The mitochondrial genome of C. antarcticus is a closed circular molecule of 15,297 bp and contains the set of 37 genes usually found in metazoans 16 (Figure 1; GenBank accession number: EU016194). The gene order is identical to Drosophila yakuba 17, and reflects the presumed ancestral condition for Pancrustacea. The majority of genes are located on the plus or J-strand, the remainder having opposite polarity and being oriented on the minus or N-strand (Figure 1 & Table 1). There are four major non-coding regions of 622, 123, 60 and 33 bp in length that are located at the gene junctions rrnS/trnI, trnS(uga)/cob, nad6/cob and trnE/trnF, respectively. The largest non-coding region most likely corresponds to the region (A+T-rich region in insects) where in other hexapods the sequences responsible for the initiation of transcription and replication of the entire genome are found 1.

Canonical initiation codons (ATA or ATG), encoding the amino acid methionine, are used in 9 PCGs (atp8, cob, cox1, cox3, nad2–5 and nad4L), whereas four other genes start with non-standard codons (atp6, cox2, nad1 and nad6) (Table 1) as it often happens in animal mtDNAs 18. Only five PCGs terminate with the complete termination codon TAA (atp8, atp6, cox3, nad4L and nad6). In all other cases, stop codons are truncated (T or TA) and their functionality probably recovered after a post-transcriptional polyadenilation 19. These abbreviated stop codons are found in PCGs that are followed by a downstream tRNA gene, suggesting that the secondary structure information of the tRNA genes could be responsible for the correct cleavage of the polycistronic transcript. It is noteworthy that in all cases an additional complete stop codon is present a few bases downstream of the incomplete stop codon, as if a short overlap would be necessary to stop the RNA polymerase in case translation should begin before mRNA cleavage 320. tRNA genes are usually interspersed among protein coding genes, their secondary structure acting as a signal for the cleavage of the polycistronic primary transcript 1922. However, there are also direct junctions between two PCGs (atp8/atp6, atp6/cox3, nad6/cob and nad4L/nad4) where other cleavage signals, different from tRNA gene secondary structures, may be involved in the processing of the polycistronic primary transcript 23. In this respect, hairpin structures, frequently observed at the 3'-end of a PCG abutting the 5'-end of a neighbouring one, may serve as signal to direct immature mRNA cleavage 242526. Experimental analyses of cDNA pools have demonstrated that some mtDNA genes (i.e.: atp8/atp6 and nad4L/nad4) are recovered as bicistronic units in human HeLa cells 21 and in the dipteran Anopheles funestus 25. In the C. antarcticus mtDNA, with the single exception of cox3, a complete stop codon (TAA) is only present when two PCGs abutt directly (Table 1). The gene pair nad4L/nad4 is composed of a single in-frame coding unit (the two genes are separated by 6 nucleotides), whereas atp8 and atp6 overlap (7 nucleotides) as is almost universally found in metazoans 25.

All the 22 tRNA genes typically found in metazoan mtDNAs were identified according to their secondary structure and primary sequence of the corresponding anticodon (Figure 2). The only peculiar feature is represented by an additional (and apparently non-functional) copy of the tRNA gene for the amino acid leucine (trnL(uag)P), that is located within a non-coding spacer (123 bp) between trnS(uga) and nad1 (Figure 1). This tRNA pseudogene may represent the degenerating vestige of a duplication event that occurred in collembolan mtDNA early in the evolution of the group, given that in the same position a similar spacer is also present in G. hodgsoni 5 (50 bp) and Podura aquatica 8 (150 bp). Other unusual features are the lack of the DHU arm in trnS(gcu) and the reduction of the TΨC arm in trnG and trnV . Moreover, some tRNA genes have few mismatches in the acceptor and/or the discriminator arms (trnR, trnD, trnC, trnE, trnQ, trnH, trnL(uaa), trnLuag, trnK, trnS(gcu) and trnW). In similar cases (in metazoans but more commonly in plants and Protozoa) base pairing is restored post- or co-transcriptionally with an RNA-editing mechanism 2728. Overlaps between tRNA genes were not frequent (4 instances) and then limited to 1 or 2 bases (Table 1).

