Identification and comparison of key RNA interference machinery from western corn rootworm, fall armyworm, and southern green stink bug

RNA interference (RNAi)-based technology shows great potential for use in agriculture, particularly for management of costly insect pests. In the decade since the insecticidal effects of environmentally-introduced RNA were first reported, this treatment has been applied to several types of insect pests. Through the course of those efforts, it has become apparent that different insects exhibit a range of sensitivity to environmentally-introduced RNAs. The variation in responses across insect is not well-understood, with differences in the underlying RNAi mechanisms being one explanation. This study evaluates eight proteins among three agricultural pests whose responses to environmental RNAi are known to differ: western corn rootworm (Diabrotica virgifera virgifera), fall armyworm (Spodoptera frugiperda), and southern green stink bug (Nezara viridula). These proteins have been identified in various organisms as centrally involved in facilitating the microRNA- and small interfering-RNA-mediated interference responses. Various bioinformatics tools, as well as gene expression profiling, were used to identify and evaluate putative homologues for characteristics that may contribute to the differing responses of these insects, such as the absence of critical functional domains within expressed sequences, the absence of entire gene sequences, or unusually low or undetectable expression of critical genes. Though many similarities were observed, the number of isoforms and expression levels of double-stranded RNA-binding and argonaute proteins varied across insect. Differences among key RNAi machinery genes of these three pests may impact the function of their RNAi pathways, and therefore, their respective responses to exogenous RNAs.


Introduction
, and the authors may not be able to provide materials including third party genetic elements to the requestor because of certain third-party contractual restrictions placed on the author's institution. In such cases, the requester will be required to obtain such materials directly from the third party. The authors and authors' institution do not make any express or implied permission(s) to the requester to make, use, sell, offer for sale, or import third party proprietary materials. Obtaining any such permission(s) will be the sole responsibility of the requestor. siRISC binds cellular ssRNA exhibiting high complementarity along the full length of the guide strand [24]. Once bound, such ssRNA is cleaved by AGO2; the 21-25 nt complementarity typically results in a one-to-one pairing of siRNA and target [24].
It is clear from the study of both model and non-model systems that basic RNAi mechanisms are present in all plants and animals. However, differences across organisms in the protein machinery involved can affect various characteristics of the response, including type of mediating sRNAs, life stages and tissues in which different pathways function, and overall regulatory outcomes [8,10,25]. More specific to the application of RNAi for insect control, response to environmentally-introduced dsRNA (environmental (e)RNAi) is known to vary greatly across species. Many insects from the order Coleoptera tend to show robust activation of their RNAi pathways via oral feeding on transgenic plants expressing dsRNAs, whereas even in a laboratory setting, dsRNA elicits little response in Lepidoptera or Hemiptera (reviewed in [26,27]). A variety of factors have been proposed to contribute to these differences, such as disparity in dsRNA uptake mechanisms and nuclease content of saliva and gut fluid [26]. Much focus has been placed on these initial barriers to treatment with insecticidal dsRNA [28], though less is known about downstream factors which may also play a role.
Understanding differences in insect response to eRNAi is central to the development and proper implementation of RNAi-based crop protection. This need, along with a general lack of research in prominent agricultural pests, led to focus of the current study on the core RNAi machinery of Diabrotica virgifera virgifera (western corn rootworm-WCR), Spodoptera frugiperda (fall armyworm-FAW), and Nezara viridula (southern green stink bug-SGSB). These three representative pests were used to explore another possible source of variability in RNAi efficacy: differences in the presence or absence, modifications to, or expression levels of core RNAi pathway proteins [26,27]. Reports describing predicted protein features and phylogeny of a handful of core RNAi machinery genes in WCR and FAW are available [29][30][31][32]. However, given the complexity revealed through decades-long study of Dme RNAi, more work is needed to understand the differences that might exist between core RNAi components of insects that respond well to control via eRNAi and those that do not.
This study is aimed at exploring differences in the core RNAi machinery of WCR, FAW, and SGSB. Toward that end, identification of core mi-and siRNA components in each insect was performed, followed by prediction and comparison of protein domain structure, phylogeny, and inspection of expression patterns across life stages. Furthermore, a direct comparison of the baseline expression levels across insects was conducted for each core component. While the putative homologues identified in these pests are similar to reference sequences and demonstrate some consistency across insect in expression patterns and levels, detailed examination reveals intriguing differences. The number, features, and expression of the dsRBP and AGO isoforms show variations that may influence basic functioning of the RNAi pathways in these insects.

In silico identification of core RNAi machinery
Putative homologues of all eight core RNAi machinery genes were mined from WCR, FAW, and SGSB complementary DNA (cDNA) datasets through a series of iterative searches beginning with Dme query sequences. Potential candidates were translated in all three coding frames, protein domain structure was predicted and compared with Dme sequences, and those exhibiting apparently complete coding DNA sequences (CDSs) and domain structure were selected. Homologues present in each of the three insect pests under investigation are presented in S1 Table, along with translated protein length and accession number. One homologue per pest was found for Drosha, DCR-1, and DCR-2. Multiple versions of potential homologues were identified for Pasha, LOQS, R2D2, AGO1, and AGO2, which may represent products of alternative transcriptional start sites, alternative splicing, or duplicated genes. Regardless of the sources of these differences (i.e. paralogues versus orthologues), for the purposes of this study all versions of the same protein are referred to as isoforms.

