Expression of Small RNA in Aphis gossypii and Its Potential Role in the Resistance Interaction with Melon

Background The regulatory role of small RNAs (sRNAs) in various biological processes is an active area of investigation; however, there has been limited information available on the role of sRNAs in plant-insect interactions. This study was designed to identify sRNAs in cotton-melon aphid (Aphis gossypii) during the Vat-mediated resistance interaction with melon (Cucumis melo). Methodology/Principal Findings The role of miRNAs was investigated in response to aphid herbivory, during both resistant and susceptible interactions. sRNA libraries made from A. gossypii tissues feeding on Vat+ and Vat− plants revealed an unexpected abundance of 27 nt long sRNA sequences in the aphids feeding on Vat+ plants. Eighty-one conserved microRNAs (miRNAs), twelve aphid-specific miRNAs, and nine novel candidate miRNAs were also identified. Plant miRNAs found in the aphid libraries were most likely ingested during phloem feeding. The presence of novel miRNAs was verified by qPCR experiments in both resistant Vat+ and susceptible Vat− interactions. The comparative analyses revealed that novel miRNAs were differentially regulated during the resistant and susceptible interactions. Gene targets predicted for the miRNAs identified in this study by in silico analyses revealed their involvement in morphogenesis and anatomical structure determination, signal transduction pathways, cell differentiation and catabolic processes. Conclusion/Significance In this study, conserved and novel miRNAs were reported in A. gossypii. Deep sequencing data showed differences in the abundance of miRNAs and piRNA-like sequences in A. gossypii. Quantitative RT-PCR revealed that A. gossypii miRNAs were differentially regulated during resistant and susceptible interactions. Aphids can also ingest plant miRNAs during phloem feeding that are stable in the insect.


Introduction
Gene regulation at the post-transcriptional level has garnered considerable attention in recent years due to an increased understanding of the regulatory roles of small RNAs (sRNA). The first regulatory sRNA was discovered in Caenorhabditis elegans almost twenty years ago [1], and since then, progress in deep sequencing technologies combined with sophisticated bioinformatic analyses have facilitated the identification of a large number of sRNAs. Three main groups of animal sRNAs: microRNAs (miRNAs), endogenous small interfering RNAs (endo-siRNA) and Piwi-interacting RNAs (piRNAs) have been classified, based on their distinctive characteristics, biogenesis processes, and association with Argonaute proteins [2]. Among these three classes, miRNAs are the most studied and best understood. The biogenesis of miRNAs involves the action of two RNase III proteins, Drosha and Dicer, that are required for processing the hairpin shaped structures to 22-23 nucleotide (nt) long mature miRNAs. The 21 nt endo-siRNAs in insects and mammals are produced in a RNA-dependent RNA polymerase (RdRP) independent manner, requiring a Dicer-dependent process [2]. Endo-siRNAs mainly originate from transposon transcripts, intergenic repetitive elements, and endo-siRNA cluster loci. piRNAs also originate from intergenic repetitive elements, including retro-transposons, but do not require Dicer for processing. piRNAs, originally reported from Drosophila melanogaster germ cells [3], are easily distinguished from the other two classes due their longer size (24-29 nt). This group of sRNA interacts with Piwi proteins and silences selfish genetic elements contributing towards germ line stability [4,5]. Recently piRNAs from neural cells were shown to play a role in the epigenetic control of memory related synaptic plasticity [6].
Traditionally most of the well characterized insect sRNAs were restricted to miRNAs sequences from Drosophila spp.; however with the recent availability of RNA-seq and whole genome sequencing data, miRNA sequences from other insect species such as silkworm (Bombyx mori), honeybee (Apis melifera), mosquito (Anopheles gambiae), locust (Locusta migratoria), beetle (Tribolium castaneum), pea aphid (Acyrthosiphon pisum), German cockroach (Blatella germanica) and brown planthopper (Nilaparvata lugens) have been deposited in miRBase [7,8]. Interestingly, the orders Diptera (Drosophila spp. and A. gambiae) and Lepidoptera (B. mori) contribute most of the insect miRNA sequences in miRBase. The depth of sRNA sequence information from the other insect orders is still limited. Pea aphid and brown planthopper are the only Hemipterans represented in miRBase [9,10] and A. pisum is the only aphid species to have miRNA sequences reported [9]. The order Hemiptera includes harmful agricultural insect pests such as aphids, whiteflies, scales, and hoppers. Many of the insects belonging to the order Hemiptera possess specialized mouth parts capable of piercing plant tissues and feeding directly from the vasculature. Piercing-sucking insects not only deplete photoassimilates, but also transmit pathogenic plant viruses. Hemipterans are hemimetabolus as they do not undergo complete metamorphosis like other insect groups. Aphids are typically parthenogenetic during most part of their life cycle and give live birth to genetically identical nymphs. The unique feeding habit combined with the ability to rapidly reproduce makes these insects some of the most damaging pests of economically important crops worldwide.
Aphis gossypii (cotton-melon aphid) is a destructive insect pest on a wide range of economically important host plants. Chemical control measures for A. gossypii are becoming limited as this species developed resistance to a wide variety of insecticide classes [11]. Therefore, significant research effort has focused on identifying and developing host-plant resistance. Resistance to A. gossypii imparted by the Vat (virus aphid transmission) gene has been identified in several geographic Cucumis melo plant introductions and introgressed into commercial melon lines [12,13,14,15,16]. The Vat gene is a member of CC-NBS-LRR family of plant resistance (R) genes, which confers dual resistance to A. gossypii and non-persistent viruses transmitted by this aphid [12,17]. Resistance against A. gossypii exhibits two possibly overlapping, modes of action; antixenosis (non-preference of the aphids) and antibiosis (reduction in aphid performance and fecundity) [18]. The antixenotic component of the resistance trait is characterized by delays in sustained phloem sap ingestion on resistant plants due to rapid interruption in feeding after the stylets penetrate the phloem [19]. The antibiotic component extends the pre-reproductive period of the aphid and shortens both the reproductive and postreproductive periods resulting in fewer progeny [19]. The overall life span of an individual aphid is reduced and after the final molt, aphids feeding on the resistant Vat + plants are smaller in size than those feeding on the susceptible Vat 2 melon plants. Thus, important reproductive, developmental, and morphological changes occur in aphids in response to Vat-mediated resistance. In this study, a survey of sRNAs from A. gossypii feeding on Vat + and Vat 2 melon plants was conducted to better understand the role of these regulatory, non-coding RNAs in the developmental changes that occur in response to the Vatmediated resistance mechanism.

