference genomes using bwamem (0.7.12) (Li and Durbin 2010) under default parameters. Duplicate reads were marked using picard (2.8.0) (https://broadinstitute.github.io/picard/). Sites were only considered for further analysis if the mapping quality was above 30, base quality was above 20, per-base alignment quality was above 15, and the sum of the depth of coverage for all individuals was about five times the number of individuals and less than twice the average population coverage. Variants were called using bcftools (Li et al. 2009) and filtered using vcftools (Danecek et al. 2011). Only those sites were considered whose quality value (–minQ) was above 20. Genotypes whose genotype quality score (–minGQ) was <30 or those that were supported by <4 reads (–minDP) were excluded or considered missing.

Mitochondrial Genome Annotation

Genome annotation was carried out as follows. Protein-coding genes were identified by generating all ORFs longer than 60 amino acids in all six frames, using the Mold, Protozoan, and Coelenterate Mitochondrial Code (i.e., UGA codes for W instead of being a stop codon) and all alternative start codons specific to Paramecium (AUU, AUA, AUG, AUC, GUG, and GUA), and retaining the longest ORFs associated with each stop codon. BlastP was then used to identify homologs of annotated mitochondrial proteins in P. tetraurelia and P. caudatum. Additional ORFs were identified by imposing the requirement that their length exceeds 100 amino acids, and subsequently annotated using BlastP against the nonredundant protein sequences (nr) database and HMMER3.0 (Eddy 2011) scans against the PFAM 27.0 database (Finn et al. 2014). tRNA genes were annotated with tRNAscan-SE (Schattner et al. 2005; version 1.21), using the “Mito/Chloroplast” source. rRNA genes were identified using Infernal (Nawrocki et al. 2009; version 1.1.1).

Identification of Telomeres

Telomeric repeats were identified as follows. The first and the last 200 bp of each de novo assembled mitochondrial genome were used as input to the MEME de novo motif finding program (Bailey et al. 2009; version 4.6.1), which was run with the following parameters: -maxw 25 -dna -nmotifs 5 -mod anr. The repetitive units defined that way were then manually aligned to each other and refined to arrive at final telomeric repeats comparable across all species.

RNA-Seq Analysis

For each species, sequencing reads were aligned against a combined Bowtie (Langmead et al. 2009) index containing both the nuclear and mitochondrial genomes using TopHat2 (Kim et al. 2013; version 2.0.8) with the following settings: –bowtie1 –no-discordant –no-mixed –microexon-search –read-realign-edit-dist 0 –read-edit-dist 4 –read-mismatches 4 –min-intron-length 10 –max-intron-length 1000000 –min-segment-intron 10 –min-coverage-intron 10. Custom python scripts were then used to identify sequence variants relative to the mitogenomes assemblies.

Building Phylogenetic Trees

Nucleotide sequences were extracted, aligned using MUSCLE, and concatenated. Missing data were encoded as “N.” RAxML was used to build the tree with GTRGAMMA as the substitution model. Bootstrap values for 1,000 replicates were obtained via the fast method recommended by RAxML with the following command line:

raxmlHPC -f a -s sequences.fasta –n sequences_boot -m GTRGAMMA -T 50 -p 31 -x 7777 -N 1000

Estimation of dN/dS, D_n, D_s, P_n, P_s, and π_n/π_s

dN/dS was estimated across the phylogeny using CODEML, PAML (Yang 2007; version 4.9a) and for the P. aurelia species it was also estimated pairwise with respect to the closest outgroup species (denoted as d_N/d_S) using yn00, PAML. π_n/π_s was obtained for all protein-coding genes in the four species—P. tetraurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum, where total number of changes in synonymous sites was >1. This filter was executed in order to reduce errors in π_n/π_s due to very low values of synonymous polymorphisms in a gene.

D _n and D_s, the number of nonsynonymous and synonymous changes, were inferred by performing ancestral reconstruction at each site, and then counting branching-specific substitutions. Ancestral reconstruction (PAML, baseml, GTR model) was conducted over the phylogeny of all available taxa. For comparing statistics between mitochondrial and nuclear genes, the ancestral reconstruction was performed over the same set of taxa for both, that is, over P. tetraurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum. For all analyses that involved ancestral reconstruction, only sites whose posterior probability of the inferred ancestral state ≥ 0.85 were used in the analyses—this filter was imposed for counting synonymous and nonsynonymous polymorphisms (P_s and P_n, respectively) as well as divergent sites (D_n and D_s). There may be some concern that ancestral reconstruction could end up biasing the ratio of D_n/D_s as many more changes at synonymous sites might result in lower confidence in inferring ancestral states at synonymous but not nonsynonymous sites. Such a bias would increase D_n relative to D_s and thus decrease values of NI. Therefore, all analyses involving D_n and D_s were also performed including all sites, with no filter, and results remained unchanged.

Calculation of Multiple Estimators of NI and Statistical Tests

Several estimators of NI have been proposed in order to counter different biases. The estimators we used were calculated as follows:

Simple neutrality index, NI = (P_n/P_s)/(D_n/D_s) (Rand and Kann 1996).

NI_π = (π_n/π_s)/(d_N/d_S), where d_N/d_S was calculated pairwise, with respect to closest outgroup species (Betancourt et al. 2012).

