Sequencing and assembly.

Using physical isolation methods, MAC were purified from a culture of T. thermophila strain SB210 and used to create multiple differentially sized shotgun sequencing libraries (Table S1). Construction of large (greater than 10 kb) insert libraries was not successful—a common problem in working with AT-rich genomes. Approximately 1.2 million paired end sequences were generated from the libraries and assembled using the Celera Assembler [37]. In an initial assembly, the mitochondrial genome (mtDNA; which was present due to some contamination of the MAC preparation with mitochondria) and the highly amplified rDNA chromosome did not assemble well compared to the published sequences of these molecules [38,39]. This was probably because contigs from these molecules had higher depths of coverage than those from other chromosomes, which caused the Celera Assembler to treat them as repetitive DNA. Thus we divided sequence reads into three bins (mtDNA, rDNA, and bulk MAC DNA) and generated assemblies for each bin separately. This resulted in a moderate improvement, and the three separate assemblies were thus used for all subsequent analyses. Detailed sequence and assembly information is presented in Tables 1 and S2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1.

Important Genome Statistics

https://doi.org/10.1371/journal.pbio.0040286.t001

The bulk MAC assembly contains 1,971 scaffolds (contigs that have been linked into larger pieces by mate pair information) with a total estimated span of 104.1 Mb. Perhaps most important, using a combination of computational and experimental identification of telomeres, we have found that many scaffold ends correspond to chromosome ends. One hundred twenty-five scaffolds, encompassing 44% of the assembled genome length, are telomere-capped at both ends and thus likely represent complete MAC chromosomes. One hundred twenty additional scaffolds, encompassing another 31% of the genome, are telomere-capped at one end (Tables 1 and S3).

Assembly accuracy and completeness.

Overall, all analyses indicate that the bulk MAC assemblies are highly accurate. For example, all 75 MAC loci that are in distinct genetic co-assortment groups (and thus should be on different chromosomes [40]) map to different scaffolds, and all pairs of loci that coassort (and thus should be on the same chromosome) either map to the same scaffold or to two non–fully capped scaffolds whose cumulative size is less than that of the corresponding MAC chromosome (Table S4). For the 24 completely assembled chromosome scaffolds for which we know the corresponding chromosome physical size, there is a very strong correlation between physical size and assembly length. In addition, there are no cases where a scaffold is significantly longer than the physical size of the corresponding chromosome (Figure 3A). Finally, all of the 96 MIC sequences known to be adjacent to Cbs sites [24,41,42] that matched to a MAC scaffold did so only at the scaffold's end.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Scaffold Sizes

(A) Scaffold sizes versus MAC chromosome size. Blue diamonds represent scaffolds capped by telomeres on both ends. Red squares and green triangles represent incomplete scaffolds capped by telomeres at one or neither end, respectively.

(B) Size distribution of scaffolds capped by telomeres on both ends.

https://doi.org/10.1371/journal.pbio.0040286.g003

The general accuracy of the assemblies indicates that many of the potential difficulties discussed in the Introduction were not significant. For example, we see little evidence for polymorphism among reads, which is likely a reflection of the use of an inbred strain and the process of phenotypic assortment, which leads to whole-genome MAC homozygous lineages [43]. Also, searches for known MIC-specific sequences indicate that the amount of MIC contamination is very low (e.g., Cbs junctions are at 0.044× coverage which is approximately 200-fold less than the bulk MAC chromosomes) and limited to small contigs (most less than 5 kb). The uniform depth of contig coverage and accuracy of assemblies also suggest that the chromosomes are present in roughly similar copy number and that only limited amounts of repetitive DNA are present in the MAC, both of which are discussed further below.

The total scaffold length is much smaller than the predicted genome size of 180 to 200 Mb [14]. Given the accuracy of the assemblies, the large number of chromosomes partially or completely capped, and the fact that all (more than 200) known MAC DNA sequences are found in the assemblies, we conclude that the assemblies represent a very large (more than 95%) fraction of the genome. We conclude therefore that previous genome size estimates were inaccurate (which is not surprising given that they were made almost 30 years ago) and that the genome is close to 105 Mb in size. It is possible, however, that some chromosomes or regions were underrepresented in our libraries due to purification or cloning bias, and thus one cannot infer the absence of any particular gene or feature simply due to its absence from our current assemblies.

Estimating the number of MAC chromosomes.

The total number of MAC chromosomes is unknown. The telomere-capping of scaffolds allows us to place a minimum boundary on this number at 185 (125 plus half of 120). One way of estimating the actual number is through analysis of the non–rDNA telomere-containing reads; 3,328 such reads can be linked to a total of 370 scaffold ends. This corresponds to approximately 9-fold coverage (3,328/370), which is not significantly different from the bulk MAC chromosome coverage of 9.08, indicating that there is no significant underrepresentation of telomere reads (Tables 1 and S3). Thus since there are 4,058 such reads total (the others could not be linked), we estimate that there are approximately 451 telomere ends (4,058/9), and thus that there are approximately 225 chromosomes (451/2). An independent estimate of the actual chromosome number can be made by assuming that the size distribution of fully capped chromosomes (see Figure 3B) is representative of the genome as a whole. Since these 125 capped chromosomes represent 43.5% of the total assembly length, this would predict 287 chromosomes in total (125/0.435). This is likely to be an overestimate, since larger chromosomes are statistically less likely to be in the completely assembled set. Indeed, the average size of completely assembled chromosomes is 359 kb, whereas estimates of the average MAC chromosome size obtained through pulsed-field gel electrophoresis are substantially higher [29,41]. Thus, we conclude that there are between 185 and 287 chromosomes, most likely somewhere near 225.

Absence of many standard global features of eukaryotic chromosomes.