Classification and analysis of putative homologues
Translated putative isoforms for each protein follow Dme nomenclature, which was assigned by conducting a blastp search against a database of corresponding Dme proteins. A total of 9 ribonuclease III domain-containing (RNaseIII), 21 dsRBP, and 12 AGO sequences were found across WCR, FAW, and SGSB. A series of bioinformatics tools was then used to examine putative homologues for similarity to each other, to sequences previously reported in related insect species, and to sequences reported from model organisms. Tribolium castaneum (Tca), Bombyx mori (Bmo), and Acyrthosiphon pisum (Api) sequences were included to represent closelyrelated members of the coleopteran, lepidopteran, and hemipteran insect phylogenetic orders, respectively. Homo sapiens (Has), Cel, and Dme sequences serve as well-characterized reference core RNAi proteins, and represent more distant relations to putative homologues from the pests of interest. Final analyses were conducted by partitioning translated reference sequences and candidate homologues into three groups: RNaseIII proteins, dsRBPs, and AGO proteins. To the extent possible, the same putative isoform per protein was selected and used for domain analysis and the reconstruction of phylogenetic trees shown in Figs 1-3. Results of these analyses show that proteins identified in each insect exhibit phylogenetic behavior and domain structure comparable to each other and to reference sequences. Results of domain analyses conducted for all identified sequences are summarized in S2 Table.

Ribonuclease III domain-containing proteins
Translations of the RNaseIII sequences were scanned against the Pfam database to determine protein domain architecture. Results show that each of the translated pest sequences contain the same number and type of domains as reference sequences (Fig 1A). These sequences display two RNase III domains and-with the exception of Dme and Bmo DCR-1-a C-terminal dsRNA binding motif (DSRM). The DSRMs of the DCR proteins are predicted with less confidence, showing much higher expect (e-)values (approaching 1) than any of the other domains identified in this study. The DCR proteins also contain a DCR dimer motif, an RNA binding domain common to PIWI, AGO, and Zwille proteins (PAZ), and a helicase C domain. The DCR-2 proteins additionally contain the N-terminal DEAD-box helicase domain consistent with the role of Dme DCR-2 translocation along a dsRNA substrate [33]. Under the conditions used in this study for protein domain prediction (see "Methods"), the DEAD-box helicase domain registers as the closely-related bacterial restriction III enzyme domain (ResIII) in the lepidopteran and hemipteran sequences.
Next, parsimony informative sites were identified from an RNaseIII protein multiple sequence alignment (MSA) and used in phylogenetic tree reconstruction. Putative homologues cluster as would be expected for correctly assigned sequences, with Drosha, DCR-1, and DCR-2 from each insect appearing in three distinct protein clades along with all corresponding reference sequences (Fig 1B). The DCR proteins from Hsa and Cel show a domain structure closer to the siRNA-specific DCR-2, but cluster with the miRNA-specific DCR-1. The WCR, FAW, or SGSB sequences also cluster with high frequency (bootstrap values >91% excluding SGSB Drosha) on the same branch as the reference sequence from their respective phylogenetic order. Overall, length of translated amino acid sequence, domain structure, and phylogenetic analysis agree well with previously reported results for select WCR and FAW RNaseIII sequences [30][31][32].
Double-stranded RNA binding proteins. The dsRBP sequences were analyzed for protein domain arrangement. Each identified pest dsRBP contains two to three DSRMs, in agreement with reference sequence scans (Fig 2A). Pasha, the partner protein of Drosha, is the longest dsRBP and contains two C-terminal DSRMs-with the exception of Cel PASH-1. The Hsa and Cel DGCR8 and PASH-1 sequences are predicted to contain a tryptophan-rich (WW) motif possibly responsible for mediating specific protein-protein interactions with Drosha [12], which does not register in any of the insect sequences. The LOQS-PB sequences each contain three DSRMs, consistent with proposed functions of binding dsRNA (DSRM1 and DSRM2) and interaction with DCR-1 (DSRM3) [7]. Insect-specific R2D2 is the shortest dsRBP and contains two N-terminal DSRMs. A final FAW R2D2 sequence could not be confidently selected from among harvested candidates, though its presence has been reported in FAW ovary-derived Sf21 cells [32]. The sequence analyzed here as FAW R2D2 represents that which agreed most consistently with reference sequences through bioinformatic evaluations utilized during the identification process (described under "Methods").
All residues of the dsRBP sequences were used to reconstruct phylogenetic relationships. Generally, the newly identified pest sequences cluster within the same clade and on the same branch as the sequence from their closest related reference organism ( Fig 2B). The WCR and FAW Pasha and LOQS homologues show bootstrap values >98% for clustering with Tca and Bmo sequences. Most R2D2 proteins also cluster within a clade separate from LOQS, though there appears to be lower bootstrap support for the distinct separation of these two dsRBP clades than any other protein groups examined within this study. Each SGSB dsRBP proves an exception by appearing on a branch separate from the available Api sequences, with R2D2 appearing in a separate clade altogether. Predicted domain arrangement and phylogeny agree with previous reports of the FAW dsRBPs [32].
Argonaute proteins. Protein domain analysis of the AGO sequences reveals each WCR, FAW, and SGSB sequence exhibit the same predicted domain structure as reference sequences ( Fig 3A). Generally, all sequences include an N-terminal domain of argonaute (ArgoN), two argonaute linker domains (ArgoL1 and ArgoL2), a PAZ domain, a Mid domain of argonaute (ArgoMid), and a Piwi domain. ArgoL1 was not detected in Cel RDE-1 and ArgoMid domains were not detected in FAW, SGSB, Bmo, and Tca AGO2 or Cel RDE-1. The failure to detect an ArgoMid domain in some AGO2 sequences could be due to the scan conditions used-such as software package and settings-or deviation from the classically recognized amino acid sequence such that this domain is not being recognized by the scanning algorithms. A true absence would be very unusual, as ArgoMid has been observed in crystal structures to form extensive critical interactions with the Piwi domain, and also to contain residues essential for recognition and binding of the guide RNA 5' terminal phosphate [34][35][36].
All residues of the AGO sequences were used in phylogenetic tree reconstruction. The WCR, FAW, and SGSB AGO1 and AGO2 sequences cluster appropriately into each of two clades, and most also appear on the same branch as the relevant reference sequence (Fig 3B). Bootstrap testing gives lower support values for co-clustering of the SGSB and Api sequences. Despite being the only human AGO displaying endonucleolytic activity [37], Hsa AGO2 helicase C (PF00271), a PAZ (PF02170), and either a ResIII (PF04851) or DEAD-box helicase (PF00270) domain. E-values for domains predicted in the WCR, FAW, and SGSB proteins range from 4.0×10 −10 to 1.1×10 −36 , with the exception of the DCR-1 and DCR-2 C-terminal DSRMs. (B) Maximum likelihood phylogenetic tree topology of translated RNaseIII protein-coding sequences (1000 bootstrap replications). Black symbols above each entry indicate phylogenetic order as follows: circles (•) for Diptera, squares (■) for Lepidoptera, diamonds (◆) for Coleoptera, teardrops (S) for Hemiptera, xrhombus (❖) for Primate, and sunburst (✸) for Rhabditida.  clusters with insect AGO1, in agreement with previous observations [38,39]. The endonucleolytic activity of Dme AGO1 is not involved in the canonical role this protein plays regarding silencing of target RNAs [40]. Though domain structure and phylogeny generally agree with previous reports of select WCR and FAW AGO proteins, the WCR AGO2 sequences reported here are longer [30][31][32]. These length differences are due to a combination of missing sequence and an assembly error that caused up to a 365 residue truncation of the N-terminus in the previously reported sequence [30], likely because of the highly repetitive nature of this region [41,42]. Additional internal sequence information allowed the error to be identified and manually corrected, resulting in the true full-length WCR AGO2 sequences reported here.