The putative secondary structures of the complete small and large ribosomal subunits (rrnS and rrnL) have been reconstructed for a limited number of insect species, and detailed studies are essentially available only for Drosophila virilis, D. melanogaster 29 and Apis mellifera 30. Within basal hexapod groups, the architectures of the ribosomal subunits have been tentatively reconstructed only for the two species Campodea fragilis and C. lubbocki 31, and a more detailed analysis has been performed only for domain III of the small RNA subunit of several Arthropleona species 32. Additional comparative analyses in basal hexapod lineages are required to allow an accurate comparison with current secondary structure models of holometabolan insects 30, and we therefore made no attempt to reconstruct the structure of the rrnS and rrnL rRNA here.

Previous analyses of the Drosophila A+T-rich region identified the signals responsible for the origin of replication of the major and minor strands (O_J and O_N, respectively) and confirmed that, at least in this genus, the mtDNA replicates with a strand-asynchronous, asymmetric mechanism 1. Mapping of O_J and O_N in Drosophila and other insect A+T-rich regions has recently allowed the definition of primary sequences and secondary structure elements involved in the initiation of the replication process 133343536. In holometabolan insects, two distinct thymine stretches, one located near trnI on the N-strand and another in the center of the A+T-rich region on the J-strand, appear to be responsible for O_J and O_N replication, respectively. Similarly, a thymine stretch plays an essential role in the initiation of the replication system of the Light strand (O_L) in mammals (although in this case the two O_R are separated by 2/3 of the genome), and is also believed to be involved in protein binding and primer RNA synthesis. Nevertheless, the conservation of thymine stretches among insects appears to be a peculiarity of more derived orders, with alternative sequences probably responsible for O_R in basal hexapod lineages 1.

The A+T-rich region of C. antarcticus is 622 bp; it shows six "TA" tandem repeats of variable length (from 5 to 7 units) and a low similarity in primary sequence with other collembolan specimens. The identification of sequences homologous to those observed in holometabolan insects involved in O_R was not completely successful in defining potential regulatory elements. In this respect, four thymine stretches of variable size (4 to 6) could be observed on the N-strand, within 100 bp of trnI, but none of them appears to correspond, in terms of size and position, to that observed in Drosophila. In addition, in agreement with previous reports 1 there is no thymine stretch longer than 6 bp on this strand. Conversely, in the C. antarcticus J-strand, a longer thymine stretch (14 bp) is present between nucleotides 14996 and 15009 and this appears to be assimilable (both in primary sequence and position) to the one observed in the central portion of the Drosophila A+T-rich region. Similarly to Drosophila, this portion of the C. antarcticus A+T-rich region may also act as the O_N, given that the motif "ACTATTT", frequently present in Drosophila (in 8/12 different species; see: Figure 6, in 1), is found 11 bp downstream of the 14-bp thymine stretch. If this evidence is confirmed by additional data on basal hexapod mtDNA sequences, the role of conserved primary sequence stretches in the initiation of the replication mechanism might be confirmed as a common feature of all hexapod groups (at least for the N-strand), from the most basal to the most derived lineages. The presence of hypothetical secondary structures involved in the initiation of the origin of replication process of the C. antarcticus mtDNA is difficult to establish. In this species the entire A+T-rich region can be folded in several paired structures, and additional comparative studies will be needed to clarify their role.