Sequence analysis of the aphid libraries
Two A. gossypii sRNA libraries were generated from aphids collected after 48 hours of feeding on Vat + (Vat + aphid library) and Vat 2 (Vat 2 aphid library) melon plants, respectively. The libraries were sequenced using Illumina GAII analyzer and the library data is available at NCBI Gene Expression Omnibus (GEO) under the series GSE38641. Illumina sorts the sequence data in a ''Tag_Count'' file format, where each sequence read (tag) is provided with the number of times (counts) it appeared in the library. The sRNA sequencing data in the tag count files have been used for in depth analyses.
The Vat + aphid library generated approximately 2.5 million reads, and after adapter trimming and removing redundant sequences, the number of usable sequences was 918,621. Similarly, for the Vat 2 aphid library approximately 1.2 million reads were obtained and the usable number of sequences was 974,781. Before the libraries were searched for putative miRNA sequences, rRNAs and tRNAs were discarded and the number of usable sequences after this step was 32,668 and 45,898 for Vat + and Vat 2 aphid libraries, respectively (Table 1). Reads with counts of 10 or more in the size range of 18-30 nt were retained and designated as distinct reads (File S1). The distribution of sRNAs was strikingly different in two aphid libraries. The size distribution pattern of the reads revealed that Vat + aphid library was over-represented by 26-27 nt sequences, whereas the Vat 2 aphid library was dominated by 22 nt sequences ( Figure 1). Both the libraries were further analyzed to determine their count complexity ( Figure 2). In the Vat + aphid library, reads in the 26 nt size category showed very high abundance, with two reads accounting for $100,000 counts each, comprising 45% of that size category, thereby making this size category the most prominent in the Vat + aphid library (Figure 2A). In the 27 nt size category; a single read accounted for over 100,000 counts (12.5%), four reads contributed to counts in the range of 50,000-99,999 each (18%) and another four reads recorded counts in the range of 25,000-49,999 each (8%) (Figure 2A). The Vat 2 aphid library showed a bias towards 22 nt sequences ( Figure 1); based on the count distribution seven reads recorded counts of $100,000 each (48%) and eight reads contributed to counts in the range of 50,000-99,999 each (10%) ( Figure 2B). The 23 nt size category in the Vat 2 aphid library was represented by six reads with counts $100,000 each (50%), one read in the 50,000-99,999 count category (2.3%), and seven reads in the 25,000-49,999 count category (6.7%). The number of reads in each count category for both the libraries is listed in File S2.