We note that we searched for but could not find many of what are considered standard global features of eukaryotic chromosomes. For example, we could not find sequence or structural features shared across multiple chromosomes that could be considered candidates for centromeric regions. This is consistent with experimental studies [44]. In addition, although in many eukaryotes certain genes and repeat elements cluster near telomeres [45–51], we cannot detect any such clustering here. This is not because there is no variation in these features; for example, GC content (Figure S1) and gene density (Figure S2) do vary greatly. Instead, the absence of similar global structure between MAC chromosomes is likely due to the absence of the processes that help generate the key features of normal eukaryotic chromosomes (e.g., mitosis and meiosis, which in T. thermophila are confined to the MIC).

MAC chromosome copy number is uniform.

The high quality and completeness of the assemblies suggest that copy number variation among at least most MAC chromosomes is relatively small since otherwise the assembler would have treated contigs from overrepresented chromosomes as repetitive DNA. Such uniform copy number is consistent with genetic experimental data for six chromosomes [31], but its generality for all chromosomes has been unknown. We realized that the relative chromosome copy number could be estimated from depth of coverage in our assemblies (assuming that cloning and sequencing success were relatively random). When all scaffolds are examined, the depth of coverage is remarkably uniform (Figure 4). The decrease in uniformity and coverage seen as scaffold size decreases is likely a reflection of both chance low coverage of some regions and some of the small scaffolds being MIC contaminants. When only scaffolds capped by telomeres at both ends are included in the analysis, observed sequence coverage is even more uniform (red diamonds in Figure 4). Although we cannot rule out that some smaller, incompletely assembled chromosomes are maintained at different copy numbers, the observed uniformity indicates that the replication and/or segregation of most or all bulk MAC chromosomes is under coordinated regulation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Depth of Coverage versus Scaffold Size

Black diamonds indicate all scaffolds; red diamonds, scaffolds capped with telomeres on both ends.

https://doi.org/10.1371/journal.pbio.0040286.g004

Protein coding gene predictions.

We identified 27,424 putative protein-coding genes in the genome (Table 2), a high number for a single-celled species. These gene models were tested by aligning expressed sequence tags (ESTs) to the genome assemblies using PASA [52]. We note that most of these ESTs were generated after the models were built (Table S5). Of the 9,122 EST clusters identified, most have either no conflicts with the gene models (49.5%) or relatively small ones (17.7% have a missed exon and 9.8% suggest the models need to be merged or split). Only 408 (4.4%) clusters are intergenic relative to the gene models. Although these could represent missed genes or gene regions, they could also be noncoding RNAs (ncRNAs) or genomic DNA contamination of cDNA libraries. In addition, the predicted and EST-derived introns are quite similar in size distribution except at the short and long extremes (Figure S3), GC content (16.3% versus 16.7%), and splice sites [only a small number (85) of EST-based introns have exceptions to the 5′-GT…AG-3′ junctions assumed by the model—these could simply be sequencing errors]. These analyses indicate that the gene models are relatively robust and should be more than sufficient for making general predictions about the coding potential of this species.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2.

Characteristics of Ab Initio Predicted Genes

https://doi.org/10.1371/journal.pbio.0040286.t002

Two other lines of evidence suggest the predicted gene number is not inflated. First, a large number of the predicted genes have matches to known or predicted genes from other species (14,916 have a BLASTP match with an E-value better than 10⁻¹⁰), and second, experimental studies of mRNA complexity predict transcription of at least 25,000 genes of an average size of 1,200 bp [53]. We also note that the sequence of the largest MAC chromosome of another ciliate, Paramecium tetraurelia, indicates a high coding density, and extrapolation to the complete genome predicts at least 30,000 protein-coding genes [54].

ncRNAs and the use of all 64 codons to code for amino acids.

The ncRNAs found in the genome are listed in Table S6. We call attention to a few new findings. Of the 174 putative 5S rRNA genes (Table S6A), 19 do not correspond to any of the four previously reported T. thermophila sequences [55,56]. These 19 differ from one another by single nucleotide substitutions at 34 positions, as well as by various insertions, deletions, and truncations and may represent pseudogenes. In addition, there are two forms of U2 snRNA present (Table S6C), which we have termed U2 (four genes) and U2var (five genes). Functional RNA gene families are expressed ubiquitously during the T. thermophila life cycle and under stress conditions as well (representative data shown in Figure S4). The largest class is tRNAs with 700 identified (Tables S6B and S6D), a number consistent with hybridization-based estimates [57].

One of the more unusual features of T. thermophila and certain other ciliates is the use of an alternative genetic code in which the canonical stop codons UAG and UAA code for glutamine [58]. The importance and age of this alternative code are reflected in the genome by the presence of 39 tRNAs for these codons. Remarkably, analysis of the genome has also revealed the presence of a tRNA that is predicted to decode the remaining stop codon, UGA. Multiple lines of evidence indicate that this is a functioning tRNA for selenocysteine (Sec), the so-called 21st amino acid. In those eukaryotic species that use Sec, most UGA codons still cause translation termination while those mRNAs that encode Sec-containing peptides have a characteristic stem-loop sequence motif in the 3′ UTR region that directs Sec incorporation [59,60]. The putative T. thermophila tRNA-Sec was identified by analysis of the genome sequence and shown to be transcribed and acylated [61], and we have found that it is expressed and charged and that its charging may be under distinct regulatory control from other tRNAs (Figure S4A). In addition, we identified six T. thermophila genes with in-frame UGA codons that align (after editing of the gene models) with known Sec codons of their homologs from other eukaryotic species and that have the stem-loop consensus and thus are likely to encode selenoproteins. Thus we conclude that UGA is almost certainly translated into Sec, which would make T. thermophila the first organism known to use all 64 triplet codons to specify amino acid incorporation.

