The mosquito Aedes aegypti has a large genome size and high transposable element load but contains a low proportion of transposon-specific piRNAs

Research into the genome-wide regulation of transposons in model organisms such as Drosophila melanogaster, zebrafish and mice has revealed the importance of two small RNA pathways for controlling their movement thereby preserving genome stability [1–8]. This is especially important considering the abundance and diversity of transposons in eukaryote genomes in which unregulated movement of active elements, or the non-autonomous sequences they can re-mobilize, would lead to insertion mutagenesis throughout the genome resulting in a decrease in genetic fitness. In D. melanogaster the Piwi-interacting RNAs (piRNAs) appear to function primarily in the regulation of transposons in both the germ line and in somatic tissues that envelope the ovaries although it is clear that transposons are not the only genomic targets of piRNAs [9–13]. In addition, endo-siRNAs have been shown to target transposons in somatic tissues [10]. However, the D. melanogaster genome has a relatively small transposon load (only 3.86% of the 120 Mb euchromatic DNA and 77% of 24 Mb of sequenced heterochromatic DNA making an average of 15.8% across 144 Mb of the sequenced genome)[14, 15] compared with many other organisms. It is not clear how similar the transposon regulatory mechanisms between it and insects with much larger genome sizes and higher transposon loads might be.

The mosquito Aedes aegypti has a genome size of 1.38 Gb of which nearly half (47%) is composed of transposons [16]. It is a vector of several human pathogens, most notably RNA viruses responsible for dengue and yellow fever, and so is an insect pest of high medical importance. Ae. aegypti is somewhat amenable to modern genetic analysis through the use of transposon-mediated genetic transformation, site-specific recombinases and RNAi leading to the emergence of novel control strategies based on the manipulation of its genome [17–20]. A curiosity is that, while genetic transformation of Ae. aegypti has been achieved using the exogenous transposons piggyBac, Mos1 and Hermes, none of these appear to re-mobilized at a frequency which allows the implementation of transposon-based genetic strategies such as gene tagging, and gene and enhancer trapping [21–24]. Indeed the failure of piggyBac to retain even somatic activity in transgenic lines of Ae. aegypti in which the piggyBac transposase is expressed is in contrast to the use of piggyBac as an incisive genetic tool in D. melanogaster, Triobolium castaneum and mice in which its mobility properties allow the identification of genes and regulatory sequences based on function [24–27]. Understanding the basis of the inactivity of these exogenous transposons in Ae. aegypti is important since the ability to use these (or other transposons) as mutagens is preventing the implementation of transposon-based genetic screens used to identify genes and sequences based on function. One possible explanation for the immobility of these three exogenous transposons once they have integrated into the Ae. aegypti genome is that they are rapidly silenced by the host's small RNA system.

Our knowledge of small RNA silencing of transposons in insects is based on studies from D. melanogaster. Two pathways, the piRNA and endo-siRNA pathway, are involved however many aspects of how these pathways actually function remain unknown. The endo-siRNA pathway is dicer-dependent and generates 21 nt small RNAs that target transposons in somatic and germ line tissues, including the follicle cells of the ovary [10, 12]. They are also generated in Kc and S2 cells as well as the heads of adult flies and so are most likely dispersed through the somatic tissue of the insect [10].

The piRNA pathway generates small RNAs that are typically 24-30 nt in length and in Drosophila appears to be mainly devoted to the silencing of transposable elements ([1, 2, 6–8, 28, 29]. In D. melanogaster it requires the action of three genes, Piwi, aubergine (aub) and Argonaute 3 (Ago3), the expression of the later two being confined to the germ line [6, 30]. The piRNA pathway itself is proposed to consist of two pathways, the primary piRNA pathway and the amplification or ping-pong pathway, the later acting in germ line tissues [31, 32]. In D. melanogaster the primary piRNA pathway utilizes antisense transcripts generated from chromosomal clusters of piRNA target sequences or transposons that are then loaded onto the Piwi protein or, to a lesser extent the Aub protein, the later acting in the germ line. These piRNAs are 2'-O-methylated and act as guides to transcripts generated from either invading transposons or transposons (or genes) located in the clusters, thereby achieving suppression of transposition [33, 34]. The ping-pong pathway requires the action of the AGO3 and Aub proteins with sense transcripts derived from piRNA target sequences or transposons being loaded onto the AGO3 protein where their 3'ends are also 2'-O-methylated. Antisense transcripts originating from the same loci are loaded onto the Aub protein. Working in concert, both complexes are then capable of recognizing and slicing transcripts arising from target sequences and, at the same time, generating piRNAs that fuel subsequent amplification cycles. The outcome is an effective means by which transposons are silenced in the germ line thereby increasing genome stability. The presence of piRNA clusters in the genome also provides a memory of transposon invasions of the genome that is preserved in the female germ line and so can provide some level of immunity to the subsequent invasion of transposons recognized by the incumbent piRNA machinery [9, 35]. However in D. melanogaster it is believed that this immunity takes more than one generation to develop before it affords resistance to at least some transposons [35]. Cytotype regulation of P transposon transposition has been proposed to be controlled by, in part, the generation of piRNAs to the P transposon [35, 36].