NI_TG = Σ_i [D_si.×P_ni /(P_si + D_si)] /Σ_i [P_si.×D_ni /(P_si + D_si)], where i is the ith gene (Tarone 1981; Greenland 1982).

NI_DoS = [D_n/(D_n + D_s)] − [P_n/(P_n + P_s)] (Stoletzki and Eyre-Walker 2011).

For MK test (McDonald and Kreitman 1991), Fisher’s exact test in R (R-Core-Team 2014) was used to test significance. In all cases, P value was corrected by Bonferonni–Holm method (Holm 1979) for multiple tests, using R.

Mutation Spectrum from MA Lines

MA line experiments for P. tetraurelia had previously been published (Sung et al. 2012), and results from their analysis were used directly in this study.

MA experiments carried out in order to obtain nuclear mutation rates in P. biaurelia and P. sexaurelia (Long, Doak, et al. 2018) were reanalyzed as follows. Sequencing reads were assembled for each MA line individually and mitochondrial contigs identified as described above. A composite consensus mitochondrial genome sequence was then created from the individual assemblies by creating multiple sequence alignments of all mitochondrial contigs using MAFFT (Katoh and Standley 2013; version 7.221) and retaining the most frequent base for each alignment column (with the exception of telomeres, which were manually curated). Adapter-trimmed reads were then aligned in a 2 × 100 bp format against a combined Bowtie index, containing a combination of the nuclear and consensus mitochondrial genomes, allowing for up to three mismatches and retaining only unique reads. Putative mutations were identified by requiring that any variant is supported by at least three nonredundant read pairs on each strand, is supported by not more than four times more reads on one strand than on the other, and is also observed in ≥5% of reads covering a given position. Telomeric sequences were excluded due to an excessively high number of sequence variants observed in those regions.

Mutation Spectrum from Population Genomics Data

For each SNP, the ancestral allele was inferred by performing ancestral reconstruction on the 13-taxa phylogeny to predict the nucleotides on internal nodes (see above). The ancestral allele was used to infer the derived allele segregating in P. tetraurelia, P. sexaurelia, P. caudatum, and P. multimicronucleatum. This analysis was restricted to sites where the ancestral nucleotide was inferred with confidence score ≥0.90, where the derived allele was present in only a single individual, that is, was a singleton, and was at 4-fold degenerate site. Of these, we counted all mutations that were from G/C to A/T or from A/T to G/C, with respect to the total number of utilizable sites that were counted according to the same criteria as above. Care was also taken to remove all sites that were part of overlapping ORFs.

Calculation of Bias in Mutation Spectrum

Mutation rate of A/T → G/C (u) and G/C → A/T (v) mutations was calculated as follows:

u = (number of A/T → G/C mutations)/(total number of utilizable A/T sites).

v = (number of G/C → A/T mutations)/(total number of utilizable G/C sites).

Mutation bias toward A/T (m) was calculated as, m = v/u, and the expected equilibrium G/C content was calculated as 1/(1+m) following Lynch (2007).

S = 4N_es (or 2N_es), is the population-scaled strength of selection toward A/T nucleotides and can be calculated using the equation, P_AT = 1/(1 + m⁻¹e^−S), where P_AT is the observed fraction of A/T sites and S is the average selective advantage of A/T over G/C nucleotides (Bulmer 1991). Thus S = −ln[(m.(1−P_AT))/P_AT] = ln(P_AT/P_GC/v/u).

Recombination Analyses

All analyses to detect recombination were restricted to SNPs that were biallelic, homozygous, and had a known ancestral state. The statistic (r²) to measure LD was calculated as r² = (f_Aa – f_A × f_a)²/[f_A × f_a × (1 − f_A)×(1 − f_a)]. The program “pairwise” in LDhat 2.2 (McVean et al. 2002) was used to infer recombination rates using the permuted composite likelihood test as well other permutation tests. These tests were performed under both the gene conversion (average tract length: 500) and crossover models with 2 values of θ (=4N_eμ) for each species: The closest allowed θ value lower than that estimated from nucleotide diversity values, and the closest higher value.

Identifying Nuclear Genes Belonging to OXPHOS Pathway and Ribosomal Complex

KEGG (Kanehisa et al. 2017) was used to obtain the full list of genes that are part of complexes involved in OXPHOS (complexes I–V) for P. tetraurelia. A total number of 87 genes in P. tetraurelia were obtained, and their corresponding orthologs were identified in other species. For genes encoding proteins that are part of the ribosomal complex, we used the PANTHER (Mi et al. 2017) annotation obtained in a previous study (McGrath, Gout, Johri, et al. 2014) for all species and selected all genes that were structural constituents of ribosome. This allowed us to start with a set of 585 genes in P. tetraurelia, 564 in P. sexaurelia, and 213 genes in P. caudatum. However, analyses requiring orthologs from all three species were conducted with a smaller subset of genes.

Data Availability

Assembled mitogenomes and protein-coding gene annotations can be accessed through Zenodo (https://doi.org/10.5281/zenodo.2539699). Mitochondrial variant files can be accessed through Github (https://github.com/paruljohri/Paramecium_mitochondrial_SNPs).

Supplementary Material

Supplementary data are available at Genome Biology and Evolution online.