Our initial investigation focused on tRNAs with properties that place them at the extremes of archaeal tRNA characteristics. tRNA introns identified in sequenced archaeal genomes have sizes ranging from 11 to 129 nucleotides with a median of 15 nucleotides 2. Besides introns of tRNA^Trp(GUC) that encode an embedded C/D box small RNA (sRNA) 151617, only one tRNA gene, tRNA^Asp(GUC) of Aeropyrum pernix, contains an intron exceeding 100 nucleotides (Table S1 in Additional file 1). Using an archaeal-specific version of snoscan 18, and manual alignment to other predicted C/D box sRNAs in A. pernix and other archaeal species, we failed to find any trace of a C/D box sRNA that might explain the long tRNA intron. Promoter analyses revealed a strong signal in the middle of the tRNA^Asp(GUC) intron, including both the transcription factor B response element and TATA box, encoded on the same strand as the tRNA gene. This high-confidence promoter prediction scores better than 88% of all predicted transcripts in the genome (Figure S1 in Additional file 1), better than 63% of the predicted promoters for annotated tRNA genes in A. pernix, and is highly similar to the promoter upstream of the 5' end of the tRNA^Asp(GUC) gene (Table S2 in Additional file 1). The promoter's placement predicts a transcription start site 20 to 25 nucleotides upstream of the 3' end of the intron, consistent with production of a leader sequence of length and high G/C content that is similar to leaders found in the 3' halves of trans-spliced tRNA genes 58

With the aid of an improved version of tRNAscan-SE 1920, we found that tRNA^Asp(GUC) can be modeled as a combination of two separate transcripts joined between position 37 and 38, the canonical intron position (Figure 1a, Table 1). Similar to other split tRNAs, a canonical BHB (helix-bulge-helix-bulge-helix - hBHBh') secondary structure at the exon-splicing junction is observed, followed by a 14-bp G/C-rich stem. RT-PCR with transcript-specific primers and sequencing of PCR-derived clones verify expression of the mature tRNA and the two tRNA halves (Figure 2a). To further confirm that the two halves are separate transcripts, we conducted northern analysis with specific probes antisense to the two halves and the full-length 199-nucleotide precursor tRNA. Results show that the two pre-tRNA halves are present at their expected sizes, as is the predicted mature tRNA (Figure 2b). Interestingly, an amplified RT-PCR product corresponding to the size of a 199-nucleotide pre-tRNA^Asp(GUC) was observed, but its absence in the northern analysis suggests this is a very low abundance read-through transcript that may be spliced, but does not contribute significantly to the mature tRNA abundance. Thus, this tRNA may be a true evolutionary intermediate between the intron-spliced and trans-spliced forms.

tRNA^Asp(GUC) in A. pernix is the first example of a trans-spliced tRNA encoded by separate, but directly adjacent transcripts. Upon re-examination of all tRNAs in Archaea for a similar pattern, we found that the tRNA^Asp(GUC) ortholog in Thermosphaera aggregans, a crenarchaeon in the same phylogenetic order Desulfurococcales, also appears to be split with the two halves joined at the canonical intron position (Figure 1a, Table 1). Similar to the arrangement in A. pernix, the 5' half of the T. aggregans split tRNA is located adjacent to the 3' half. Unlike A. pernix, the two halves are encoded on opposite strands in a convergently transcribed orientation (Figure 3).