Codon and amino acid usage bias.

Although T. thermophila can use all 64 codons, it does not use all equally. The most significant aspect of the codon usage in this species is that the AT-rich codons tend to be used more frequently than others [62,63]. Thus although the AT bias in the genome is strongest in noncoding regions, where selection is thought to be relaxed, it is seen even in coding regions. In fact, the AT pull is so strong in coding regions that amino-acid composition of proteins is shifted toward those coded by codons with high AT content, as seen in other species with extreme AT bias (e.g., [64]). Although the overall codon usage is biased against GC-rich codons, on a gene-by-gene level there is significant variation in the degree of bias. We have identified two dominant patterns to this gene-by-gene variation. The major pattern is that for most genes, the codons used are simply a reflection of the overall AT content of the gene (Figure 5). The variation among genes is due to genomewide variation in AT content (see Figure 5A), although we have been unable to discern a mechanism underlying this variation (e.g., there is no clustering of high or low AT genes near telomeres). There is, however, a less common pattern in the gene-by-gene variation that is very important. There exists a subset of genes (shown in red) that use a common preferred codon set that is different from that of the average gene, and the codons in this set are not strongly correlated to the genes' AT content. Although the existence of such a preferred codon set for this species has been reported [62,63], analysis of the genome allows the set and the genes that use it to be more precisely defined. In total, using a relatively conservative cutoff (Figure 5B), we have identified 232 such genes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Codon Usage

(A) Effective number of codons (ENc; a measure of overall codon bias) for each predicted ORF is plotted versus GC3 (the fraction of codons that are synonymous at the third codon position that have either a guanine or a cytosine at that position). The upper limit of expected bias based on GC3 alone is represented by the black curve; most T. thermophila ORFs cluster below the curve [red dots as in (B)].

(B) Principal component analysis of relative synonymous codon usage in T. thermophila. The 232 genes in the tail of the comma-shaped distribution (those with the most biased codon usage) are colored red.

(C) Principal component analysis of relative synonymous codon usage in P. falciparum.

https://doi.org/10.1371/journal.pbio.0040286.g005

The use of preferred codons by a gene is thought to allow for more efficient or accurate translation [65]. This appears to be the case here as, of the predicted genes using the preferred subset, many have likely housekeeping functions, and, although they account for only 0.85% of all predicted genes, 12.5% of all ESTs map to them (Table S7). Although some do not have EST matches and theoretically could represent falsely predicted genes, it seems unlikely that spurious genes would use the preferred codon set. Thus we predict that these outlier genes are either highly expressed (in at least some of the conditions normally encountered by the organism) or have some critical function requiring accurate translation.

Codon usage differences between genes are thought to have only small fitness effects. For natural selection to effectively work on codon usage differences and to thus create a preferred subset, factors that enhance genetic drift (e.g., small population sizes, inbreeding) must be weaker than the selective forces [66]. Thus although codon usage is probably under selective pressure in all species, not all are able to evolve preferred codon sets. For example, although it has a similar AT bias to T. thermophila, no preferred set could be detected in the apicomplexan Plasmodium falciparum (Figure 5C), possibly a reflection of its parasitic lifestyle and limited effective population size. The presence of a preferred subset in T. thermophila is likely a reflection of a large effective population size due to its free-living, sexually reproducing lifestyle (see [66,67] for additional discussion on the large population size of this species).

No plastid-derived genes can be identified.

One question of particular interest that the T. thermophila genome might shed light on relates to the timing of the origin of the plastids found in apicomplexans and dinoflagellates, the other members of the alveolates [68,69]. Although the plastids in these lineages differ (e.g., that in apicomplexans, known as an apicoplast, is not even involved in photosynthesis), both are thought to be of red algal origin [70]. This has led to the proposal that the plastids in these lineages are the result of a single endosymbiotic event between an ancestor of apicomplexans and dinoflagellates and a red alga, with the algal nucleus being lost and the algal plastid being kept. A key question is whether this secondary endosymbiosis occurred before or after the ciliates split off from the other two lineages. The possibility that it occurred before the ciliate split is known as the chromalveolate hypothesis [71].

For the chromalveolate hypothesis to be correct, plastid loss would have to have occurred in ciliates, most likely at the base of the ciliate tree since no modern ciliates are known to harbor plastids. If the ancestor of ciliates once had a plastid, it is possible that some plastid-derived genes would have been transferred to the nuclear genome (as has occurred in many lineages including apicomplexans and dinoflagellates [72]), and furthermore that some such genes would still be found in T. thermophila. To test this possibility, we built phylogenetic trees of all genes in the genome and searched for those with a branching pattern consistent with plastid descent (see Materials and Methods). For T. thermophila, we do not see any signal for genes of plastid descent that rises above the noise seen in such automated phylogenetic analyses.

Several lines of evidence suggest that this is not a general flaw in the phylogenetic approach used here. For example, we have used the same approach to identify and catalog the plastid-derived genes in other lineages including the plant Arabidopsis thaliana and the apicomplexan P. falciparum. In addition, such an approach has been used to detect past endosymbioses in other eukaryotic lineages [73]. Finally, using the same approach we identified 91 likely mitochondrion-derived genes (Table S8) in the T. thermophila nuclear genome. This is significant because mitochondrion-derived genes are generally more difficult to identify than plastid-derived genes [74], in part because the plastid symbiosis was more recent [75].