Interestingly the Ae. aegypti genome contains an expansion of the Piwi gene family with there being a single Ago3 gene and six Piwi genes [37]. Our analysis of Ae. aegypti Piwi1 indicates that it is a truncated gene and so may not be functional. It is not possible to exactly discern which Piwi corresponds to Drosophila Aub although, based on sequence similarity, we believe Piwi2 is the most likely candidate. Ae. aegypti contains single dicer 2, ago 2 and dicer 1 genes but two ago1 genes. By inference from studies in D. melanogaster these most likely play roles in the siRNA (dicer 2 and ago2) and miRNA (dicer 1 and ago1) pathways [38].

We wished to determine the piRNA complement of Ae. aegypti and to examine whether this small RNA regulatory pathway may be responsible for the control of transposons in this mosquito, whether piRNAs were generated from a few large clusters as they are in D. melanogaster, and whether piRNAs are also generated to protein coding genes. We report the results of sequencing seven Ae. aegypti small RNA libraries from five Ae. aegypti lines (including four transgenic lines) and one D. melanogaster library using high-throughput sequencing. We show that piRNA sequences are generated from piRNA clusters and from certain protein coding genes. Remarkably for an organism with such a high transposon load we show that a much lower percentage of Ae. aegypti piRNAs map to transposons than in D. melanogaster. Indeed the majority of piRNAs appear to be targeted to protein coding genes, some of which are of viral origin.

High-throughput sequencing technology has greatly increased the opportunities to study the possible regulatory roles of piRNA and siRNA sequences. In arthropods most of the research has focused on the piRNA sequences of D. melanogaster (reviewed in [13, 58]), where it has emerged that a major role of piRNAs is the regulation of transposons. However, D. melanogaster has a relatively small percentage of its genome composed of transposon derived sequences and has no described MITEs which are present in many of other insect genomes (including other drosophilids). We have conducted an analysis of piRNAs in the mosquito Ae. aegypti, which is a vector of many arboviruses and contains a genome rich in transposon derived sequences (47%) which includes a significant percentage of MITEs derived sequences (16%) [16].

Sequencing of seven whole tissue libraries sampled only a portion of all available piRNA sequences in this mosquito. Based on sequence overlaps between these libraries we estimated that the total diversity of piRNA sequences in Ae. aegypti was within the same order of magnitude as the piRNA diversity of D. melanogaster. Given this similarity in overall piRNA diversity between Ae. aegypti and D. melanogaster it was surprising that only 19% of Ae. aegypti piRNAs mapped to annotated transposon sequences, compared to 50% of piRNAs in our D. melanogaster library. Barring a large number of unrecognized transposons in Ae. aegypti, this suggests that the role of piRNAs in transposon regulation in Ae. aegypti follows the D. melanogaster model in some, but not all, respects.