Identification of conserved miRNA in aphid libraries
Conserved A. gossypii miRNAs were identified by searching the sub-set of sRNAs, designated as distinct reads, against the repository of animal miRNAs in miRBase (Release 18, http:// www.mirbase.org/). During homology searches, two mismatches were allowed, and 81 miRNA sequences belonging to 56 miRNA families were identified ( Table 2) from both libraries combined. As observed from the size distribution counts, the number of 22 nt sequences was noticeably lower in the Vat + aphid library, which was reflected by the number of miRNA sequences identified in that library. In comparison, to the 77 miRNA sequences identified in the Vat 2 aphid library, only 69 miRNA sequences were reported from the Vat + aphid library ( Table 2). Several miRNA families including miR-100, miR-137, miR-307, miR-316, miR-3041 and miR-3042 were absent in Vat + aphid library. Certain miRNAs like miR-133a, miR-1357, miR-29b and miR-87b were not found in Vat 2 aphid library. Interestingly, some miRNA families (miR-10, miR-184, miR-2, miR-276, miR-87, miR-9 and let-7) were represented by fewer members in the Vat + aphid library, whereas in Vat 2 aphid library these families were strongly represented by more members. The abundance of these miRNA sequences in each library is reported in 'transcripts per million' (TPM). The TPM profile represents a measure of miRNA expression levels [20], and the results show different patterns of accumulation for the conserved miRNAs between the two libraries ( Table 2). A general trend towards lower miRNA accumulation was observed in Vat + aphid library. However, some miRNA families like miR-1357, miR-1692, miR-184, miR-310, miR-996 and miR-998 showed enhanced accumulation in Vat 2 aphid library, as suggested by the library counts ( Table 2). Library counts have been useful to survey expression profiles for further investigation [10,21]. In addition to conserved miRNA sequences, miRNA* sequences were also identified for miR-276 and miR-1 from both the libraries.
Twelve aphid-specific miRNAs were identified in the aphid libraries (Table 3). Till date, aphid miRNAs have been reported only from A. pisum [9]. The TPM profile revealed that eight out of the twelve aphid-specific miRNAs were expressed in Vat + aphid library (Table 3). Ap-miR-X15, Ap-miR-X43, Ap-miR-X71, and Ap-miR-X81 were not reported from the Vat + aphid library ( Table 3).

Identification of novel A. gossypii miRNAs
Sequences in both the libraries were analyzed for the presence of novel A. gossypii miRNAs. A. gossypii EST sequences (http:// www.aphidbase.com/aphidbase/downloads) and the genome sequence of A. pisum (http://www.aphidbase.com/aphidbase/ downloads) were used by the miRDeep algorithm to map the precursor sequences in order to identify novel miRNAs. The miRDeep algorithm uses the probabilistic model of the miRNA biogenesis to score the compatibility and the position of the sequence sRNA to the secondary structure of the miRNA precursor [22]. Because animal miRNAs have a small precursor length, 100 bp regions were folded into stem-loop structures to identify the putative novel miRNA sequences ( Figure 3). Nine candidate novel miRNAs were identified from A. gossypii (Table 4). Stem-loop quantitative real-time PCR (qPCR) was performed to validate the expression of these candidate novel miRNAs. Seven out the nine yielded PCR products that could be detected ( Figure 4). Ago-miR-C1, Ago-miR-C2 and Ago-miR-C4 were significantly (p#0.05) up-regulated in Vat + aphid library. Ago-miR-C3 was also up-regulated, with a p-value of 0.051 that was just below the significance threshold. Ago-miR-C7 was the only novel miRNA in the Vat + aphid library to show significant downregulation (p#0.001).

Identification of plant miRNA families in the aphid libraries
Six plant miRNA families were identified from the aphid libraries; miR156/miR157, miR166, miR168, miR2911 and miR2916 ( Table 5). All of these plant miRNAs were also present in melon sRNA libraries [23,24]. Two of the six, miR166 and miR168, were identified in the Vat + aphid library. However, confirmatory experiments using stem-loop qPCR could detect only miR166, miR168 and miR2911 in both Vat + aphid and Vat 2 aphid tissues (data not shown). The plant derived miR156/ miR157 and miR2916 did not yield detectable amplification products. To further validate if sRNAs could be transferred to aphid tissue via phloem feeding, an in-vitro feeding assay was performed with 59 [c -32 P]-ATP labeled synthetic double stranded (ds) 21 nt oligomers. The [c -32 P]-ATP labeled ds-21 nt was presented to the aphids over a four day period in a liquid diet composed of 0.5 M sucrose enclosed within two parafilm layers ( Figure 5A). A disc of blotting paper placed at the bottom of the container was used to collect honeydew (excreta) droplets. After four days of feeding, phosphor-imaging detected strong signals on the discs in patterns that were consistent with the distribution of honeydew ( Figure 5B). RNA extracted from both aphid tissues and honeydew eluted from the discs was separated by PAGE and exposed to X-ray film revealing the [c -32 P]-ATP labeled synthetic double stranded (ds) 21 nt oligomers in each sample ( Figure 5C).

