Argonaute-CLIP delineates versatile, functional RNAi networks in Aedes aegypti, a major vector of human viruses

Summary

Argonaute (AGO) proteins bind small RNAs to silence complementary RNA transcripts and are central to RNA interference (RNAi). RNAi is critical for regulation of gene expression and antiviral defense in Aedes aegypti mosquitoes, which transmit Zika, chikungunya, dengue, and yellow fever viruses. In mosquitoes, AGO1 facilitates miRNA interactions while AGO2 mediates siRNA interactions. We applied AGO-crosslinking immunoprecipitation (AGO-CLIP) for both AGO1 and AGO2 and developed a universal software package for CLIP analysis (CLIPflexR), identifying 230 small RNAs and 5,447 small RNA targets that comprise a comprehensive RNAi network map in mosquitoes. RNAi network maps predicted expression levels of small RNA targets in specific tissues. Additionally, this resource identified unexpected, context-dependent AGO2 target preferences, including endogenous viral elements and 3′UTRs. Finally, in contrast to current thinking, mosquito AGO2 repressed imperfect targets. These findings expand our understanding of small RNA networks and have broad implications for the study of antiviral RNAi.

eTOC Blurb

In the arbovirus vector Ae. aegypti, RNAi regulates gene expression and defends against viral infection. Rozen-Gagnon et al. developed CLIP and a software package, CLIPflexR, to map transcriptome-wide RNAi networks in mosquitoes. Network maps revealed that AGO2, the effector of antiviral RNAi, has flexible target preferences and represses imperfect targets.

Graphical Abstract

Introduction

RNA interference (RNAi) is an essential biological process that regulates gene expression and silences target RNAs. RNAi includes both microRNA (miRNA) and short interfering RNA (siRNA) pathways (Bartel, 2004; Hutvágner and Zamore, 2002; Meister, 2013). In both pathways, double-stranded RNAs (dsRNAs) are processed into small, single-stranded RNAs, which are then loaded onto Argonaute (AGO) subfamily proteins to form RNA-induced silencing complexes (RISCs). RISCs then target cognate transcripts via sequence complementarity, decreasing transcript levels by cleavage, mRNA decay and/or translational inhibition (Bartel, 2018).

AGO-crosslinking and immunoprecipitation (AGO-CLIP) paired with high-throughput sequencing has allowed systemic mapping of small RNA interactions in model organisms (Chi et al., 2009; Hafner et al., 2010; Helwak et al., 2013; Taliaferro et al., 2013; Wessels et al., 2019). Despite interest in using AGO-CLIP to examine RNAi networks in important non-model organisms, lack of suitable reagents and computational pipelines remain obstacles (Fu et al., 2017; Zhang et al., 2017). The Aedes (Ae.) aegypti mosquito transmits human pathogens including dengue, chikungunya, Zika, and yellow fever viruses and significantly contributes to human disease burden (Farazi et al., 2008; Olmo et al., 2018). Unlike in mammals, where four AGOs function redundantly in the miRNA pathway (Liu et al., 2004; Meister et al., 2004), in mosquitoes, AGO subfamily proteins are functionally bifurcated: AGO1 facilitates miRNA interactions and AGO2 mediates siRNA interactions (Czech et al., 2009; Fu et al., 2017; Ghildiyal et al., 2010; Hammond et al., 2001; Kawamura et al., 2008; Okamura et al., 2004). AGO specialization is especially relevant in mosquitoes, where AGO2 is the main effector of the antiviral response (Blair, 2011; Mongelli and Saleh, 2016). In antiviral RNAi, AGO2 uses exogenous, viral-derived siRNAs (vsiRNAs) to target and cleave perfectly complementary viral RNAs. Mosquito AGO2 also serves analogous endogenous roles, using endogenous siRNAs (esiRNAs) to suppress invasive genetic elements, such as transposons, in the mosquito genome (Biryukova and Ye, 2015; Olmo et al., 2018). Therefore, due to its specialized AGO functions and transmission of human viruses, the Ae. aegypti mosquito is a prime target for systematic mapping of RNAi networks. Without a reliable network map that links small RNAs to their targets, study of host and viral small RNA expression profiles yield limited insight into the roles of small RNAs during viral infection.