The majority of the 19% of piRNAs that did map to transposons in Ae. aegypti mapped to the antisense transcription strand, a pattern that was previously observed in D. melanogaster, and is consistent with having transposon transcripts sliced by PIWI proteins loaded with piRNAs that are antisense to the transposon transcript [31]. Ae. aegypti piRNA sequences mapped to all transposon classes including MITEs (Table 2). MITEs may be transcribed but do not produce a functional protein product and so, in order to be mobilized require the presence of a corresponding full length and active transposase [67]. Ae. aegypti MITEs show little sequence similarity to full length transposons making it unlikely that many of the observed piRNAs mapping to MITEs are simply remnants of full length transposon inactivation systems. However, only 0.26% of Ae. aegypti piRNAs mapped to MITEs despite them comprising 16% of the assembled genome. Conversely 4% of Ae. aegypti endo-siRNAs mapped to MITEs which is perhaps not surprising given that transcription along the length of a MITE would produce a foldback dsRNA sequence that would elicit the siRNA response. A link between transposons, miRNAs and gene regulation was first proposed from studies in humans and has been refined, based on analyses performed in plant genomes, to include both siRNAs and MITEs [68–71].

We were able to identify a number of piRNA clusters that generated a large percentage (84%) of observed piRNAs. A number of lines of evidence supported these genomic locations as having a similar role to D. melanogaster clusters. First, previously described D. melanogaster clusters were also recovered using our cluster discovery procedure applied to our D. melanogaster library (Table 3). Second, all Ae. aegypti libraries supported the same basic piRNA cluster locations (Table 4). Third, the top 30 Ae. aegypti clusters appeared to overlap with transcribed portions of the genome (Figure 5). Finally, these clusters shared some common features with previously described D. melanogaster piRNA clusters. These included: roughly similar ranges in individual cluster lengths; a mixture of clusters with piRNAs mapping almost exclusively to one strand and clusters with piRNAs mapping to both strands; and similarities in the relative orientation of piRNA, transposon, and transcription orientation for at lease some clusters. Some piRNA clusters in D. melanogaster and mouse are transcribed unidirectionally. This is the case for the D. melanogaster flamenco locus in which 99% of piRNAs map to the sense strand of transcription, while all the transposon sequences are oriented in the direction opposite to transcription [31]. Many of the top Ae. aegypti piRNA clusters also contained transposon sequences predominantly oriented in the same direction (Table 5). Unfortunately, overall cluster transcription direction could not be determined for all clusters. However, among the 30 piRNA clusters in Table 4 three overlapped with protein coding genes that were also identified as generating piRNAs (cluster 1, gene AAEL010454; cluster 3, genes AAEL007861 and AAEL007866; cluster 29, gene AAEL006159; Tables 4 and 7). If we assume that the clusters are transcribed in the same orientation as these genes we see a similar pattern in clusters 1 and 3 to the D. melanogaster flamenco locus: piRNAs are on the sense strand of transcription while transposon sequences are oriented in the opposite direction to transcription (in cluster 29 transposon were not predominantly oriented in any one direction). These observations further reinforce the similarity between Ae. aegypti and D. melanogaster piRNA clusters. However, while piRNA clusters in D. melanogaster and mouse generate sequences that predominantly map to transposons [4, 6] fewer than a quarter of potential piRNAs generated from Ae. aegypti clusters matched known transposons. Furthermore, while many previously described piRNA clusters contained a high density of transposon sequences we did not detect significantly higher levels of transposon sequences within Ae. aegypti clusters compared to random portions of the genome (Table 5). However Ae. aegypti piRNA clusters covered a greater portion of the assembled genome than D. melanogaster clusters and so may be more widespread. These results suggest that while Ae. aegypti and D. melanogaster share many features of their piRNA clusters, the role these clusters have in transposon inactivation may not be completely identical between these species. The nature of this difference has yet to be determined.

We examined the piRNA density maps to seven Ae. aegypti transposons and found these to be similar in their patterns to equivalent density maps from D. melanogaster [6, 9, 35]. All but the non-LTR MosquI element contained at least on ping-pong amplification overlap suggesting that these could be silenced by this pathway. Notably these ping-pong signatures were present in representatives of the two LTR elements that together account for 58% of piRNAs. The most marked anti-sense strand bias was observed for the non-LTR JuanA element which has been proposed to be recently active in mosquitoes [52]