Unlike those of plants, animal mitochondrial genomes are very compact, with a high proportion of coding vs non-coding sequences 1637. Intergenic spacers are usually limited in number and size, and their occurrence is believed to be the result of errors of the mtDNA replication system (i.e. duplication of parts of the mitochondrial genomes, with redundant copies becoming pseudogenes and disappearing). Apart from the A+T-rich region, the largest (123 bp) non-coding sequence of the C. antarcticus mtDNA is located between trnS(uga) and nad1. Within this region, an additional copy of trnL(uag) (Figures. 1 and 2) can be identified. Given that a functional copy of this gene is present in its canonical position between nad1 and rrnL genes (as in other arthropod mtDNAs), and that there are several mismatches in the acceptor arm that might compromise the correct folding into a typical cloverleaf structure, we believe that trnL(uag)P may represent the molecular trace of an older duplication event that occurred during the evolution of this species (there are also alternative interpretations, see below). It is noteworthy that this genomic region is also subject to molecular rearrangments in other collembolan species (i.e. one tRNA translocation in Tetrodontophora bielanensis and Onychiurus orientalis) or shows intergenic spacers similar to those observed in C. antarcticus (i.e. 48 bp in G. hodgsoni). This observation suggests that the DNA fragment encompassing cob and nad1 may represent a "hot spot" for rearrangements in the collembolan mitochondrial genome, and that either several taxa have independently undergone rearrangements or these have occurred before the differentiation of the major collembolan lineages.

It is well known that tRNA genes are more prone to translocations than protein coding genes, with "hot spots" of rearrangement being described in some lineages 3839. In the case of the collembolan mtDNA, we speculate that a replication error (probably due to slipped strand mispairing) may have generated an intermediate state with at least a duplicated copy of the trnL(uag) gene. Afterwards, the superfluous copy(ies) may have been transformed into a pseudogene, whereas the original copy could have conserved its functionality (therefore leading to no changes in the gene order). Presently, it is only possible to conclude that: 1) similar (for size and position) intergenic spacers have been also found in two species of the genus Campodea (Diplura 31), another basal hexapod group probably as primitive as Collembola; 2) the gene junction trnS(uga)/nad1 corresponds to the shift of the coding polarity between two blocks of genes (nad6-cob-trnS(uga) oriented on the J-strand, and nad1-trnL(uag)-rrnL-trnV-rrnS on the N-strand). Interestingly, experiments conducted on transcriptional termination factors in D. melanogaster have identified the positions of some binding sites that are responsible for the control of multiple transcription units on the mtDNA of this species 40. One of the most interesting results of these studies is that, unlike human, and along with sea urchin mtDNA 41424344, in Drosophila one of these terminator sites is not located at the 3'-end of the ribosomal genes. Conversely, two stretches of non-coding sequence would correspond to the binding sites for the Drosophila mitochondrial transcription termination factor (DmTTF), a nuclear-encoded protein involved in the processes of attenuation and/or termination of mRNA transcripts 40. Intriguingly, the short stretches of non-coding sequences identified in Drosophila as DmTTF binding sites are located at the gene junctions trnE/trnF and trnS(uga)/nad1, in the same positions where C. antarcticus shows intergenic spacers (Table 1). Not only their position is identical, but also these sequences are located in a position of abrupt shift in the transcription polarity of the mtDNA genes (a suitable position for an attenuator/terminator of mRNA synthesis). Nevertheless, the non-coding regions of Drosophila and C. antarcticus, show no discernible sign of homology, leaving open the question of whether they do play the same role. It should be noted, however, the signalling role may be played by secondary structure motifs, like those observed between trnS(uga) and nad1 in Collembola and Diplura mtDNAs, rather than primary sequence.

Two perfect palindromic sequences were observed. One is 38 bp long and spans the junction between the nad2 and trnW (positions 1196–1233). The second is 18 bp long and lies 42 bp from the 3'end of rrnS (14033–14050).

Three repeated sequences were found in different areas of the genome. One is 22 bp long and encompasses the initial part of both trnL(uaa) and trnL(uag) (note that these latter genes are oriented on opposite strands), from base 3 to base 24 (with one mismatch). Interestingly, although duplication and remolding events of leucine tRNA genes have probably occurred independently several times within some major animal lineages 45, the observed 22 bp-long fragment is highly conserved in all collembolan species investigated (data not shown) suggesting that primary sequence and precise conformation of this region is important for the building of both tRNA genes and/or for their interaction with the ribosomal complex.