Expression patterns of core machinery across insect development
Following identification of core RNAi machinery, expression of these genes was evaluated across the life cycles of WCR, FAW, and SGSB through the use of RNA sequencing (RNA--Seq). As used in this study, the term 'expression' refers to normalized transcript abundance levels derived from RNA-Seq experiments. To serve as a point of comparison, expression values of the Dme reference machinery were extracted from results of the Dme developmental transcriptomes generated as part of the modENCODE project and are also displayed [43,44]. Details of the 14 (WCR), 10 (FAW), 9 (SGSB), and 30 (Dme) life cycle points from which expression data were collected are described in S3 Table. Core machinery of the miRNA pathway for all four insects has been separated into the microprocessor complex of drosha and pasha (Fig 4), and the downstream genes dcr-1, loqs, and ago1 ( Fig 5). Core machinery of the siRNA pathway for all four insects are shown together (Fig 6).
Expression patterns of the miRNA machinery show similar trends within insects. The drosha and pasha transcripts display comparable patterns-these transcripts are most abundant early in the egg but are at lower levels in remaining life stages (Fig 4). Normalized expression values for these transcripts are on average the lowest observed for any of the core machinery, though drosha has higher estimated abundance than pasha. Expression patterns of the partner proteins dcr-1 and loqs are also generally similar within each insect, though comparison of patterns across insects are more difficult to make ( Fig 5). Highest dcr-1 expression occurs in early to mid-age eggs in all species. While the stage with highest loqs levels varies across insects, this transcript reaches the highest value of any core miRNA gene within WCR, FAW and Dme. Expression patterns of ago1 are similar across insects, showing highest levels in early egg followed by lower expression. It also expresses highest of any miRNA transcript in SGSB.
Expression patterns of siRNA machinery transcripts are also consistent within insect ( Fig  6). Across insect, these transcripts are variably expressed through life stages rather than peaking early in development. Normalized WCR and Dme values are high in egg, dip in late larval and pupal or adult stages, and increase in pre-pupal, early pupal, or actively reproducing adults-especially pregnant females. The FAW transcripts spike in early larval instars and adults. The SGSB transcripts are increasingly expressed from egg to adult, with pregnant females showing the highest expression of dcr-2 and ago2. In WCR, FAW, and Dme, dcr-2 exhibits the lowest expression of the core siRNA transcripts, while in SGSB r2d2-the partner protein of dcr-2 in Dmel-is lowest. The most consistently abundant of the siRNA machinery transcripts across the life cycles of WCR, FAW, and Dme, is ago2, but in SGSB dcr-2 is SGSB range from 1.5×10 −3 to 9.2×10 −15 . (B) Maximum likelihood phylogenetic tree topology of translated dsRBP protein-coding sequences (1000 bootstrap replications). Black symbols above each entry indicate phylogenetic order as follows: circles (•) for Diptera, squares (■) for Lepidoptera, diamonds (◆) for Coleoptera, teardrops (S) for Hemiptera, xrhombus (❖) for Primate, and sunburst (✸) for Rhabditida.
https://doi.org/10.1371/journal.pone.0203160.g002 generally the most abundant. Expression values for the FAW r2d2 transcript are not included due to uncertainty in choosing a sequence from among available candidates, though expression of the top candidate sequence was confirmed in whole FAW using reverse transcription polymerase chain reaction (RT-PCR) (S1 Fig). Expression of this transcript was observed in early and late egg, third instar, pupal, and adult female stages-bands were most intense at early egg and third instar; stages that match expression peaks of FAW dcr-2 and ago2.