To better understand the role of the 81% of Ae. aegypti piRNAs that did not map to transposons we examined possible associations between protein coding genes and piRNA sequences. In addition to their role in transposon transcript degradation, piRNA sequences have been demonstrated to silence protein coding genes in D. melanogaster [60, 72]. The Supressor of Stellate (Su(ste)) and Traffic Jam (tj) genes contain piRNA sequences on their sense strand that can be used to degrade the transcripts of the Stellate (Ste) and Fasciclin 3 (Fas 3) gene transcripts respectively [60, 72]. Ae. aegypti contained a number of genes with piRNAs mapping almost exclusively to their sense strand that are therefore unlikely to be involved in the regulation of the host gene transcript (Table 7). Instead, they may be used to regulate other genes, the identity of which cannot be deduced using the present data. However, it is interesting to note that just as the D. melanogaster tj gene has been identified as a source of piRNAs, a possible ortholog of tj in Ae. aegypti may also be a source of piRNA sequences (AAEL007686, Table 7). Furthermore, a putative ortholog to the D. melanogaster Fas 3 gene has been identified in Ae. aegypti (AAEL003044, OrthoDB http://cegg.unige.ch/orthodb4) suggesting that the Ae. aegypti AAEL007686 gene has at least the potential to act in a similar way to the D. melanogaster tj gene. One difference in the location of the sense-strand piRNAs arising from both Ae. aegypti and D. melanogaster is that for Ae. aegypti tj they are located upstream from the translation termination codon rather than being located within the 3' UTR.

The location of piRNAs to the sense strand of the 3' UTR of the Ae. aegypti maelstrom (mael) gene suggests that these piRNAs may also be involved in the regulation of downstream genes, as had been proposed for D. melanogaster tj [60] However as yet we are unable to identify these target genes. In D. melanogaster, mael is associated with both the nucleus and nuage of germ line cells [64]. Mutations in Mael, as well as in other components of the nuage, such as vasa and Krimper, have been shown to reduce the levels of piRNAs associated with the HeT-A, roo and I transposons suggesting that these genes play a role in the suppression of transposition in the D. melanogaster female germ line [66]. Consistent with this is the role of Mael in mouse spermatogenesis in which its absence led to a 100-fold increase in L1 expression and a 3-5 fold increase in the expression of the unrelated IAP element [73]. As was seen for both Ae, aegypti and D. melanogaster tj, the piRNAs map to the sense strand of mael but differ from Ae. aegypti tj in that they are located in the 3' UTR. We find it encouraging that two of the top 30 piRNA generating Ae. aegypti genes have previously implicated in the regulation of either piRNAs or transposons in D. melanogaster and so suggests that our own bioinformatics screening of these libraries is generating valid targets. Seven other protein coding genes also generated piRNAs only to the sense strand and all remain unannotated (AAEL011224, AAEL017228, AAEL005277, AAEL005213, AAEL009263, AAEL013013, AAEL011027)(Table 7). It remains to be determined if piRNAs from many of these genes arise from the exonic sequence since many current Ae. aegypti gene annotations have not been manually curated and are based mostly on relatively poor EST datasets. Robine et al. (2009) noted that in mouse piRNA libraries many piRNA clusters that were once believed to be exonic could, upon reexamination, be reclassified as 3'UTR directed. This same phenomenon of high density of piRNAs on the sense strand of the 3' end of the ORF was also observed in the case of the putative maelstrom homolog gene described above.

Ae. aegypti is a vector of many RNA viruses, some of which cause severe disease in humans. Eight of the 30 top piRNA generating genes in Ae. aegypti are apparently of viral origin (Table 7). Three of these generate piRNAs only to the antisense strand (AAEL007866, AAEL017001, AAEL007844,) while another (AAEL009873) generates 99% of it's piRNA to this strand. The remaining three (AAEL017355, AAEL000120, AAEL001772) generate piRNAs mapping to both strands, although for each it is the antisense strand that dominates. This may indicate that most of the piRNAs generated to viral-like genes function by, in association with the appropriate Piwi protein, slicing the viral gene transcripts. As such this mechanism is entirely different to that operating for tj, mael and the seven other protein coding genes in which the piRNAs are generated exclusively by the sense strand. All of the remaining 12 protein coding genes that generate piRNAs have these mapping primarily to the antisense strand.