Analysis of the longer (26-27 nt) sRNAs
The Vat + aphid library had a strong representation of the longer sRNA sequences (26-27 nt) ( Figure 1). Remarkably, the accumulation of these 26-27 nt sequences was about 30 times higher in Vat + aphids in comparison to Vat 2 aphids. In silico experiments were performed to identify these longer sRNA sequences and to map their origin. Previous studies have shown well-characterized piRNA sequences tend to have lengths ranging between 26 and 30 bases and are often derived from repetitive elements or originate from transposons. In the absence of an assembled and annotated genome database for A. gossypii, homology-based searches for transposon-like origins for the subset of sRNA sequences with lengths greater than 26 nt was performed using the available transposon database of A. pisum. Results of homology searches from the Vat + aphid library revealed 942,169 reads out of the total 2,064,014 (46%) in 26-27 nt long read category that were mapped to transposons (data not shown). Similarly for Vat 2 aphid library, 51% of the reads in the 26-27 nt category originated from the transposable elements. Based on the A. pisum annotation, the putative repeat classes identified from the distinct sequences include Polintons, miniature inverted repeat-transposable elements (MITEs), Helitrons, terminal inverted repeats (TIRs), long terminal repeats (LTRs), long interspaces nuclear elements (LINEs), short inverted nuclear elements (SINEs) and a large diverse class of unclassified transposable elements (NoCats). An informal examination of the base composition of the sequences shows a depletion of adenosine nucleotides. A thorough exploration of the sequences using the DREME program [25] in the MEME motif finding software package (version 4.8.1) [26] discovered an overrepresentation of GC-rich motifs in both libraries with the top five motifs found in each library shown in Table 6. The longer sRNA sequences were also analyzed to determine if any of the sequences originated from the endosymbiont genome. Aphids accommodate the primary endosymbiont Buchnera aphidicola within specialized cells called bacteriocytes that contribute to the overall nutritional health of the aphid [27]. Homology searches allowing one mismatch with the Buchnera genome revealed 4.6% of the 26-27 nt sequences in the Vat + aphid library and 0.85% in the Vat 2 aphid library were of bacterial origin (data not shown).

Target prediction of miRNAs
In silico target prediction analysis was conducted to further understand the role of the different sRNA molecules identified in the aphid tissue. While several computational methods exist for predicting mRNA targets of metazoan miRNAs [28,29], the PITA program was selected because the prediction is not based on target-site conservation, yet the program can detect species-specific miRNA targets [30]. As a control, the PITA program was used to compute miRNA targets in the well characterized D. melanogaster genome. All the sequences for the 304 miRNAs and 39-UTRs from the D. melanogaster genome database [31,32] were used for the control analysis. The D. melanogaster miRNA-target prediction results corroborated with the results from previous D. melanogaster miRNAs target studies (data not shown). A total of ,2,600 targets were predicted for all the identified A. gossypii miRNAs, including the conserved and novel sequences (Table 7) using the A. gossypii Unigenes. Since a single miRNA can target more than one mRNA, the number of miRNA-target pairs was about 4,130 ( Table 7). The PITA program predicted several isoforms and  multiple UTRs as targets for particular miRNAs. After removing redundancy, the number of targets for A. gossypii conserved miRNAs identified from the A. pisum 39 UTR sequences was approximately 4,500. Following similar analyses, 814 targets were predicted for the nine candidate novel miRNAs (File S5). The biological functions of the predicted targets of the A. gossypii miRNAs were analyzed using the currently available gene ontology (GO) annotations of the A. pisum genes. The A. pisum genome database contains 46,360 GO terms for 11,440 genes. Using protein homology information, GO terms associated with all the A. gossypii sequences with significant hits to the A. pisum proteome were extracted. Table 8 summarizes the homology and annotation results for all four sequence data sets used. Similar patterns of functional annotation were observed for all the four databases (File S3) used in target predictions of the A. gossypii miRNAs (File S6). The largest single group of the target genes (9.2% of the total annotated GO terms) identified was annotated to have functional roles in morphogenesis and anatomical structure determination ( Figure 6). Genes belonging to signal transduction pathways (8.7%), cell differentiation (8.4%), and catabolic processes (6.6%) were also frequently identified as targets.
The RNAhybrid program [33] was also used to obtain a more conservative set of miRNA target predictions for A. gossypii. GO annotations for these datasets are provided in File S7. Between 38-43% of the A. gossypii mRNA targets were identified in both RNAhybrid and PITA predictions when searched against A. gossypii ESTs and unigenes and A. pisum genome. The number of overlapping targets increased to 46% when using the M. persicae EST database with both methods. These results reflected differences in the two prediction algorithms and their cut-off threshold values and metrics.