A second perfect direct repeated sequence, 18 bp long, is observed at 105 bp from the 3'-end of rrnS (14096–14113), and at 7 bp from the 3'-end of nad6 (10309–10327). A partially overlapping sequence of 17 bp is also repeated in inverted orientation in the middle of the A+T-rich region.

The mtDNA of many arthropod lineages is characterized by a strong compositional bias showing high values of A% and T% vs. G% and C%. This trend in nucleotide composition (A+T bias) is remarkable in some insect groups (i.e. holometabolan orders) and, to a lesser extent, is also apparent in the more basal hexapod lineages 31. The C. antarcticus mtDNA is 70.9% A+T-rich, and therefore in the range observed for other collembolan species (A+T contents spanning from 65.5% to 74.1%) 58. The observed A+T content strongly influences the use of two- and four-fold degenerate synonymous sites in protein coding genes, with relaxed pressure probably responsible for the most frequent use of NNA and NNU codons (Table 2).

Another remarkable molecular feature of metazoan mtDNAs is the asymmetry in the composition of the nucleotide content between the two strands 46. Usually, A% and C% are higher than T% and G% on the J-strand (appropriately named plus strand because of its positive AT and CG skew), whereas the reverse is observed in the N-strand (hence named minus strand). Asymmetry in nucleotide composition among strands may be due to the peculiar replication and transcription mechanisms 47. In mammals, it has been demonstrated that the duplication of the mitochondrial genome (usually defined as the 'strand-displacement' or 'leading and lagging' model) is asynchronous and requires extensive displacement of the H strand, that therefore transiently is found in a single-strand state (48; but see also alternative views in 49). The longer the time this strand is unpaired during replication (D_SSH), the more pronounced is the compositional asymmetry between H and L strands 50. In the few cases where mtDNA replication has been studied in insects (essentially Drosophila), the mechanism is also "more asynchronous" given that both origins of replication (O_N and O_J) are located in the A+T-rich region (rather than at 3/4 of the genome, as in the strand-displacement model of replication) 1, and synthesis of the major coding strand begins after 97% of the minor strand has been replicated 1251. Being the N-strand, during its single-stranded status, more prone to deaminations than double stranded DNA, A→G and C→U transitions are more likely to occurr. This mutational pressure generates inequalities in the nucleotide composition between the two strands with positive AT-skew and CG-skew values resulting on the J-strand 2047.

Remarkable strand compositional biases have been previously observed in the mtDNA of some basal hexapod lineages 314752. In the J-strand of the C. antarcticus cytosines are always more frequent than guanines, leading to a positive CG-skew (0.148). Conversely, the AT-skew is only moderately positive, as calculated over the entire genome (0.058) and in III codon positions of the J-strand PCGs (0.048), or negative, as calculated over entire PCGs (-0.082 and -0.27 for protein coding genes oriented on the J-strand and N-strand, respectively). On the other hand, remarkable negative values of AT and CG skew are observed on the N-strand (AT-skew = -0.27 and CG-skew = -0.193, for all PCGs positions; AT-skew = -0.227 and CG-skew = -0.189, for 3^rd positions only). This latter bias may be due to the high frequency of mutational events leading to deamination of A and G nucleotides of the displaced N-strand, occurring during the replication of mtDNA. Different rates of transitions (A→G < C→U; 44) may have eventually led to the asymmetry (negative AT and CG skews) of nucleotide composition and to the higher T% vs A% of PCGs oriented on the N-strand.

In this respect, the frequency of adenines and thymines is approximately equal in the genes encoded on the J-strand, whereas a higher percentage of A vs T (counting on the J-strand) is observed in those oriented on the N-strand (Figure 3). If the strand displacement model of mtDNA replication described in mammals and Drosophila also holds for Collembola, we can speculate that the block of genes spanning from rrnS to trnF (all genes oriented on the N-strand except for trnT, nad6, cob and trnS(uga); Figures 1 and 3) is more prone to the deamination type C→U than the opposite strand (due to higher D_SSH), and that transitions between C and T are more frequent than those between A and G. Perhaps this trend, observed only in the PCGs oriented on the N-strand, is dependent on a higher number of mutations occurring during transcription, rather than replication, and is consequently less etched in the PCGs encoded on the J-strand. In addition, it has been demonstrated that cytosine is the most unstable of the four bases and the number of C deaminations is more pronounced in single-strand DNAs 53.