Nevertheless, since it is possible that our phylogenomic screen might have missed some plastid-derived genes, we also did a targeted search for genes that might be expected to be retained, using the apicoplast as a model. Apicoplasts are involved in biosynthesis of fatty acids, isoprenoids, and heme. Fatty acid and isoprenoid biosynthetic pathways are of special interest because the plastid-derived pathways are distinct from analogous pathways in the eukaryotic cytoplasm [76]. In the case of isoprenoid biosynthesis, genes for proteins in the canonical eukaryotic cytosolic mevalonate pathway are present as expected based on experimental studies [77–79], but no enzymes involved in the plastid-derived DOXP pathway were evident. For fatty acid biosynthesis, while T. thermophila does not require an exogenous supply of fatty acids for growth, no evidence for a complete version of a type I (normally cytosolic) pathway could be found. Although at least some genes for a type II pathway are present, these are insufficient for de novo fatty acid synthesis and appear more likely to be derived from the mitochondrion than a plastid.

Based on the general and targeted searches, we conclude that there is presently no evidence for a plastid or ancestrally plastid-derived genes in T. thermophila. This does not preclude the possibility that other ciliates have plastid-derived enzymes or even a plastid, but there is presently no evidence to suggest this despite extensive ultrastructural observations [80,81]. If ciliates do lack all evidence of a plastid, it could either mean that the hypothesized early origin of the chromalveolate plastid is incorrect or that an ancestor of T. thermophila (and perhaps all ciliates) lost its plastid and all detectable plastid-derived genes outright. The latter possibility is not without precedent, as some apicomplexans such as the Cryptosporidia have lost their apicoplasts and have few, if any, plastid-derived genes in their nuclear genomes [82,83]. This loss has been suggested to be the result of metabolic streamlining in response to its parasitic lifestyle. Resolving whether a plastid was present in the ancestor of ciliates will be important to our understanding of the evolution of plastids and their biochemical relationship with eukaryotic hosts.

IES excision targets foreign DNA rather than repetitive DNA per se.

As discussed in the Introduction, there are multiple parallels between the IES excision process and other repeat element silencing phenomena such as RIP and heterochromatin formation. Despite these parallels, the processes differ significantly in their mechanisms of action and therefore likely have different short- and long-term evolutionary consequences. For example, in species with RIP, all repetitive DNA becomes a target for mutational inactivation, which has resulted in a drastic suppression of evolutionary diversification through gene duplication [84,85]. The IES excision process results in the exclusion of certain MIC DNA sequences from the transcriptionally active MAC. Experimental introduction of foreign transgenes into the MIC has shown that as MIC copy number increases, so does the efficiency of transgene excision [86]. One might therefore predict a similar suppression of gene duplication as in RIP. However, rather than targeting repetitive DNA per se, it has been proposed that IES excision specifically targets foreign DNA that has invaded the germline MIC but is not represented in the MAC [35,87,88]. MIC gene duplication and functional diversification should still be possible under this scenario as long as, at each conjugation event, the gene copies have not diverged in sequence enough to be recognized as foreign and excluded from the MAC; since sex is frequent in natural populations of T. thermophila [89], this should be the case. We therefore sought to use the genome sequence data to both test the foreign DNA hypothesis and to examine what the consequences of the IES excision process have been on the evolution of the T. thermophila genome.

Analysis of the genome reveals several lines of evidence that provide strong support for the foreign DNA hypothesis. First, small but nevertheless significant amounts of repetitive DNA are present in the MAC. This is best seen in analysis of the scaffolds that correspond to complete MAC chromosomes which are unlikely to contain MIC IES contamination. These scaffolds contain dispersed repeats that make up 2.3% of the total DNA. This means that some repetitive DNA bypasses the IES excision process. The second line of evidence comes from examining the small contigs and singletons (nonassembled sequences) in the assembly data. Known MIC-specific elements such as the REP and Tlr1 transposons [90,91] are found only in these small contigs, which are thus clearly enriched for MIC-specific DNA (and also for repetitive DNA; see Figure S5). In fact, the small contigs contain homologs of an unusually wide range of transposable element (TE) clades for a single-celled eukaryote [92,93] including many previously unreported in Tetrahymena (Table S9). We do not find any good matches to TEs in any of the large contigs. Thus, transposons in general appear to be filtered out very efficiently by the IES excision process. The tandem and dispersed repeats in the MAC appear to correspond to noninvasive DNA (e.g., the 5S rRNA genes). Taken together, the fact that mobile (and likely invasive) DNA elements are kept out of the MAC, combined with the fact that both tandem and dispersed noninvasive repeats avoid the excision process, indicates strong support for the foreign DNA hypothesis.

In organisms with RIP, since all duplicated DNA is targeted [94], gene diversification by duplication is suppressed. For example, the fraction of all Neurospora crassa genes found in paralogous families is only 19%, a value that falls below the overall correlation line between this fraction and total gene number [84]. In addition, very few gene pairs share greater than 80% amino acid sequence identity [84]. Consistent with the foreign DNA hypothesis, we do not see such signs of suppression of gene family diversification in T. thermophila. Large numbers of paralogous genes are found in the genome (1,970 gene families including 10,851 predicted proteins) (Table 3). The fraction of genes in such families in T. thermophila (39%) is much higher than that seen in N. crassa. Although this fraction is not as high as would be predicted from the observed correlation between total number of genes and the fraction found in paralogous families [84], the fraction of gene pairs sharing greater than 80% amino acid identity is much higher than in N. crassa and similar to that found in other sequenced eukaryotes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3.

Gene Families

https://doi.org/10.1371/journal.pbio.0040286.t003

Since it is possible some of the 1,970 gene families could have originated by duplications that occurred prior to the origin of the IES excision process, it is more useful to examine recent duplications. We searched for such duplications in multiple ways, including the identification of genes duplicated in the T. thermophila lineage relative to other lineages for which genomes are available (Table S10) and by searching for pairs of paralogs with very similar sequences. Both of these classes are abundant in T. thermophila, further indicating that the IES excision does not significantly affect expansion of gene families of “native” genes. Thus the ciliate system of targeting invading DNA has significantly different consequences than RIP.