Discussion
The regulatory role of sRNAs in biological processes such as development, metabolism and environmental stress responses has been well documented across different taxa. In recent years, the involvement of insect miRNAs has been studied in the context of the well-defined process of metamorphosis [34,35,36,37]. However, there has been limited progress in understanding the roles of miRNAs and their target genes in insect-host interactions, particularly those interactions that involve genetic host plant resistance. Host plant resistance to aphids in melon is due to the presence of the R-gene Vat. Interesting phenotypic differences were observed in aphids during the resistance interaction [38]. When aphids are forced to survive on the resistant host plant they show reduced growth, retarded development and decreased reproduction as compared to aphids thriving on the susceptible host. This study examined the molecular mechanism of aphid responses to R gene-mediated host plant resistance by comparing two aphid sRNA libraries generated from A. gossypii feeding on Vat + (resistant) and Vat 2 (susceptible) melon plants.
An interesting feature of the two libraries was the different size distribution patterns found in aphids feeding on susceptible or resistant plants. The Vat 2 aphid library generated from aphids feeding on a susceptible host plant showed an unimodal distribution that was highly represented by 22-23 nt sequences of non-coding RNA that were identified as miRNAs ( Table 2). Aphids feeding on resistant plants showed a bimodal size distribution pattern of sRNAs with additional enrichment of 26-27 nt sequences (Figure 1). Unimodal size distributions with miRNAs as the most abundant size class has been observed in German roach and Aedes mosquito [39,40], whereas, bimodal size distribution with abundance peaks at 20-22 nt and 27-29 nt have been reported during different developmental stages in silkworm and brown planthopper and for blood fed female Culex quinquefasciatus [10,37,40]. A shift in size distribution from unimodal to bimodal occurred between two developmental phases in migratory locust [35]. In locusts, the 22 nt peak was identified as miRNAs and the peak of longer sRNAs was due to the presence of piRNA-like sequences. Wei and co-workers (2009) [35] concluded that 26-29 nt sRNAs are involved in the process of phase changes in locust, with the development of each form (gregarious and solitary phase) showing a distinct expression pattern of specific types of sRNA. Like locusts, aphids exhibit a high degree of phentoypic plasticity and can develop different forms (alate versus apterous and sexual versus asexual). The 22 nt peak in the aphid libraries have been confirmed as miRNAs ( Table 2). In the absence of A. gossypii genome sequence the 26-27 nt category of longer sRNAs were mapped to the repeat elements from A. pisum. Mapping analysis revealed that 46% of the longer sRNA from Vat + aphid library originated from repeat elements in the genome and could be classified as piRNA-like elements. The abundance of piRNA-like sequences in the Vat + aphid library raises questions about their involvement in aphid responses to Vat-mediated resistance. A detailed analysis of the miRNAs expressed in the two aphid libraries also showed differences in accumulation of miRNAs between the two libraries. The proportion of miRNAs in the Vat 2 aphids was approximately five times greater than the Vat + aphids (Figure 1). Aphids feeding on resistant Vat + plants lacked many of the miRNAs that were expressed in the Vat 2 aphids (Table 2).
Differences in the accumulation pattern were also observed between both the libraries for most of the conserved miRNAs; the Vat + aphid library showed a general trend towards reduced accumulation of miRNAs ( Table 2). Similar variability was also observed for aphid-specific miRNAs between the Vat + and Vat 2 aphid libraries, where four of the twelve families were not detected in Vat + aphids (Table 3). However, some miRNAs (miR-133a, miR-1357, miR-184a, miR-184b, miR-2b, miR-310, and miR-998) showed enhanced accumulation in the Vat + aphid library, indicating their potential role in the resistance interaction. Further evaluation is required to determine the importance of the differential expression of these miRNAs and their involvement in the resistance mechanism. The expression profile of the novel miRNAs also showed differences between the libraries. Quantitative PCR analysis detected most of the novel miRNAs identified in the A. gossypii libraries. The expression of Ago-miRC1, Ago-miRC2 and Ago-miRC4 were significantly up-regulated in the Vat + aphids ( Figure 4); however, amplification products were not detected for Ago-miRC5 and Ago-miRC6, which could be the result of limited primer specificity.