Comparison of core machinery expression levels between insect
Expression of core RNAi machinery was directly compared in WCR, FAW, and SGSB to determine baseline levels (Fig 7). Expression of the core RNAi machinery genes in each insect was likelihood phylogenetic tree topology of translated AGO protein-coding sequences (1000 bootstrap replications). Black symbols above each entry indicate phylogenetic order as follows: circles (•) for Diptera, squares (■) for Lepidoptera, diamonds (◆) for Coleoptera, teardrops (S) for Hemiptera, xrhombus (❖) for Primate, and sunburst (✸) for Rhabditida.
https://doi.org/10.1371/journal.pone.0203160.g003 with drosha marked by black squares and pasha with grey circles. Normalized count for WCR, FAW, and SGSB transcripts was estimated using RSEM and modeled using DESeq2. The median value across sequencing samples (n = 2 to 4 for pest, n = 30 for Dme) is shown, with error bars representing the median absolute deviation (MAD). Dme data were obtained from the modENCODE project [33,34]. Normalized count (in reads per kilobase million) for Dme transcripts was generated by adjusting for read depth on a per million scale and length of each target gene in kilobases. Expression of drosha is scaled on the left axis in all graphs, and pasha on the right for WCR and FAW; axes colors also reflect gene target scaling. pasha-RA is shown for WCR and FAW, and-RAa for SGSB. Specific Dme isoforms are unknown.
https://doi.org/10.1371/journal.pone.0203160.g004 measured using semi-quantitative RT-PCR with cDNA prepared from samples of the same starting mass and with the same amount of isolated RNA. The life stage chosen for this analysis was midpoint of the first post-hatch stage during which each insect would begin to feed on host crops: first instar for WCR and FAW, second instar for SGSB. Expression of dcr-1, dcr-2, and ago2 is similar between insects at this stage, as is r2d2 in WCR and SGSB (quantitative expression of the top FAW r2d2 candidate was not measured). A moderately lowered expression of drosha and pasha in SGSB and ago1 in WCR is observed when compared with the other two insects. The greatest difference is seen in levels of loqs transcript, which are highly elevated in WCR compared with those of FAW or SGSB.