There has been some recent evidence implicating piRNAs in the recognition of arboviruses in Ae. aegypti and Aedes albopictus. In addition to siRNAs, small RNAs 24-30 nt in length to the sense strand of the dengue virus genome were recovered from infected Ae. aegypti and none showed a bias for uracil at the 5' end and little bias for adenine at the 10^th position although these authors stated that unpublished data revealed that small RNAs of the same size distribution generated to the sense strand of Sindbis virus did show a U1 bias [63]. Aedes albopictus C6/36 cells have been found to lack an siRNA response to infection by West Nile virus, Sindbis virus and La Crosse virus, but do generate small peaks of smaller RNAs 24-28 nt in size to Sindbis and La Crosse virus infections yet no such peak is generated by infection with the dengue virus [62, 74]. Interestingly small RNAs within this size range generated to the inadvertent infection of C6/36 cells by Cell Fusing Agent virus showed a strong preference for adenine at the 10^th position which is consistent with them being piRNAs that interact with, in D. melanogaster, the AGO3 protein [74]. Taken together these data from two difference Aedes species indicate that a piRNA response to arboviral infection may be generated and, if so, implicate this pathway in an anti-viral response. Taken in this context, the piRNAs generated by the viral-like sequences identified here may be further evidence of the role that this small RNA pathway may play in anti-viral defense in this mosquito. Ae. aegypti may thus provide important and novel information concerning how this small RNA pathway interacts with both transposons and viruses, both of which are abundant in this insect, especially in comparison with D. melanogaster.

A large portion of transposon-matching Ae. aegypti piRNAs mapped inside genic sequences for reasons that remain unclear (Figure 1A). Some protein coding genes are likely origins of piRNAs (Table 7), but it appears unlikely that these would be sufficient in number to account for this observation. A possible, but as yet untested, explanation might be that these regions contain a high level of active transposons. Genic regions of the genome are more likely to be transcribed, which may increase the chances that an inserted transposon will actively transpose. This in turn could produce a higher response of piRNA silencing mechanism to these transposons.

As a vector of human disease pathogens there is interest in developing highly robust genetic tools for Ae. aegypti. While germ line transformation is possible (albeit a low rate compared to D. melanogaster) efforts to remobilize transposons in Ae. aegypti have occurred at very low rate suggesting the presence of a transposon silencing mechanism [21, 23, 41, 75]. We examined the piRNA content of mosquito lines that had been transformed with transposon sequences and found preliminary evidence that piRNA sequences mapping exclusively to the transformed sequences had been produced. Since these piRNAs did not match the current Ae. aegypti assembly their presence in the transformed lines was likely explained in one of two ways: 1) they had been maternally inherited and perhaps amplified via the ping-pong cycle, or 2) new piRNAs were being generated from introduced sequence. In either case these data are suggestive that a component of the piRNA pathway was activated by the insertion of foreign DNA into the genome although we have no information as to how rapid this response would have occurred. The full kinetics of the piRNA response to transgenic sequences need to be explored in association with genome-wide transcriptional analyses which should shed light on the relationship between transgenesis, the small RNA response and viral infection in this mosquito

We analyzed piRNA and endo-siRNA sequences from Ae. aegypti, a mosquito that is a significant vector of human pathogens and has a large genome size with a correspondingly high transposon content. Unlike D. melanogaster, Ae. aegypti contains MITEs and we found higher levels of siRNAs targeted to these than piRNAs. The terminal inverted repeats of MITEs most likely enables foldback RNAs to be formed from unidirectional transcripts, leading to the induction of the siRNA pathway, which is associated more with anti-viral defense. Despite having an abundance of transposons, the majority of piRNAs in Ae. aegypti were targeted to non-transposon sequences, many of which were protein-coding genes. As such the piRNA profile of this mosquito is more similar to that of mice than D. melanogaster in which the majority of piRNA sequences map to transposons. The majority of piRNAs in this mosquito were 28 nt in length and so longer than those seen in D. melanogaster but contained the U1 or A10 sequence bias seen in other organisms in which piRNAs have been sequenced. Two genes targeted by piRNAs in Ae. aegypti have been implicated in piRNA biogenesis or function while the function of the majority of them remain unknown. Several others were of viral origin suggesting that the piRNA response may extend into anti-viral defense in this insect. piRNAs were also generated to introduced transposons. The diversity of endogenous transposons present in this mosquito, together with the corresponding diversity and number of piRNAs and siRNAs mapping to them suggests that these small RNA pathways may of some importance in maintaining the integrity of its genome in the presence of numerous transposons and viruses.