High gene count in T. thermophila.

The expansion of gene families helps explain the high gene count in T. thermophila, which is higher than that of other protists and even surpasses that of some metazoans (Table 4). The duplication events appear to be spread out over evolutionary time with some being ancient and some quite recent. We searched for but did not find evidence for either whole genome or segmental duplications. We do find extensive numbers of tandemly duplicated genes. In total, 1,603 tandem clusters of between two and 15 genes were found, comprising 4,276 total genes; 67% of these clusters are simple gene pairs and 96% contain five or fewer genes. Thus it appears many of the paralogous genes in T. thermophila are the results of separate small duplication events.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4.

Numbers of Protein-Coding Genes in Various Eukaryotes

https://doi.org/10.1371/journal.pbio.0040286.t004

The high gene count in T. thermophila relative to some other single-celled eukaryotes is not simply a reflection of gene family expansions. For example, when recent gene expansions are collapsed into ortholog sets, we find that humans and T. thermophila share more orthologs with each other (2,280) than are shared between humans and the yeast S. cerevisiae (2,097) or T. thermophila and P. falciparum (1,325) (Figure 6), despite the sister phyla relationships of animals and fungi on the one hand and ciliates and apicomplexans on the other. We note that this does not mean that humans and T. thermophila are overall more similar to each other than either is to species in sister phyla. For example, humans and S. cerevisiae do share some processes that evolved in the common ancestor of fungi and animals. In addition, for orthologs found in all eukaryotes, the human and S. cerevisiae genes are more similar in sequence to each other than either is to genes from T. thermophila. The higher number of orthologs shared between humans and T. thermophila is a reflection of both the loss of genes in other eukaryotic lineages and the retention of a variety of ancestral eukaryotic functions by T. thermophila. Consistent with this conclusion, there are 874 human genes with orthologs in T. thermophila but not S. cerevisiae, 58 of which correspond to loci associated with human diseases (Table S12). Thus genome analysis reveals many cases where T. thermophila can continue to complement experimental studies of yeast as a model system for eukaryotic (and human) cell biology [13].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Orthologs Shared among T. thermophila and Selected Eukaryotic Genomes

Venn diagram showing orthologs shared among human, the yeast S. cerevisiae, the apicomplexan P. falciparum, and T. thermophila. Lineage-specific gene duplications in each of the organisms were identified and treated as one single gene (or super-ortholog). Pairwise mutual best-hits by BLASTP were then identified as putative orthologs.

https://doi.org/10.1371/journal.pbio.0040286.g006

Signal transduction and the expansions of kinase families.

A variety of genes with putative roles in signal transduction were identified in our screens of paralogous genes. Of these, we chose to perform an in depth analysis of the kinases because they are such a diverse family of proteins and because they have been found to have critical roles in sensory and regulatory processes across the tree of life. In total, 1,069 predicted protein kinases (Tables 5 and S11A) were identified in the genome. This corresponds to approximately 3.8% of the predicted proteome, a fraction significantly larger than the approximately 2.3% in fungi, Drosophila, and vertebrates [104]. Among these, representatives were found of 54 of the known kinase families and subfamilies [105]. Some families found in a wide diversity of eukaryotes [106] were not detected. This includes the checkpoint kinase CHK1/RAD53, the PI3 kinase–related kinase TRRAP, two cyclin-dependent kinases (CDK7 and CDK8, which may be functionally replaced by the related expanded CDC2 family), and two poorly conserved classes (Bub1 and Haspin) that may have been missed by sequence homology searches. Despite the reported presence of phosphotyrosine in T. thermophila [107], no clear members of the tyrosine kinase group could be identified. However, the genome encodes some proteins that might be alternative tyrosine kinases including multiple dual-specificity kinases (e.g., Wee1, Ste7, TTK, and Dyrk) as well as five members of the related TKL group, which may mediate tyrosine phosphorylation in the slime mold Dictyostelium discoideum [106]. Twelve kinase classes are found in T. thermophila and humans but not yeast, and thus are apparent examples of the retention of ancestral eukaryotic functions discussed above. Several of the genes in these classes have been implicated in the etiology of human disease (Dyrk1A, DNAPK, SGK1, RSK2, Wnk1, and Wnk4) [108].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5.

Distribution of Selected Protein Kinase Classes in T. thermophila and Other Classified Kinomes

https://doi.org/10.1371/journal.pbio.0040286.t005

A key feature of the T. thermophila kinome is the expansion of several kinase classes relative to other sequenced organisms (Table 5). The implications of some of these expansions can be predicted based on the known functions of family members. For example, the mitotic kinase families Aurora, CDC2, and PLK are all substantially expanded, perhaps reflecting the additional signaling complexity required by two nuclei that simultaneously engage in very different processes within the same cell cytoplasm. Also expanded are multiple kinases that interact with the microtubule network [109,110] [e.g., Nima-related kinases (NRKs) and the ULK family], possibly reflecting diversification of cytoskeletal systems (discussed more below). Of the kinase families with known functions, the most striking expansion is the presence of 83 histidine protein kinases (HPKs), which are generally involved in transducing signals from the external environment [111]. HPKs are found predominantly in two-component regulatory systems of bacteria, archaea, protists, and plants and are absent from metazoans. Most of the T. thermophila HPKs have substrate receiver domains, and many are predicted to be transmembrane receptors.