Discussion
Interest in the use of RNAi-based technology for insect control has increased the necessity of understanding RNAi pathways in non-model insects, especially those of damaging agricultural pests [2]. Examples of such pests include WCR, FAW, and SGSB. These pests also represent phylogenetic orders whose reactions to eRNAi greatly differ. While many coleopterans show Study of RNAi gene sequences in WCR, FAW, and SGSB robust knockdown of target genes in response to eRNAi, lepidopterans and hemipterans show variable or weak knockdown [26,27]. Reflecting the response of other beetles, successful control of WCR via knockdown of important genes by transgenic crops expressing dsRNAs was reported ten years ago [6]. In contrast, successful expression knockdown of a FAW gene target via laboratory dsRNA feeding has only been reported twice [45,46], and has not been reported for SGSB. Differences between characteristics of the core RNAi machinery have not been extensively explored as a potential contributor to the observed responses in these species. Presented here is the first identification and preliminary evaluation of all eight of the core mi-and siRNA pathway genes and their potential isoforms in WCR, FAW, and SGSB. WCR drosha, dcr-1, dcr-2, ago1, ago2, and FAW drosha were also confirmed to agree with previously reported partial sequence [29][30][31][32], and the FAW sequences searched for hits against the recently published genome (S1 File) [47]. All sequences were examined and compared with one another and with reference sequences to determine whether differences exist in presence, number of isoforms, protein domain structure, expression pattern, or baseline expression levels. It was hypothesized that variation in these natural characteristics could lead to variation in efficacy of insecticidal RNAs. Study of RNAi gene sequences in WCR, FAW, and SGSB At least one sequence homologous to each core component of both the Dme mi-and siRNA pathways was identified in WCR, FAW, and SGSB (S1 Table). As defined by Dme, these pathways are intact and would be expected to function in generally the same manner if evaluated solely by gene presence. Beyond basic presence of pathway machinery in an insect's genome, the number of genomic copies has been suggested to confer graded sensitivity to exogenous dsRNAs [48]. The sensitivity of Tca appears to be increased beyond the response of several other studied coleopterans, and this insect reportedly has two genomic copies specifically of ago2 [48]. Homologues of Dme ago2 have been reported in several lepidopteran and hemipteran pest species [49][50][51], though a FAW ago2 homologue was not included in a previous list of RNAi pathway genes detected in Sf21 cells [32]. Interestingly, AGO1 has also been reported to contribute to the response of a coleopteran cell line to exogenous dsRNA [52]. Although exact genomic copy number was not determined in this study, complete absence of any one core RNAi component-including both AGOs-cannot explain the difference in eRNAi response observed between WCR, FAW, and SGSB.
It is well-established in Dme that different isoforms exist for core RNAi machinery, and that they impact functioning of the RNAi pathways as discussed below. In an effort to understand differences that may exist between the RNAi pathways in WCR, FAW, and SGSB, it was important to include isoform identification in the current study-though the isoform numbers reported here may not be complete due to limitations of the cDNA databases used. Study of RNAi gene sequences in WCR, FAW, and SGSB Additionally, isoform designations were assigned based on in silico predictions only; they do not guarantee any discrete functionalities-or lack thereof-that have been demonstrated for these proteins in Dme and other organisms. Exactly one homologue for each of the RNaseIII proteins was identified in each pest, a result in line with reports from Dme and other relevant insects [32, 48-51, 53, 54]. This does not exclude the possibility that different isoforms of these or other core RNAi machinery genes may exist under different conditions, a state which has been observed for mammalian Drosha and DCR [55,56]. Several different isoforms for the dsRBPs and AGOs were discovered for WCR, FAW, and SGSB, also consistent with the Dme RNAi pathway machinery.
Different isoforms of the dsRBP Pasha may be required for localization in either the nucleus or cytoplasm to facilitate distinct functions of Drosha. Most Drosha functions have been found to depend on Pasha [11,12,57], and this RNaseIII protein-typically thought to function in the nucleus-has been implicated in the cytoplasmic antiviral response of Dme cells [58]. It then follows that Pasha should also likely be present in the cytoplasm under those circumstances. Two versions of Pasha have been reported for Dme, which differ from one another at the N-terminus: PA/C and PB (S2 Fig). A nuclear localization signal (NLS) is predicted in the first 50 amino acids for Dme Pasha-PA/PC-a region that is absent in the Pasha-PB isoform. This may suggest a nuclear function for one isoform and a cytoplasmic function for the other. Both Pasha isoforms were identified in FAW, with PA containing a predicted Nterminal NLS that is missing in PB (S2 Fig). Only one Pasha isoform was classified in WCR and SGSB. The WCR Pasha was classified by homology as a PB isoform, but is predicted to contain an NLS and was therefore designated a PA isoform (S2 Fig). Three variants of Pasha-PA were found in SGSB (designated PAa, PAb, and PAc); they deviate at the amino acid level in their DSRMs, but non-N-terminal NLSs are predicted in all three (S2 Fig). Isoforms of the dsRBP LOQS have been demonstrated to play distinct functional roles in Dme RNAi pathways. Both the PB and PA isoforms contain three DSRMs, but differ in both their interaction with DCR-1 and expression by sex. The PB isoform exhibits high binding affinity for DCR-1, is the primary dsRBP facilitating dicing of many pre-miRNAs, and shows higher expression than PA in female flies [17]. The PA isoform is lower affinity, can rescue the phenotype of PB-deficient flies but in some cases produces miRNAs of different length and seed sequence, and shows higher expression than PB in male flies [17]. Sequences classified as LOQS-PB and -PA isoforms were found in WCR, FAW, and SGSB. In all three cases, the PA isoform contains a shortened linker region between the second and third DSRMs, consistent with Dme LOQS-PA (S2 Table, S3A Fig). In flies this region encodes a motif essential for forming a hydrophobic interface with DCR-1, and its absence in PA leads to lowered affinity and modified miRNA processing [17,59].
The Dme LOQS-PC and -PD isoforms lack the C-terminal third DSRM, and while PC is thought to be an aberrant transcript the protein for which has yet to be detected in any fly stage or tissue, PD is known to interact with DCR-2 [60][61][62]. One significant difference between FAW and the other two pests is the apparent presence of an isoform that is missing the third DSRM, similar to LOQS-PD (S2 Table, S3B Fig). In the Dme PD isoform, the third DSRM is replaced with a region responsible for DCR-2 interaction [60,61], and the putative FAW LOQS-PD isoform also shows an exchange of the third DSRM for novel sequence. The presence of LOQS-PD has not been confirmed outside of Drosophilidae, and has been proposed to be an adaptation specific to that family [63]. The LOQS-PA isoform appears to fill the role of -PD in Aedes aegypti and potentially in other insects by participating in the siRNA pathway with DCR-2, and additionally exerting a regulatory effect on miRNA production [63]. It is unclear whether the putative LOQS-PD transcript in FAW represents a bona fide isoform. In addition to LOQS-PD, the dsRBP R2D2 also interacts with and modifies the activity of DCR-2 in Dme [21,33]. One R2D2 homologue was identified in WCR, but more than one version was detected in FAW and SGSB.
The importance of AGOs to RNAi has been well characterized in model systems (reviewed in [7,8]), and all three pests in this report appear to have several different isoforms of both AGO1 and AGO2. The isoforms of Dme AGO1 and AGO2 differ at the N-terminus, presumably due to alternative transcriptional start sites. This appears to be consistent for putative isoforms identified in WCR, FAW, and SGSB (S4A and S4B Fig). AGO2 isoforms within these species show a high number of N-terminal amino acid differences. The N-terminus of Dme AGO2-PB and -PC exhibits a long, unstructured, glutamine-rich repeat region which is absent in the PE isoform. This feature is common across many arthropod AGO2 sequences [41], and indeed it appears in the AGO2 sequences identified for WCR and SGSB (S4B and S4C Fig). Previous reports of WCR AGO2 did not include this repetitive N-terminus, likely due to a combination of missing sequence and assembly errors [29][30][31]. The FAW AGO2 sequences identified in this study are smaller than those of WCR and SGSB, and assuming no missing sequence, their N-termini do not contain a high proportion of glutamine residues (S4 Table). They instead contain a higher proportion of lysine and glutamic acid residues and were more homologous to the Dme AGO2-PE isoform, which completely lacks the glutamine repeat region (S4C Fig, S4 Table). No reliable cDNAs equivalent to AGO2-PE were identified for WCR or SGSB. Although fitting with the known variability of this region even within members of the same genus and species [41,42], the importance of such differences is not clear. It has been shown that this region interacts with AGO1 early in Dme development [64]. Another proposed function is direct interaction with viruses, which could drive its reported rapid evolution [41,42]. While interesting from the perspective of development and possible adaptation to viral evolution, it is unknown whether these differences-or differences in AGO1 isoforms-would affect insect response to eRNAi.
A source of variation beyond the presence, number of isoforms, and protein domain structure of the core RNAi machinery across WCR, FAW, and SGSB could be their expression levels in each insect. It is possible that differences in expression may promote contrasting responses, even under circumstances where the same proteins exist across species and serve identical functional roles. Examination of expression patterns of the core RNAi machinery across life stages of WCR, FAW, SGSB, and Dme reveals surprisingly similar trends (Figs 4-6). Within and across insects, most transcripts whose protein products partner together-and those that cooperate in the same pathway-show similar patterns of expression at the same stage or within a one-stage delay. In several cases, that pattern roughly propagates across species. It is important to note that changes in expression of siRNA factors have been observed upon viral infection and eRNAi in other insects [65][66][67][68], and baseline expression across the four insects shown here would not reflect differences in changes occurring in response to various stressors such as ingestion of insecticidal RNA.
In addition to expression patterns across life stage, direct comparison of transcript levels across a field-relevant WCR, FAW, and SGSB life stage revealed one major difference between WCR, an insect with robust response to exogenous dsRNAs, and FAW and SGSB that do not: an increased loqs expression (Fig 7). Variations in the roles of LOQS may exist in these insects, as LOQS isoforms are known to perform different functions in both the mi-and siRNA pathways of other insects. Expression data for FAW r2d2 were not included, but expression of this dsRBP does not differ between WCR and SGSB. Poor expression of r2d2 has previously been suggested as a potential explanation of the insensitivity of a lepidopteran ovarian cell line to dsRNA [53]. Furthermore, a previous direct comparison of the expression of several core RNAi components in immortalized coleopteran pupal and lepidopteran ovarian cell lines showed universally lower expression in the lepidopteran cells, which was proposed to partially explain the observed discrepancy in dsRNA sensitivity [52]. Results from the current study indicate that expression levels of RNAi genes may not be a consistent source of disparity in whole insects, though r2d2 does show lower expression relative to the other core machinery genes. It is possible that induction of r2d2 or specific loqs isoforms may occur under conditions of dsRNA challenge; these types of responses have been observed for dcr-2 and ago2 in lepidopteran and hemipteran insects, but those studies did not include evaluation of r2d2 [66,69,70]. It is also possible that expression level has fundamentally different effects in each insect that would be undetectable from these data. For example, differences in correlation with translation or intrinsic activities of each insects' proteins would not be apparent. Expression of RNAi proteins specifically in gut tissues may not be comparable to evaluation using whole organisms; however, this seems improbable considering oral ingestion is a primary route of insect exposure to entomopathogenic viruses for other insects [71]. The expression pattern, transcript, and protein abundance of LOQS isoforms and R2D2 in WCR, FAW, and SGSB must be further evaluated under both baseline conditions as well as under dsRNA challenge.
Recent research suggests nuclease content and dsRNA uptake mechanisms are important factors in determining the effectiveness of eRNAi across different insects [48,[72][73][74][75][76], but limited investigation has occurred on the role the core RNAi machinery may also play. Several differences between the core RNAi machinery of WCR, FAW, and SGSB were identified in the present study. Although relevance of these differences to eRNAi is unknown, based on the information presented here they cannot be ruled out as potential contributors to the differing responses of these insects. Purified proteins for in vitro experimentation and whole organisms under conditions of viral or insecticidal dsRNA challenge would assist in parsing the mechanisms and interactions of the core RNAi machinery in these three pests. The information provided here may serve as a basis for such future work.