To get an evolutionary perspective on the lineage of this tRNA gene, we examined the syntenic region in Desulfurococcus kamchatkensis, the closest sequenced relative to T. aggregans. In both of these species, a sequence of three genes is present: eIF-2A, Nop10p, then tRNA^Asp(GUC) (Figure 3). However, the end of the syntenic region occurs in the middle of the tRNA^Asp(GUC), where apparently the ancestral, uninterrupted form of tRNA^Asp(GUC) observed in D. kamchatkensis has been split and inverted in T. aggregans, by an unknown series of genome re-arrangement events. Examination of the tRNA^Asp(GUC) genomic regions for syntenic blocks in the eight sequenced Desulfurococcales species showed that every species had genome rearrangements downstream of tRNA^Asp(GUC), relative to every other species, and three had rearrangements upstream as well. The split tRNA^Asp(GUC), along with many other tRNAs 2122, appear to be common positions for genome recombination and/or viral integration. Although A. pernix and T. aggregans are both Desulfurococcales, they are separated by four other species that are more closely related, but lack the split tRNA^Asp(GUC). Given that none of the other six Desulfurococcales species have a split tRNA^Asp(GUC), the most parsimonious explanation for these observations is that a tRNA-splitting event happened relatively recently in two independent instances among the Desulfurococcales (Figure 4). The agent causing the splitting event could be specific for these tRNA sequences, as there are only four nucleotide changes between the A. pernix and T. aggregans tRNA sequences. Notably, there are many more changes in length and sequence identity (20 nucleotide differences) in the complementary region required to join the split halves, thus disfavoring the possibility of recent lateral transfer of a precursor split tRNA gene. As such, these are likely to be the first compelling examples of 'late' acquisition of split tRNA genes based on proximal tRNA halves. Late acquisition has also been suggested recently based on comparative analyses of non-canonical tRNA introns 23.

Using an improved version of the tRNAscan-SE computer program 1920 (manuscript in preparation), we predicted 46 tRNAs in Staphylothermus marinus, including a missing tRNA^Lys(CUU) gene and an extra, low-scoring tRNA^Leu(CAA) prediction (26.23 bits versus all other archaeal tRNAs score >30.0 bits). This suggested a potential tRNA isotype mis-assignment. Indeed, closer examination of the low-scoring tRNA^Leu(CAA) revealed an almost exact match to tRNA^Lys(UUU) from position 30 to the 3' terminus, with no similarity upstream of position 30. The single difference between these two sequences is a U to C change, aligned to anticodon position 34 in tRNA^Lys(UUU), consistent with the fragment being the 3' end of the missing tRNA^Lys(CUU). Additional sequence similarity searches identified a candidate 5' half fragment of the tRNA^Lys(CUU) encoded 4,500 nucleotides away (Table 1; Figure S2a in Additional file 1), ruling out the possibility of an extremely long polycistronic intron. Strong promoters matching the promoter consensus (better than 40% of other identified tRNA promoters) were found at the expected distance upstream of both candidate loci.

Intriguingly, the two halves of the candidate split tRNA^Lys(CUU) in S. marinus join between position 30 and 31, precisely the same position as in the split tRNA^Lys(CUU) of distantly related N. equitans (Table 1) 5. A canonical BHB (hBHBh') secondary structure at the exon-splicing junction is also observed, just as in N. equitans tRNA^Lys(CUU), followed by a perfect 13-nucleotide trans-pairing between the regions downstream of the 5' half and the upstream of the 3' half (Figure 1b). Fortuitously, the genome of another species in the genus Staphylothermus, S. hellenicus, was recently sequenced, enabling identification of the orthologous split tRNA^Lys(CUU), closely resembling the counterpart in S. marinus (Figure 1b, Table 1). While the sequences of 3' halves in the two species are identical, there is one nucleotide difference near the end of the 5' tRNA half. The difference conserves a base pair in the trans-paired region, changing the predicted G-U pairing in S. marinus to a G-C pair in S. hellenicus (Figure 1b), further suggesting selective pressure to maintain a perfect 13-nucleotide trans interaction.

These two closely related Staphylothermus species give the first case in which the issue of genome stability and split tRNAs may be examined. The neighboring genes located upstream of the 5' half (Smar_1316-Smar_1321, Shell_1133-Shell_1128), in between the 5' and 3' halves (Smar_1322 - Smar_1326, Shell_1127 - Shell_1123), and downstream of the 3' half (Smar_1327 - Smar_1332, Shell_1122 - Shell_1116) retain complete synteny between species, showing a lack of recombination or integration events in this region (Figure S2 in Additional file 1). However, tRNA genes are known to be positions of genome rearrangement due to processes including transposon and viral integration 21242526. For context, we examined genome rearrangement events adjacent to the other 45 ortholog pairs of tRNAs, and found that 20 of them (44%) had a break in synteny either upstream, downstream, or on both sides of the tRNAs. Thus, the split tRNA^Lys(CUU) arrangement since the divergence of S. marinus and S. hellenicus has been preserved with no local recombination, and is consistent with hypotheses proposing enhanced genome stability from split tRNAs.