The full meaning of the kinome diversity in T. thermophila is hard to predict as a great deal of the diversification has occurred in classes for which the functions are poorly understood. For example, in many of the known kinase families, the T. thermophila proteins are highly diverse in sequence, both relative to those in other species as well as to each other (e.g., see Figure S6). The scope of the diversification in T. thermophila is perhaps best seen in the fact that 630 (approximately 60%) of the kinases could not be assigned to any known family or subfamily [105]. Overall, 37 novel classes of kinases and hundreds of unique proteins were identified in this genome. The presence of so many novel kinases and expansions in many known classes of kinases is both an indication of the versatility of the eukaryotic protein kinase domain seen in other lineages [112] and suggestive of a great elaboration of ciliate-specific functions.

Diversification of membrane transport systems.

Many of the most greatly expanded T. thermophila gene families encode proteins predicted to be involved in membrane transport. Membrane transporters play critical roles in responding to variations in the environment and making use of available resources. We therefore conducted a more thorough analysis of the predicted transporters in this species. Overall, T. thermophila possesses a robust and diverse collection of predicted membrane transport systems (Tables 6 and S11B). Comparison to other eukaryotes [113] reveals some interesting differences in terms of both classes of transporters and predicted substrates being moved. For example, T. thermophila has more representatives in each of the four major families than do humans. In addition, it encodes a much higher number of transporters in the ABC superfamily, voltage-gated ion channels (VICs), and P-type ATPases than any other sequenced eukaryotic species (Table 6) including the other free-living protists, the diatom Thalassiosira pseudonana, and the slime mold D. discoideum. Regarding substrates, an extremely extensive set of transporters likely specific for inorganic cations has been identified (Table 6). Most of these are channel-type transporters and cation-transporting P-type ATPases. Interestingly, despite the apparent massive amplification of cation transporters, T. thermophila has a very limited repertoire of transporters for inorganic anions: only one member each for sulfate, phosphate, arsenite, and chromate ion were identified, and there are no predicted anion channels. The reason for the difference in the amplification of cation versus anion transporters is unclear.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6.

Comparison of the Numbers of Membrane Transporters in T. thermophila and Other Eukaryotes by Family and Predicted Substrate

https://doi.org/10.1371/journal.pbio.0040286.t006

As with kinases, some of the most interesting properties are revealed by examination of the lineage-specific duplications of transporters. The recent clusters include K⁺ channel proteins (285 members), ABC transporters (152 members), cation-transporting ATPases (59 members), K⁺ channel beta subunit proteins (22 members), oxalate:formate antiporters (24 members), sugar transporters (22 members), and phospholipid-transporting ATPases (20 members). The expansion of the K⁺ channel proteins, which are VIC-type transporters, was particularly large and was pursued further.

In total, 308 VIC-type K⁺-selective channels have been predicted, many more than in any other sequenced species and over three times as many as identified in humans (89). A multigene family of potassium ion channels has also been identified in P. tetraurelia [114] and thus may be a general characteristic of some ciliates. Some lines of evidence suggest that this expansion in ciliates could be adaptive. First, K⁺ channels control the passive permeation of K⁺ across the membrane, which is essential for ciliary motility [115]. Second, a novel adenylyl cyclase with a putative N-terminal K⁺ ion channel regulates the formation of the universal second messenger cAMP in ciliates and apicomplexans [116,117], which could assist in responding to sudden changes of the ionic environment. T. thermophila encodes six homologs of this adenylate cyclase/K⁺ transporter, whereas the parasitic apicomplexans P. falciparum and Cryptosporidium parvum encode only one each.

The robust transporter systems present are likely a reflection of T. thermophila's behavioral and physiological versatility as a free-living single-celled organism and its exposure to a wide range of different substrates in its natural environment. Examination of the specific types of expansions suggests that functions associated with transport of K⁺ and other cations have been greatly diversified. Thus such functions may play a role in many of the unique aspects of the biology of this species and ciliates in general.

Proteolytic processing.

T. thermophila is a voracious predator and thus might be expected to have a wide diversity of proteolytic enzymes. Analysis of the predicted proteins in T. thermophila reveals some conflicting results relating to this idea. On the one hand, many of the largest clusters of lineage-specific duplications are of proteases (e.g., papain, leishmanolysin). On the other hand, the total number of proteases identified (480) is relatively low in terms of the fraction of the proteome (1.7%) compared to other model organisms that have been sequenced and annotated [118–120]. The conflict is most likely a reflection of the diversity of physiological processes in which proteases function [121]. Thus we examined the subclassification of types of proteases present in more detail.

Using the Merops protease nomenclature, which is based on intrinsic evolutionary and structural relationships [119] the T. thermophila proteases were divided into five catalytic classes and 40 families. These are: 43 aspartic proteases belonging to two families, 211 cysteine proteases belonging to 11 families, 139 metalloproteases belonging to 14 families, 73 serine proteases belonging to 12 families, and 14 threonine proteases belonging to the T1 family (Tables 7 and S11C). Some unique features of T. thermophila can be seen by comparison to P. falciparum which is the most closely related sequenced species to have a detailed analysis of its proteases published [122]. Twenty-one protease families are present in both genomes. For example, the highly conserved threonine proteases and the ubiquitin carboxyl-terminal hydrolase families (C12 and C19) reflect the crucial role of the ATP-dependent ubiquitin-proteasome system, which has been implicated in cell-cycle control and stress response [123]. Nineteen protease families are present in T. thermophila but not P. falciparum. One of these includes leishmanolysin (M8), originally identified in the kinetoplastid parasite Leishmania major and thought to be involved in processing surface proteins [124–126]. This family is greatly expanded (to 48 members, including 15 in a tandem array) in T. thermophila and suggests that surface protein processes may be important here, although the functions of leishmanolysin-related proteases in nonkinetoplastid eukaryotes remain unclear. The carboxypeptidase A (M14) and carboxypeptidase Y (S10) families are expanded to 28 and 25 members, respectively, in T. thermophila, which may reflect numerous and diverse functions. Only four protease families present in P. falciparum are not found in T. thermophila. Among these are metacaspase (C14), an ancestral type of caspase that is characteristic of apoptosis or apoptosis-like signal transduction pathways [127].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7.