Homologue identification
Putative homologues for core RNAi machinery from WCR, FAW, and SGSB were retrieved by locally querying internal cDNA databases with Dme CDSs and the tblastx algorithm with an evalue cutoff of 1×10 −10 . These internal databases had been previously assembled using Trinity (v. 2.0.6), IDBA-Tran (v. 1.1.1), Velvet-Oases (v. 1.2.10-0.2.08), and/or SOAPdenovo-Trans (v. 1.03) [77][78][79][80]. Resulting hits were translated to identify likely CDSs, and those showing adequate length were evaluated by local HMMER3.0 scans in the DoMosaics software package (v. 0.92) with Pfam 31.0 HMM libraries [81,82]. Translated sequences showing appropriate Pfam domain structure were manually corrected if misassembly was observed (i.e. appearance of two expected domains in two different reading frames with a long region of sequence overlap), and a variety of protein properties such as molecular weight, extinction coefficient, and isoelectric point were analyzed using the pepstats function of EMBOSS Explorer (v. 2.2.0) and Vector NTI Advance (v. 10.3.1) [83,84]. Final candidates were chosen based on agreement with Dme reference sequence protein domain structure and pepstats-estimated properties. If multiple candidate sequences were found containing the same number but a handful of changed non-consecutive amino acids scattered along the length of the peptide, this was attributed to population variation as the available cDNA databases did not always contain inbred lines. In these cases, only one match was selected. Final putative protein isoforms were classified by using the blastp algorithm (v. 2.2.13) with several scoring matrices (PAM30, PAM70, BLOSUM62) specifically against all reported isoforms of the corresponding Dme protein.
Sequences were matched to a Dme isoform based on highest bit score and lowest e-value. For sequences with very close or identical scoring results across more than one Dme isoform, discrepancies in peptide length and distinguishing features of MSA were used to differentiate between isoforms. If unique features were unavailable, sequences were classified alphabetically as sub-designations of the parent isoform.

Final evaluation of core RNAi machinery sequences
Protein domain analysis was conducted using DoMosaics and InterProScan (v. 4.8 with Inter-Pro database 42) with a scan cutoff e-value of 10 [85]. In some cases, multiple domain predictions overlie the same region and domains showing the lowest e-value were selected for display. NLS sequence was predicted using cNLS Mapper with default settings [86]. Multiple sequence alignment was performed using the MUSCLE algorithm in the MEGA7 software package (v. 7.0.21) [87], with default gap penalty settings, maximum iterations set to 10, and clustering method to UPGMB with minimum lambda of 24. Prior to reconstruction of phylogenetic trees, 56 different amino acid substitution models were tested in MEGA7 using a maximum likelihood fit to identify that which gave the lowest Bayesian Information Criterion score for each protein dataset. Trees were then reconstructed in MEGA7 using the maximum likelihood statistical method with Nearest-Neighbor-Interchange (NNI) heuristic and the WAG+G +F, LG+G+I, and LG+G+F substitution models for the RNaseIII, dsRBP, and AGO datasets, respectively [88,89]. Gaps were included as part of the datasets analyzed, and uncertainty in each tree was estimated using 1000 replications of the bootstrap test.