The differential accumulation of miRNAs in the two libraries suggests that these non-coding RNAs play some role in regulating aphid developmental processes in the resistance interaction. Crossgenome conservation of gene targets often aids in identifying the functionality of miRNAs. The regulatory roles of conserved miRNAs have been investigated extensively in D. melanogaster, C. elegans and mouse. Several conserved miRNAs have experimen-  Figure 5. In vitro feeding assay with A. gossypii. A. Sachet feeding system. B. Honeydew droplets collected on a blotting sheet viewed by phosphor-imager C. 15% PAGE analysis of RNA and honeydew isolated from the aphids feeding on labeled diet and exposed to autoradiograph. tally validated roles; miR-1 is involved in muscle growth and cardiogenesis, miR-14 in fat metabolism, miR-9 in cellularization, miR-311/miR-312 in dorsal closure and miR-315 in wing development [41,42,43]. The most abundant miRNA expressed in Vat 2 aphid library was miR-1, and its enhanced expression could induce muscle development by down-regulating myostatin [44]. Another miRNA with enhanced expression in the Vat 2 aphid library, miR-133, originates from the same loci as miR-1 in mouse and is transcribed in a tissue specific manner during skeletal muscle development in both mouse myoblast cells and Xenopus laevis embryos [45]. Activation of wing development by ectopic miR-315 in Drosophila, which was abundant in the Vat 2 library, could be an appropriate response for aphids feeding on a quality host plant as population growth results in crowding effects, which induce wing formation [43,46]. Loss of function experiments in Drosophila showed miR-184 plays a role in oogenesis and egg development [47]. Differential expression of miR-184 in aphids could be indicative of changes in resource allocation from or to reproduction, particularly as aphid ovariole number has been found to be influenced by host quality [48] Recently, an important regulatory role has been demonstrated for miR-310 in neurotransmitter release in Drosophila [49], and this miRNA showed enhanced expression in Vat + aphid library.
In silico target prediction for A. gossypii miRNAs were performed using PITA and RNA hybrid program [30,33]. In silico detection of bona fide targets of metazoan miRNAs is a challenging task because target sites lack extensive Watson-Crick base pairing and hence, the miRNA recognition elements are very short motifs [28,50,51]. As a result, a single metazoan miRNA gene family can be predicted to target hundreds of messages in an organism. A combination of in silico prediction and gene ontology annotation of the miRNA targets indicated that many of the target genes are involved in the regulation of anatomical morphogenesis and cell differentiation processes. Gomez-Orte and co-workers (2009) [34] have shown that metamorphosis is regulated by miRNA in the German cockroach. RNAi silenced expression of Dicer-1 in last instar nymphs led to the depletion of the mature miRNAs that impaired the ability of the nymphs to molt into adults. Nymphoids were morphologically similar to those resulting from juvenile hormone (JH) treatment [34]. The reduced accumulation of mature miRNAs in the Vat + aphids that target genes involved in morphogenesis and cell differentiation could be responsible for reduced growth and fecundity in aphids that are nutritionally challenged when forced to survive on resistant plants.
Another interesting feature of the aphid sRNA libraries was the detection of miRNAs of plant origin. It appears that during phloem sap ingestion members of several plant miRNA families (miR156/miR157, miR166, miR168, miR2911 and miR2916) were transferred into the aphid tissues or were present in the contents of the gut as illustrated by the in vitro feeding assays ( Figure 5). Three of these miRNAs (miR156/miR157, miR166 and miR168) have been reported to be present in phloem sap of pumpkin (Cucurbita maxima), apple (Malus domestica), canola (Brassica napus) and lupin (Lupinus sp) [52,53,54,55]. The presence of the remaining plant miRNAs (miR2911 and miR2916) in the phloem sap has not been confirmed; however, their identification in the aphid libraries suggests that these miRNAs are also present in the phloem sap of melon. The perfect match of these sequences to conserved plant miRNAs strongly indicates that they originate from the host plant. Furthermore, the possibility that large transcripts acquired from plant phloem sap by the aphid can be processed into miRNAs seems remote, as this would require the ingested plant pri-miRNA to be efficiently transferred and processed at high numbers in aphid gut cell nuclei. Interestingly, only two plant miRNA families (miR166 and miR168) were detected from Vat + aphids. This could reflect the well-documented observations that aphids on Vat + plants spend less time ingesting phloem sap than aphids feeding on Vat 2 plants [19,56].
In summary, this study identified 81 conserved, 12 aphidspecific and nine novel miRNAs from A. gossypii. The novel miRNAs were validated by qPCR, and it was established that these miRNAs are differentially expressed in the Vat + and Vat 2 aphids. In general, a reduced accumulation of mature miRNAs was observed for Vat + aphids, suggesting their expression is regulated in response to the Vat-mediated resistance mechanism. Targets of these miRNAs were predicted using in silico methods, and it was observed that most of the target genes were involved in morphogenesis and anatomical structure determination, indicating that these miRNAs have regulatory roles in aphid development, growth and fecundity.