Protease Complements in T. thermophila and Other Model Organisms

https://doi.org/10.1371/journal.pbio.0040286.t007

The largest clusters of expanded proteases in T. thermophila are all cysteine proteases, which comprise 44% of the total protease complement. The two most prominent families from this class are the papain family (C1), which is the most abundant and complex family, with 114 members, and the ubiquitin carboxyl-terminal hydrolase 2 family (UCH2, C19) with 47 members. It is possible that the biochemical activity among the paralogs within these families is conserved but that they are used in different parts of the cell (or outside the cell) or in different developmental stages in T. thermophila.

Cytoskeletal components and regulators.

Ciliates have highly complex cytoskeletal architecture [128] with highly polarized cell types which assemble 18 types of microtubular organelles in specific locations along the anteroposterior and dorsoventral axis. We therefore sought to determine whether this diversity was reflected in the genome. As with the protease analysis described above, initial comparisons of the number of particular types of cytoskeletal and microtubule-associated proteins was somewhat ambiguous (the numbers for humans and T. thermophila are shown in Tables 8 and S11D). For example, although kinesin and dynein motors as well as kinases associated with microtubules appear to be expanded, structural components of the cilia and participants in the intraflagellar transport pathway are not. In addition, some cytoskeletal protein types are apparently absent from T. thermophila; these include intermediate filament proteins (including nuclear lamins) as already suggested by biochemical studies [129], some microtubule-associated proteins (MAP2, MAP4, and Tau, for which no nonanimal eukaryotic homologs have been found) and some actin-binding proteins (e.g., α-actinin). To better understand what role genes involved in microtubule and cytoskeletal functions might have played in the diversification of this species, we focused analysis on some of the genes with apparent expansions: tubulins, dyneins, and regulatory proteins.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8.

Numbers of Loci Encoding Selected Types of Cytoskeletal Genes in T. thermophila and H. sapiens

https://doi.org/10.1371/journal.pbio.0040286.t008

Tubulins.

Tubulins are the key structural components of microtubules and they come in many forms in eukaryotes [130]. In the T. thermophila genome, phylogenetic analysis of tubulin homologs (Figure 7) reveals the presence of one or two genes, each within the essential alpha (α), beta (β), and gamma (γ) subfamilies (as reported previously [131–133]) and one in each of the delta (δ), epsilon (ε), and eta (η), which are found in organisms that possess centrioles/basal bodies [134–136]. In addition, T. thermophila encodes noncanonical tubulin homologs that can be divided into two categories. In the first category are genes that are most similar to the canonical α- or β-tubulins. These nine genes (three α-like and six β-like) lack characteristic motifs for the tail domain post-translational modifications (polyglutamylation and polyglycylation) that are essential to the function of their canonical counterparts [137–139]. Three of the β-like genes (BLT1/TTHERM_01104960, TTHERM_01104970, and TTHERM_01104980) form a tandem cluster with intergenic intervals of less than 2 kb. We hypothesize that these genes function, perhaps redundantly, in formation or function of some of the many highly specialized microtubule systems of T. thermophila cells. Experimental analysis of BLT1, a β-like tubulin, indicated that its product localizes to a small subset of microtubules and is not incorporated into growing ciliary axonemes (K. Clark and M. Gorovsky, unpublished data). Genetic deletion of this gene or of the α-like gene TTHERM_00647130 did not yield an obvious phenotype (R. Xie and M. A. Gorovsky, unpublished data).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. Tubulin Gene Diversity in T. thermophila

The figure shows a neighbor-joining tree built from a clustalX alignment. Species abbreviations: Hs, H. sapiens; Dm, D. melaogaster; Sc, S. cerevisiae; Tt, T. thermophila; Pt, P. tetraurelia; Cr, C. reinhardtii; Tb, T. brucei; Ec, E. coli; Xl, Xenopus laevis. A prokaryotic tubulin ortholog, Escherichia coli FtsZ, was used as the outgroup.

https://doi.org/10.1371/journal.pbio.0040286.g007

The second category of noncanonical tubulin homologs consists of three novel proteins (TTHERM_00550910, TTHERM_01001250, and TTHERM_01001260) that fall into a clade with P. tetraurelia iota tubulin. Two of these (TTHERM_01001250 and TTHERM_01001260) are closely related to each other (Figure 7) and closely linked in the genome and thus likely arose by a recent tandem duplication. The functions of these genes are unknown, but because they are, so far, unique to ciliates, they might be responsible for microtubule functions specific to this phylum.

Dyneins.

Dyneins, which were first discovered in Tetrahymena [140], are molecular motors that translocate along microtubule tracks, a process critical to many activities in T. thermophila including ciliary beating, karyokinesis, MAC division, cortical organization, and phagocytosis. Many of these activities are critical for sensing and responding to changes in the environment. Each dynein complex consists of one, two, or three heavy chains (containing the motor activity) and specific combinations of smaller subunits, including intermediate, light-intermediate, and light chains, which regulate motor activity and the tethering of dynein to its molecular cargo [141–143]. In organisms with cilia or flagella, there are multiple isoforms of dyneins, including the axonemal outer arm dyneins, the axonemal inner arm dyneins, and nonaxonemal or “cytoplasmic” dyneins. Each is specialized in its intracellular location and the cellular task it performs [144].