RNA sample preparation
Between 10 and~2500 insects were sourced from colonies maintained within an internal insectary (DuPont Pioneer, Johnston, IA) at the approximate midpoint of each life stage unless otherwise described (S3 Table). Total RNA was isolated from whole flash-frozen WCR of each of 14 life stages by first homogenizing in Buffer RLT with 0.01% PEG using the RNeasy Mini Kit (Qiagen N.V., Hilden, Germany) following manufacturer's instructions. Directly following column elution, isolated RNAs were DNase-treated using the Ambion TURBO DNA-free Kit and associated protocol (Thermo Fisher Scientific, Inc. Waltham, MA). Purified WCR RNAs were checked for quality and quantity using an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc. Santa Clara, CA) with 2100 Expert software (v. B.02.08.SI648). RNAs larger than 200 nts were isolated from whole live FAW and SGSB of each of ten and nine life stages, respectively, using the Ambion miRVana miRNA isolation kit (Thermo Fisher Scientific, Inc.). Directly following column elution, isolated RNAs were DNase-treated for 90 minutes using the RNase-free DNase kit (Qiagen N.V.) and re-purified using the Isolate II RNA Micro Kit (Bioline, London, England), both per manufacturer's instructions. Purified RNAs were checked for quality and quantity using a Fragment Analyzer (Advanced Analytical Technologies, Inc. Ankeny, IA) with PROSize 2.0 software (v. 1.3.1.1), and then each FAW and SGSB sample was spiked with diluted Ambion ERCC RNA Spike-In Mix 1 (ThermoFisher Scientific, Inc.) at a ratio of 2 μL to 1 μg RNA.

Next-generation sequencing
Sequencing libraries from purified RNAs were prepared using the TruSeq mRNA-Seq kit with associated protocol (Illumina, Inc., San Diego, CA). Briefly, mRNAs were isolated via attachment to oligo(dT) beads, chemically fragmented to a mean size of 150 nt, and reverse transcribed into cDNA via random hexamer priming. Resulting double-stranded cDNA fragments were end-repaired to create blunt-end fragments, 3' adenine-tailed, ligated with indexed Tru-Seq adapters (Illumina, Inc.), and PCR-amplified using TruSeq primers (Illumina, Inc.). Purified PCR-amplified libraries were checked for quality and quantity on a Bioanalyzer DNA 7500 chip (Agilent Technologies, Inc.) with 2100 Expert software before normalization and sample pooling. Sample pools of 10 nM were clustered and sequenced on the HiSeq 2000 (WCR) or 2500 (WCR, FAW, SGSB) system with TruSeq Sequencing By Synthesis Rapid v3 (WCR) or v4 (WCR, FAW, SGSB) chemistry (Illumina, Inc.), as per manufacturer's instructions. Samples were sequenced single-read, fifty cycles per read, to a minimum depth of five million reads per sample and a target depth of ten million reads per sample.

RNA-Seq data normalization
Raw sequencing reads were trimmed to remove bases with quality scores less than 13 and sequence tags less than 24 base pairs (bp), after which samples were deconvoluted based on sequenced index identifier. Filtered reads were aligned to transcriptome assembly references using Bowtie 2 (v. 2.2.2), and gene fragment counts were estimated using RSEM (v. 1.3.0) with default settings [90,91]. Surrogate variable analysis (svaseq) was supervised on selected measurable ERCC sequences as controls to remove batch effects [92,93], using the following model: where g ij is the expression for gene i in sample j, and b i0 , d i u j , and e ij , represent terms for baseline expression, unknown artifact, and measurement error, respectively, as previously described. DESeq2 (v. 1.10.0) was then used with the same ERCC sequences as sample scaling controls (FAW and SGSB) to model gene expression, with independent filtering to optimize power at the 95% confidence level, and with a variance stabilizing transformation to correct for over-dispersion [94], using the following model: Where K ij is the observed count for gene i in sample j following a Negative Binomial (NB) distribution, s j is the sample scale factor according to ERCC controls, μ ij , α i , and q ij are all parameters fit to the data, x j. is life stage, and β i contains the log 2 fold changes for gene i. These modelcorrected count values were back-transformed and used to generate expression pattern graphs by calculating the median and median absolute deviation for each sequence per sample type. Isoforms shown in Figs 4-6 are those most abundantly expressed in the samples and/or the one for which an available sequence appeared within the reference used for alignment. Reference sequences and raw sequencing reads that aligned to them for all relevant transcripts are available in S2 File. Pre-and post-normalized count data are available in S3 File.