The asymmetrical directional mutation pressure 54 observed in the two strands strongly influences the codon usage of PCGs oriented in opposite directions 555657. In this respect, NNA and NNC codons are more frequent than NNU and NNG in the PCGs encoded on the J-strand, whereas the N-strand genes show exactly the opposite trend (Figure 4). Codon usage may also be influenced by other molecular processes such as translational selection efficiency and accuracy 58, which apparently have a stronger influence in organisms with rapid growth rates 59. In addition, recent analyses of codon usage in four-fold degenerate codons of mammal and fish mtDNAs have demonstrated that there are context-dependent mutational effects (correlations between pairs of neighboring bases). In this respect, the second codon position base strongly influences the presence of a specific nucleotide at fourfold degenerate sites, and these latter are not independent from the first-position nucleotide of the following codon 60.

The nucleotide variability of each mitochondrial PCG has been estimated calculating gene-by-gene average values of genetic distances (Appendix 1: dist.jpg), across all collembolan taxa for which full (or nearly complete) mtDNAs are available (Appendix 2: list.exl). Distances values are higher for nad2, nad3 and cob, suggesting these genes to be the least conserved ones in the mtDNA of Collembola. Conversely, the more conserved mitochondrial genes are some of the cytochrome family (complexes 1 to 3) and nad1. The distribution of the observed nucleotide variability is somewhat difficult to explain. However, according to the proposed model of mtDNA replication mechanism (studied for a restricted number of hexapods taxa 1), the genes located in the vicinity of the A+T-rich region remain for a longer time in a single-strand state during the mtDNA replication and therefore are more exposed to hydrolytic and oxidative damages 50. This interpretation could explain the high levels of divergence observed for nad2. In addition, sites for the attenuation of the mtDNA transcription mechanism 43 (i.e. located in the vicinity of nad3 and cob) may also be important to increase the mutation load for neighbour genes. Collectively, these DNA fragments are probably more prone to higher substitution rates than the others due to their position along the mtDNA, although no experimental study has been performed to sustain this hypothesis. Alternatively, structural constraints at the protein level may result in site-specific selective pressures acting differently and in a scattered manner along the mitochondrial genome, with some genes bearing a higher mutational load than others.

Specimens of C. antarcticus were sampled from Killingbeck I. (Antarctic Peninsula: S 67° 32'; W 68° 7') during the 2002 polar expedition of BAS in collaboration with PNRA. Samples were frozen in liquid nitrogen and conserved al -80°C. Total DNA was extracted using the Wizard SV Genomic Purification System (Promega). The complete mitochondrial genome was sequenced by combining direct sequencing of three PCR fragments amplified with universal primers, and shotgun sequencing of three fragments obtained through Long-PCR amplification with specific primers. Initial amplification and sequencing of three short mtDNA fragments (cox1/cox2, cob and rrnL, totalling ≈ 3 kb) was performed using primers C1-J-1751/C1-N-2191, CB-J-10933/CB-N-11367 and LR-J-12887/LR-N-13398 61 using several DNA extractions from different C. antarcticus specimens. Six specific primers were designed to amplify (using an additional DNA extraction from a single specimen) the rest of the genome in three fragments of approximately 2 kb (1), 6 kb (2) and 4 kb (3) via Long-PCR: 1) CANcox2-3453J (CGCTTACTGGATGTAGACAATCGCACAG)/CANcox3-5245N (GAGCCGTACGCTGAGTCTGAAATTG), 2) CANcox3-5269J (CAATTTCAGACTCAGCGTACGGCTC)/CANcob-11444N (CACAATTGTTAAAATTTGTCCCAC) and 3) CANtrnSuga-11574J (AGTGATTAAGCACTTACCTTGAAAGCAAGCTAC)/CANcox1-3053N (CTAGAAGAGGAGAAGCTGCGTTTTGG). Long PCR conditions were the following: 1) fragment CANcox2-3453J/CANcox3-5245N, 94°C for 1 min, 60 for 1 min and 68° for 2 min 30 sec, 35 cycles; 2) fragment CANcox3-5269J/CANcob-11444N, 94°C for 1 min, 50 for 1 min and 68° for 7 min, 35 cycles; 3) fragment CANtrnSuga-11574J/CANcox1-3053N, 94°C for 1 min, 50 for 1 min and 68° for 8 min 30 sec, 35 cycles.