In total we identified 21 light chains, five intermediate chains, two light-intermediate chains, and 25 heavy chains (Table S13). The expression of each gene, as well as the exon/intron structures of most, was confirmed by RT-PCR and, if necessary, sequencing of the RT-PCR product. For the most part, the families of T. thermophila dynein subunits appear to be similar to those of other model organisms; however, there are some interesting differences. T. thermophila light chains LC3A and 3B are most similar to the green alga Chlamydomonas reinhardtii's LC3 and LC5 [145]. These proteins belong to the larger family of thioredoxin-related proteins, and, without biochemical evidence identifying one or both of the proteins as part of a dynein complex, it may be premature to label these as dynein components. Light chain LC4 belongs to the calmodulin-related family of proteins and may regulate calcium-dependent ciliary reversal. T. thermophila expresses two LC4 genes, perhaps providing alternative or additional ways to control ciliary motility compared to species that express only one. In other systems, LC8 is associated with several different dynein and nondynein complexes, and T. thermophila expresses one canonical LC8 as well as five divergent LC8-like genes, with unknown functions.

Perhaps the most interesting revelation is that T. thermophila expresses 25 dynein heavy chains. These include the 14 DYH genes previously described [146,147] and 11 new ones, all of which appear to be axonemal. The complexity of the DYH family may represent a mechanism by which the organism can fine-tune ciliary activity, produce specialized cilia (e.g., oral and posterior cilia), and/or generate large numbers of new cilia quickly. Along these lines, there has also been an expansion in other motor proteins. For example, there are 78 kinesins, more than in any other sequenced organism ([101] and Table 8). In addition, although there are fewer myosins than in humans (13 versus 22), 12 of 13 of the T. thermophila genes comprise a single novel myosin class not found in other organisms [102,148].

Regulation of microtubules and microtubule-associated processes.

Among the expanded genes in T. thermophila are a variety implicated in the regulation of microtubules or microtubule-associated processes. One example is the tubulin tyrosine ligase-like domain proteins of which multiple members have been identified as enzymes responsible for polyglutamylation of either α- or β-tubulin [149]. T. thermophila encodes 50 tubulin tyrosine ligase-like proteins compared with 14 in human. Another example is the NRK family of protein kinases which, as mentioned above, has undergone a large expansion in T. thermophila. NRKs are often found associated with microtubular organelles [150] such as centrioles, basal bodies, and flagella and play multiple roles, including the regulation of centrosome maturation [151] and flagellar excision [152]. We identified 39 NRKs in T. thermophila, roughly three times the number of such loci in humans. Phylogenetic and functional analyses have suggested that this diversification has adapted the members of this family for distinct subcellular localizations and cytoskeletal roles [103]. Thus, such gene expansions could allow differentially targeted protein isoforms to regulate the function of the same organelle type in different locations or generate different properties of the same structural building materials (e.g., microtubules), which are used as frameworks to build different types of organelles.

Secretory pathways and membrane trafficking.

Besides the conventional organelles, T. thermophila maintains several more specialized membrane-bound compartments, including alveoli (shared with other alveolates), a contractile vacuole (found in many protists), and separate, functionally distinct macronuclei and micronuclei [128]. It also has multiple pathways for plasma membrane internalization, as well as both constitutive and regulated exocytosis [128,153]. The sorting and trafficking of membrane components are critical functions for all these activities. Analysis of the genome reveals homologs of many of the key proteins known from other eukaryotes to be involved in vesicle formation and fusion, including all major classes of coat proteins (Table S14). One interesting finding that came from genome analysis is that T. thermophila encodes eight dynamin-related proteins, more than most other sequenced unicellular eukaryotes, and two of them, Drp1p and Drp2p, have evolved a new function in endocytosis [96] (A. Rahaman and A. P. Turkewitz, unpublished data). Furthermore, phylogenetic analysis indicated that the recruitment of dynamin to a role in endocytosis occurred independently by convergent evolution in the animal and ciliate lineages [96].

The diversification of membrane trafficking is more apparent in regard to Rab proteins, which are small monomeric GTPases that regulate membrane fusion and fission events. T. thermophila, with 69 Rabs (Table S15), has a number more along the lines of humans (which have 60) than many single-celled species, such as Saccharomyces cerevisiae, which has 11 [154] and Trypanosoma brucei, which has 16 [155]. Based on localization and functional studies, including comparisons between yeast and humans [156], Rabs have been divided into eight groups [157]. Phylogenetic analysis (Figure S7) indicates that T. thermophila encodes representatives of all but groups IV and VII, which are involved in late endocytosis and Golgi transport, respectively. For group VII this appears to reflect a lineage-specific loss, since the genomes of both T. brucei and Entamoeba histolytica have several homologs in this group. Two T. thermophila Rabs appear homologous to Rab28 and Rab32, which have not been assigned to any of these groups; Rab32 was previously thought to be restricted to mammalian lineages. Rab groups II and V, involved in endocytosis, are especially large in T. thermophila and include several Rab2, Rab4, and Rab11 homologs in group II. This may reflect the intricacy of maintaining at least two major pathways of membrane internalization. Additionally, 29 Rabs in T. thermophila fail to cluster with any of the Rab groups found more widely among eukaryotes. Within this group, 20 cluster into three clades, designated Tetrahymena clades I, II, and III in Figure S7, which may represent ciliate-specific radiations. The remaining nine are very divergent and may represent very ancient duplication events and/or changes related to recruitment for novel function. Because unambiguous alignment among such divergent Rabs is difficult, their relationships will become clearer as additional related genomes are sequenced.

Recently, large numbers of Rabs have been found in a variety of amoeboid protists including D. discoideum, E. histolytica [158], and the parabasalid Trichomonas vaginalis [159]. The diversification in these species was proposed to relate to their amoeboid lifestyle [159]. However, the presence of significant diversification in T. thermophila suggests that different protist lifestyles may be accompanied by their own brand of significant Rab diversification.