Semi-quantitative RT-PCR
Pools of RNA from whole insects of each of nine life stages per insect were prepared using the second procedure described above, and concentrations were determined using a NanoDrop 8000 UV-Vis Spectrophotometer (Thermo Fisher Scientific, Inc.) with software (v. 2.3.2). Reverse transcription was carried out using the SuperScript First-Strand Synthesis System for RT-PCR (Thermo Fisher Scientific, Inc.) by loading 125 ng RNA per reaction and following manufacturer-provided instructions for a combination of random hexamer and oligo-dT priming. For detection of FAW r2d2, one reaction per life stage was prepared with 1 μL of undiluted cDNA and primers amplifying a 283 bp transcript region. For detection of all other genes, three reactions per life stage per insect were prepared with 1 μL of a 1:10 cDNA dilution and primers amplifying 300 bp of each gene. PCR reactions were conducted using Platinum PCR SuperMix High Fidelity (FAW r2d2, Thermo Fisher Scientific, Inc.) or Phusion High-Fidelity PCR Master Mix (all others, Thermo Fisher Scientific, Inc.), according to manufacturer instructions. No template and no reverse-transcriptase controls were also prepared for each primer pair. Thermal cycling proceeded for 40 (FAW r2d2) or a target of 31 (all others) cycles in a C1000 Touch instrument (Bio-Rad Laboratories, Inc., Hercules, CA), after which the entirety of each PCR reaction was loaded onto 1.2% agarose gels and electrophoresed at 100 volts for 90 minutes. Size was indicated through use of the ZipRuler Express DNA Ladder 1 (Thermo Fisher Scientific, Inc.), and a four-point standard curve of pure 300 bp DNA was also included on each quantifying gel. Gels were post-stained with SYBR Safe DNA gel stain (Thermo Fisher Scientific, Inc.) and imaged using a FugiFilm Imager LAS-4000 and Image-Quant LAS-4000 software (v. 1.1, General Electric Corp., Boston, MA). Densitometry was performed using Carestream Molecular Imaging software (v. 5.07.23, Bruker Corp., Billerica, MA). Values for core machinery genes were assessed both with and without normalization using several reference genes confirmed by the same semi-quantitative RT-PCR technique to express at a constant level across life stages and insects. Normalization did not change the expression pattern of core machinery genes, and so directly measured values are presented. Primers and thermal cycling conditions are outlined in S5 Table. Densitometric values calculated for the core RNAi machinery are shown in S6 Table. Supporting information S1 Table. Putative isoforms of the WCR, FAW, and SGSB core RNAi machinery. Accession numbers for the coding sequences and parameters of translated peptide sequences for each insect are shown. The translated sequences used for additional in silico analysis and for which expression data are displayed are marked by asterisks ( Ã ); the specific Dme isoform associated with expression data is unknown. (DOCX) S2 Table. Protein domains predicted in isoforms of the WCR, FAW, and SGSB core RNAi machinery. Domains were predicted as described under "Methods". The translated sequences used for additional in silico analysis and for which expression data are displayed are marked by asterisks ( Ã ). (DOCX) S3 Table. Description of life cycle stages for insect expression data. The points displayed in expression graphs (Figs 4-6) correspond to each of the stage numbers in this table for each insect. Graphed points begin at stage number 1 on the left and end at the last collected stage on the right. Additional details are included above, such as insect colony used, age within each stage, insect diet, and replicate number included in expression data. Information pertaining to the Dme expression data were taken from the following sources prediction for the Pasha-PA and -PB isoforms for Dme, WCR, FAW, and SGSB were performed using ClustalW with default MEGA7 parameters and cNLS Mapper, respectively. Alignments were performed separately by insect to more clearly depict intraspecific sequence differences, which are not easy to visualize in an aggregated alignment. Alignment text colors represent biochemical properties of the different amino acids, and include the following: yellow (A, M, F, I, V, L), olive (C), green (N, Q, S, T, W), aqua (D, E), blue (P), red (R, K), fuchsia (G), teal (H), and lime (Y). Asterisks ( Ã ) above the alignment indicate identical residues, and alignment site numbers are shown at the beginning and end of each block. The highest-scoring NLS is indicated above the relevant residues and includes both the type and score. As described within [83], increasing NLS scores represent higher likelihood of nuclear versus cytoplasmic localization: 2! exclusively cytoplasmic, 3-5 both nuclear and cytoplasmic, 7-8 preferentially nuclear, 9 exclusively nuclear. (TIF)

S3 Fig. Features of insect LOQS isoforms.
Alignments were performed using ClustalW with default MEGA7 parameters. Analyses were separated by insect to more clearly depict intraspecific sequence differences, which are not easy to visualize in an aggregated alignment. Alignment text colors represent biochemical properties of the different amino acids, and include the following: yellow (A, M, F, I, V, L), olive (C), green (N, Q, S, T, W), aqua (D, E), blue (P), red (R, K), fuchsia (G), teal (H), and lime (Y). Asterisks ( Ã ) above the alignment indicate identical residues, and alignment site numbers are shown at the beginning and end of each block. A) Alignment of Dme, WCR, FAW, and SGSB LOQS-PB and -PA isoforms. The amphipathic helix responsible for higher DCR-1 binding affinity exhibited by the Dme PB isoform is indicated by arrows above the Dme sequences [56]. Red arrows point to residues for which mutations cause detrimental effects to DCR-1 binding [56]. B) Alignment showing the C-termini of the Dme and FAW LOQS isoforms. (TIF)

S4 Fig. Features of insect AGO isoforms.
Alignments were performed using ClustalW with default MEGA7 parameters. Analyses for A) and B) were separated by insect to more clearly depict intraspecific sequence differences, which are not easy to visualize in an aggregated alignment. Alignment text colors represent biochemical properties of the different amino acids, and include the following: yellow (A, M, F, I, V, L), olive (C), green (N, Q, S, T, W), aqua (D, E), blue (P), red (R, K), fuchsia (G), teal (H), and lime (Y). Asterisks ( Ã ) above the alignment indicate identical residues, and alignment site numbers are shown at the beginning and end of each block. A) Differences at the C-termini of Dme and WCR AGO1 isoforms. B) Differences at the C-termini of Dme, WCR, FAW, and SGSB AGO2 isoforms. C) Alignment of AGO2-PB sequences from Dme, WCR, and SGSB, and AGO2-PEa of FAW.