Plant and insect growth conditions
The nearly isogenic AR 5 (Vat + , resistant) and PMR 5 (Vat 2 , susceptible) melon lines were grown in controlled conditions of 23uC, 60% relative humidity and a photoperiod of 16:8 h (light:dark) for 4 weeks until they reached the four leaf stage. A. gossypii clonal colony was reared and maintained on susceptible honeydew melon plants in a controlled growth chamber at 21uC, with a photoperiod of 16:8 h (light:dark).

Insect infestations
The third and fourth leaves from four-week-old melon plants, both AR 5 and PMR 5, were used for aphid infestations. Each leaf was exposed to 50 apterous aphids confined in a 2.5 cm diameter clip cage. A mixture of aphids from different growth stages were used to mimic a natural crop infestation. Infestations were  terminated after 48 hours. After this period, all aphids were gently brushed off the leaf tissue, collected in a nuclease free microfuge tube, snap frozen in liquid nitrogen and stored at 280uC. All experiments were biologically independent and at least seven independent replicates were used for each melon line.

Cloning of sRNA and Illumina sequencing
Total RNA was isolated by TRIzolH reagent (Invitrogen) following the manufacturer's instructions for plants with high polysaccharide content, since aphid body is mostly composed of the sugary phloem. The RNA was enriched for low molecular weight (LMW) RNA by precipitation with 25% PEG and 5M NaCl [57]. The LMW RNA quality was evaluated using a RNA600 Nano LabChip kit with the Agilent Bioanalyzer system (Agilent Biotechnologies) and the concentration was measured with a Nanodrop spectrophotometer (Thermo Scientific).
LMW RNA (50 mg) from Vat + and Vat 2 aphids was separated in a 15% denaturing polyacrylamide gel. Size fractions of 20-24 nt were excised and the RNA was extracted from the gel slices with 0.3M NaCl, followed by ethanol precipitation. The RNA was ligated to a 26 bp 59 adapter, size-selected on a polyacrylamide gel and ligated to a 39 adapter. The adapter sequences and PCR primers used in sRNA library preparation are provided in File S8.
First strand cDNA synthesis was performed with the adapter ligated sRNA template using the Superscript II reverse transcriptase (Invitrogen) and a primer complementary to the 39 adapter (S 8). The cDNA was used as a template for a low cycle PCR amplification (initial denaturation of 30 sec at 98uC, 15 cycles of 98uC for 10 sec, 60uC for 30 sec and 72uC for 15 sec, followed by final extension at 72uC for 10 min) to generate sufficient template for deep sequencing. The size fractionated cDNA of ,100 bp was gel eluted and subjected to Illumina GAII analyzer for deep sequencing.
Bioinformatic analysis of the sRNA transcriptome All Illumina sequencing data was initially converted to FASTA format from FASTQ. sRNA sequences were extracted from raw reads by matching them with the first 8 nucleotides of the 39 adapter sequences. Based on the length of the mature miRNA and adapter length, sequences shorter than 18 nt and greater than 30 nt in length were removed. At this stage, the data were screened for redundant sequences. The remaining sequences were queried against ribosomal and transfer RNAs from Flybase (http://flybase.org/) and the matches to the rRNA and tRNA were discarded. Sequences having ten or more counts were assigned as unique sequences. These sequences were then queried against A. gossypii coding RNA regions (EST database) for perfect matches. Those sequences that matched to the sense strand were discarded as they were considered to be degraded mRNA; however those matching to the anti-sense strand were retained and aligned against the miRBase (http://microrna.sanger.ac.uk/ sequences) to identify the conserved miRNA (two mismatches allowed). All bioinformatic analyses were performed using custom written PERL script. Bowtie, which is an ultra fast memory efficient short read aligner [58], was used to match the sequences to A. gossypii EST, M. persicae EST, and A. pisum genome sequence databases. Unaligned sequences were potential candidates for novel A. gossypii miRNAs. Novel miRNAs were identified using the algorithm miRDeep [22]. Since the complete genome sequence of A. gossypii was not available during the processing of the dataset, the EST database was used to find novel miRNAs. The bioinformatic analysis is summarized in File S1.