We computationally screened for other atypical tRNA transcripts by aligning tRNAs and their upstream promoter regions to identify unusual spacing between candidate promoters and predicted tRNA genes. When applied to the thermophilic crenarchaeon Thermofilum pendens, we found that the promoters of 44 mature tRNA genes, out of a total of 46 in the genome, are located in the upstream region between 30 and 49 nucleotides relative to the 5' end of the mature tRNAs (Figure S3 in Additional file 1), explained by natural variation in the lengths of 5' leaders of pre-tRNAs. The promoters of two outliers, tRNA^iMet(CAU) and tRNA^Tyr(GUA), were found at positions -72 and -65, respectively, implying 5' leaders at least 16 nucleotides longer than all others. Similar to the initial prediction of the split tRNA in S. marinus, tRNA^iMet(CAU) and tRNA^Tyr(GUA) in T. pendens scored relatively low (26.23 and 32.76 bits, respectively, compared to >60 bits for other tRNA genes in this genome). Secondary structure analysis showed that these tRNA predictions fail to form requisite tRNA cloverleaf secondary structure at the T-Ψ-C loop and the acceptor stem.

Results from an improved version of tRNAscan-SE 1920 showed that the original predictions of tRNA^iMet(CAU) and tRNA^Tyr(GUA) in T. pendens are only the 5' halves of these tRNA genes. For each tRNA, a precisely matching 3' half was found between the 5' half and its predicted promoter on the same strand, suggesting that the two fragments belong to the same transcript, using a single promoter (Figure S3 in Additional file 1). Primary and secondary structure analyses strongly support that these are circularly permuted tRNAs with a BHB motif at the exon-splicing junction, located between positions 59 and 60. Unexpectedly, the split position is precisely the same as the T-Ψ-C loop-permuted tRNAs identified in the distantly related eukaryotic alga C. merolae (Figure 1c) 9. The intervening sequences between the 3' and 5' halves of tRNA^iMet(CAU) and tRNA^Tyr(GUA) in T. pendens are 7 nucleotides and 1 nucleotide long, respectively, in comparison to intervening sequences in known algal permuted tRNAs ranging from 5 nucleotides to 85 nucleotides. Both of the genes also include a canonical intron, indicating that permuted tRNA structure and normal intron splicing are not exclusive processes. Unlike the permuted tRNAs in eukaryotes, these two archaeal tRNAs have genomically encoded 3'-terminal CCA sequences, and the tRNA sequences are quite different from those found in red alga, ruling out a recent inter-domain transfer between algal and archaeal species. However, the sequences and proteins flanking these tRNAs share strongest similarity to species outside of the Thermoproteales, suggesting these may have been part of laterally transferred regions acquired after the divergence from other sequenced Thermoproteales. Both tRNA^iMet(CAU) and tRNA^Tyr(GUA) are single-copy genes in T. pendens with essential decoding functions that cannot be supplanted by other tRNAs, indicating that processing of permuted genes is an essential activity in this species.

Complete genomic sequences of A. pernix, S. hellenicus, S. marinus, T. pendens, and T. aggregans were obtained from NCBI RefSeq (accession numbers [NC_000854], [NC_014205], [NC_009033], [NC_008698], and [NC_014160]).

Trans-spliced and permuted tRNAs were predicted using an improved version of tRNAscan-SE (manuscript in preparation) 1920. Archaeal tRNA-specific and BHB motif covariance models were created for similarity searching using Infernal 1.0 42 after pre-filtering possible candidates with tRNAscan 43 and an A/B box motif detection algorithm 44. A default cut-off score was set to 20 bits.

To generate a training set for promoter identification, potential operons were predicted genome-wide with the requirement of a minimum intergenic separation of at least 100 nucleotides (on the same strand). A 16-mer motif search of the 90 nucleotides upstream of known genes (not annotated as putative or hypothetical genes) using MEME 45 was conducted to identify the consensus promoter, including the transcription factor B response element (one to three adenosines) plus the TATA box. A position-specific scoring matrix was generated from the alignments of the MEME results after manual inspection. Each organism's position-specific scoring matrix was used to scan the 150-bp upstream region of all non-coding and protein-coding genes to identify potential promoter regions. Ten virtual genomes for each target genome were generated using a fifth-order Markov chain to retain the base frequency of the target genome, and scanned to identify the score distribution of false positives. The promoter candidates identified were filtered according to expected position 46 and a threshold P-value equivalent to that of the lowest-scoring known gene.