Amplifications were performed on a Gene Amp^® PCR System 2700 (Applied Biosystem) in 25 μl reaction volume composed of: 10.75 μl of sterilized distilled water, 2.5 μl of LA PCR Buffer II (Takara), 2.5 μl of 25 mM MgCl₂, 4 μl of dNTPs mix, 1.25 μl of each primer (10 μM), 2.5 μl of DNA template and 0.25 μl (1.25 U) of TaKaRa LA Taq polymerase (Takara). Long-PCR fragments were purified using the Microcon^® Centrifugal Filter Unit (Millipore), randomly sheared to 1.2–1.5 kb DNA segments using a HydroShear device (GeneMachines). Sheared DNA was blunt end-repaired at room temperature for 60 min using 6 U of T4 DNA Polymerase (Roche), 30 U of DNA Polymerase I Klenow (NEB), 10 μl of dNTPs mix, 13 μl of 10× NEB buffer 2 (NEB) in a 115 μl total volume and gel purified using the Wizard^® SV Gel and PCR Clean-Up System (Promega). The resulting fragments were ligated into the SmaI site of a pUC18 cloning vector using the Fast-Link DNA ligation Kit (Epicentre) and electroporated into One Shot^® TOP10 Electrocomp™ E. coli cells (Invitrogen) using standard protocols. Each resulting clone was sequenced on both strands in a CEQ 8000XL automated DNA Analysis System (Beckman Coulter). Eventually, regions of "unsatisfactory coverage" and small gaps in the sequences have been amplified and sequenced using standards protocols and different Long-PCR amplifications as template. Sequences were manually corrected and assembled with the software Sequencher 4.4.2 (Gene Codes). Long-PCR products were composed with those initially generated for short fragments to provide the complete genome sequence. Hence, the final assemblies were based on a minimum sequence coverage of 5×, that is the results of a pool of individuals, although only for a very short portion of the molecule (essentially limited to small fragments of cox1/cox2 and cob, given the supposed conservation of the amplified region of the rrnL). Screening of nucleotide sequences, obtained from specimens of the same locality, provides no extreme variability (essentially limited to the 3^rd codon positions) for either cox1/cox2 and cob fragments (data not shown).

Genes encoding proteins, rRNAs and tRNAs were identified according to their amino acid translation or secondary structure features, respectively. Individual gene sequences were compared with the homologous sequence of other collembolan species available in GenBank and inspected for the presence of gene overlaps, non-canonical start codons and truncated termination codons, and unusual structures at gene junctions. The secondary structure of tRNA genes was manually reconstructed and rendered using the software Rna Viz 2.0 62. Nucleotide variability of collembolan mtDNA (see Additional file 1: dist.jpg) was assessed using thirteen alignments of the PCGs. Basic sequence statistics and genetic distances among collembolan genes were calculated using PAUP* 4b8 63 (see Additional file 2: list.xls), whereas codon usage and RSCU values were calculated using codonw 64. Strand asymmetry was measured using the formulas AT-skew = [A%-T%]/[A%+T%] and CG-skew = [C%-G%]/[C%+G%] 4647. The presence of repeated sequences within non-coding fragments was studied using the mreps software 65. The presence of additional repeats and palindromic sequences was investigated using Blast, and hits with e-value < 0,05 were retained for further analysis.