In vitro feeding assay
A sterilized 25 ml glass beaker of 4.5 cm depth and 2.5 cm diameter was used for the in vitro feeding assay. At the bottom of the beaker a blotting paper disc of 2.5 cm was placed, on top of which ten aphids of same developmental stages where gently placed with a brush. Diet sachet containing [c -32 P]-ATP labeled ds-21 nt in 75 ml of 0.5 M sterile sucrose solution was sandwiched between two parafilm layers and this diet sachet was used to seal the mouth of the beaker. Aphids were allowed to feed on this labeled diet for 4 days. The same system, minus the [c -32 P]-ATP labeled ds-21 nt was used for control assay. After the end of 4 days, the number of live aphids were counted and used for RNA extraction with TRIzolH reagent. The blotting disc at the bottom of the beaker was carefully retrieved and exposed to a phosphorimager screen and viewed in TyphoonTrio (GE Healthcare). Honeydew from these blotting discs was recovered by washing with using 0.1M Tris. The RNA extractions along with the honeydew elutions and a positive marker (labeled ds-21) was separated on a 15% PAGE, and exposed to X-ray film.
Expression profiling of novel A. gossypii miRNAs by stem loop qPCR LMW-enriched RNA was used to study the differential expression pattern of novel miRNA in the Vat + and Vat 2 aphid tissues. The miRNA expression was measured using a two-step process. In the first step, a stem-loop (RT) primer designed according to Chen et al. (2005) [59], was hybridized to the miRNA and reverse transcribed in a pulsed RT reaction [60]. In the second step, the RT reaction product was PCR amplified using a miRNA-specific forward primer and a universal primer (File S9) in real time with SYBR green chemistry using ABI 7500. The A. gossypii elongation factor-1a (eF1a) gene (GenBank Accession EU019874.1) was used as a reference. The relative changes in miRNA expression were quantified using Pfaffl's method (2001) [61]. The data for relative quantities were converted to fold differences by logarithmic transformation to express the data as a normal distribution. The data were represented as averages of seven measurements 6 standard error and the comparisons of the miRNA expression levels between Vat + and Vat 2 aphid were measured using student's t-test (p#0.05).

Analysis of the longer sRNAs
The longer sRNAs were mapped to the transposable elements of A. pisum. The available transposon database of A. pisum was downloaded, which was predicted using the REPET transposon annotation pipeline [62]. Using bowtie short reads mapper (with non-default parameters: -k 1 -p 20 -n 1 -v 2) [57], the longer (at least 25 nt long) sequences from the two separate Vat + and Vat 2 aphid libraries were mapped. The sequence composition of the longer sRNA was analyzed using the DREME program [25] in the MEME motif finding software package (version 4.8.1) [26].

Target prediction analysis
In the absence of a publicly available whole genome annotation for A. gossypii, several databases were used to predict the targets of the sequenced A. gossypii miRNAs. The first database is an extensive collection of A. gossypii ESTs [63], mRNAs, and gene fragments from the NCBI GenBank that were used in the construction of the A. gossypii Unigene build #1 (ftp://ftp.ncbi.nih. gov/repository/UniGene/Aphis_gossypii/Ago.info). From the aphid genome database resource [64] the assembled transcriptome contigs of A. gossypii and M. persicae, and the annotated gene model predictions of A. pisum were obtained. Supplementary file 3 (File S3) shows the number of sequences retrieved from each database. The BEDtools program [65] was used to extract the 39-UTR sequences of A. pisum.
PITA and RNA hybrid programs were used for target prediction of the newly identified miRNA sequences from A. gossypii. As a prelude to target prediction, the PITA program was run using the default settings; a moderately stringent setting for seed region base pairing was used [30]. The moderate stringency setting allows the use of seed lengths ranging from 6-8 bases and up to single mismatch or wobble in the 7-8 nt seeds. The results of the computational prediction produced excessive number of target mRNAs for all the sequencing datasets used. To improve on the efficacy of the target predictions, the non-default parameter settings (-l 7-8 -gu ''7;0,8;0'' -m ''7;0,8;0'') were used. This stringent setting used perfectly matching seed length of 7-8 bases.
Using the recommended miRNA-target prediction energetic score (DDG) cut-off for the PITA program, only targets with DDG scores of 210.0 or less were retained (File S4). The RNA hybrid program was run with non-default parameters: -m 100000 -s 3utr_fly -p 0.01) to obtain a conservative set of microRNA target predictions with high statistical significance (p-value#0.01) [33].
The BLASTX program from NCBI (blast.ncbi.nlm.nih.gov/ Blast.cgi) was used (e-value cut-off used was 1.0E208) to identify A. gossypii gene homologs of A. pisum [66]. Protein homology information was used to extract A. pisum GO terms associated with the A. gossypii sequences. The GO annotation for each dataset was imported in to the BLAST2GO web server [67]. For annotation, the GO Slim GO term mapper module in BLAST2GO was used and the generic GO term classification dataset was selected to define the functional categories of the predicted genes.

Supporting Information
File S1 Bioinformatic analytic scheme of sRNA libraries.