A. pernix K1 was obtained from Deutsche Sammlung von Mikroorganismen und Zellkulturen (DSMZ, Braunschweig, Germany). A. pernix cultures were grown in TY medium (0.4% tryptone, 0.2% yeast extract, pH 7) supplemented with 3.86 mM sodium thiosulfate 47. These cultures were incubated aerobically at 90°C and collected at mid- to late-log phase. Cell pellets were frozen in liquid N₂ and stored at -80°C.

Total RNA was extracted from the frozen cell pellets using a Polytron tissue homogenizer and TRI Reagent (Sigma-Aldrich, St. Louis, MO, USA). RNA samples were treated with TURBO DNase (Ambion, Austin, TX, USA) to remove any residual DNA, re-extracted with TRI Reagent, and normalized to 1.5 μg/μl.

Cell pellets were incubated overnight in SNET lysis buffer (400 mM NaCl, 1% SDS, 20 mM Tris-Cl, 5 mM EDTA) with 400 μg/ml proteinase K at 55°C. Genomic DNA was then extracted from cell lysate using phenol:chloroform:isoamyl alcohol. DNA samples were treated with RNaseA to remove any residual RNA and re-extracted with phenol:chloroform:isoamyl alcohol.

Total RNA from A. pernix was denatured at 100°C for 5 minutes and cooled on ice for 5 minutes. First strand cDNAs were synthesized from denatured total RNA using gene-specific reverse primers and Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA, USA) at 65°C for 30 minutes according to the manufacturer's instructions. These cDNA templates were PCR-amplified using forward and reverse primers spanning the mature tRNAs, the 5' tRNA halves, and the 3' tRNA halves. PCR parameters for 30 cycle amplification were as follows: denaturing at 94°C for 30 seconds, annealing at 68°C for 30 seconds, and extension at 72°C for 30 seconds, using AmpliTaq DNA polymerase (Applied Biosystems, Foster City, CA, USA). Negative controls for the reactions were conducted under the same conditions concurrently, but without the addition of reverse transcriptase at the first strand cDNA synthesis step. PCR products were cloned with the pCR-2.1-TOPO cloning kit (Invitrogen). Plasmid DNA was extracted using Zyppy plasmid miniprep kit (Zymo Research, Irvine, CA, USA). DNA samples were sequenced at the University of California Berkeley DNA Sequencing Facility. Primers used for RT-PCR are as follows: forward primer for mature tRNA^Asp(GUC) in A. pernix and its 5' half, 5'-CGCGGTAGTATAGCCTGGA-3'; forward primer for 3' half of tRNA^Asp(GUC) in A. pernix, 5'-CGGGCCTGCGGAGAG-3'; reverse primer for mature tRNA^Asp(GUC) in A. pernix and its 3' half, 5'-GCGGCCGGGATTTGAAC-3'; reverse primer for 5' half of tRNA^Asp(GUC) in A. pernix, 5'-GCGGGGCCCTTGACAG-3'.

Five micrograms of total RNA extracted from A. pernix cell cultures was denatured for 3 minutes at 90°C in an equal volume of 95% formamide gel loading buffer (Ambion), resolved on an 8% polyacryamide-urea gel by electrophoresis, and blotted onto Hybond N⁺ membrane (GE Healthcare, Piscataway, NJ, USA) by overnight electro-transfer. A 239-nucleotide sequence including the A. pernix tRNA^Asp(GUC) and the region between the two fragments was amplified by PCR using A. pernix genomic DNA and gene-specific primers Sn-Ape-tRNA-Asp (5'-CCCAGTGGTAAGATATGTGAACC-3') and Asn-Ape-tRNA-Asp (5'-GGCCGCGAGGATTATTG-3'). The resulting PCR product served as the template for generating single-stranded, [α-³²P]-ATP-labeled DNA probes by linear PCR using Asn-Ape-tRNA-Asp. Hybridizations were carried out at 42°C in UltraHyb buffer (Ambion). Hybridization patterns were determined using a PhosphorImager (Molecular Dynamics, Sunnyvale, CA, USA).