To expand understanding of small RNA networks and biology in the major human disease vector Ae. aegypti, we developed AGO1- and AGO2-CLIP along with CLIPflexR, a streamlined computational pipeline suitable for general CLIP analysis in any organism (https://kathrynrozengagnon.github.io/CLIPflexR/). We applied AGO-CLIP protocols and CLIPflexR analysis to cells and whole mosquitoes to generate a comprehensive RNAi network map. This map links an expanded repertoire of 164 previously described and 230 previously unannotated mosquito small RNAs to their 5,447 high-confidence targets. We used the reported RNAi network map to contrast the biological processes regulated by each of the mosquito’s two AGO subfamily proteins. Notably, this resource facilitated two major insights into AGO2 biology with broad implications for our understanding of the mosquito antiviral machinery. First, examination of AGO2 binding preferences revealed that the Aag2 cell line, commonly used to study antiviral response, does not recapitulate AGO2 binding in vivo. Second, we demonstrated that AGO2 frequently binds and can repress imperfect, 3′UTR targets, a previously unappreciated mode of AGO2-mediated repression in insects. This study expands the toolkit available to researchers in vector and RNA biology and highlights the insights to be gained by examining RNAi networks in important infection models.

Results

Establishing AGO1- and AGO2-CLIP to map small RNA networks in mosquitoes

To map RNAi networks in Ae. aegypti mosquitoes and Aag2 cells we developed robust AGO1- and AGO2-specific CLIP protocols (Figures 1 and 2). Here, we describe these protocols, both as a roadmap for developing new CLIP protocols in non-model organisms and to provide confidence in the RNAi network map and other findings that we report.

Figure 1. Custom AGO1 and Drosophila AGO2 antibodies immunoprecipitate endogenous mosquito AGO.

(A) AGO-CLIP workflow to immunoprecipitate (IP) AGO ribonucleoproteins. UV = ultraviolet; RNase = ribonuclease A; CIP = calf intestinal alkaline phosphatase; Ab = antibody.

(B) IP-western of lysates confirms AGO1-CLIP IP specificity. Custom anti-AGO1 Abs (AGO1 Ab +) specifically IP a smear (blue brackets) within the size range of the multiple AGO1 isoforms (black lines, ∼100 – 125 kDa, top). AGO1 isoforms (FLAG-AGO1 short and long, black triangles) were overexpressed with 3xFLAG tags as size controls (middle). AGO2 (red asterisk) is not detectable in AGO1 IP (bottom). kDa = kilodaltons; WB = western blot.

(C) As in (A), using a Drosophila anti-AGO2 Ab cross-reactive with mosquito AGO2 (red asterisk, top). AGO2 3xFLAG (middle) and AGO1 (blue bracket, bottom) control immunoblots are shown Note cross-detection of the rabbit bridging antibody used in the IP results in a smear in the AGO1 blot.

(D) Immunoblot of AGO1 depletion from lysates following AGO1 IP. Pre-IP lysates show the initial AGO1 level (IP Ab -).

(E) As in (D), confirming depletion of AGO2 using an AGO2 Ab.

Vertical lines in (B-C) indicate removed lanes. All full-length immunoblots are provided in the Supplementary Information. See also Figure S1 and Table S1.

Figure 2. Generation of AGO1- and AGO2-associated RNA libraries from mosquito-derived lysates.

(A) AGO-bound RNA was ligated to a phosphorus-32 (P32) labeled RNA linker. AGO ribonucleoproteins (RNPs) were separated via sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and visualized by autoradiography. AGO RNPs were cut from membranes and proteinase K digested to liberate the RNA.

(B) In bromodeoxyuridine (BrdU) CLIP, reverse transcription (RT) was performed using a primer containing 3′ and 5′ linker sequences, degenerate barcodes and indices using BrdUTP as a dTTP analog. Complementary DNA was pooled, purified using anti-BrdU Ab, then circularized. Libraries were PCR amplified, further pooled and sequenced. In standard CLIP, RNA was ligated to a 5′ RNA linker with a degenerate sequence and RT was performed using the 3′ linker sequence. In the 1st PCR, cDNA was amplified and size selected. Indices were added with a 2nd PCR and libraries were sequenced.

(C) Autoradiogram of P32-labeled AGO1 RNPs from cells using the custom anti-AGO1 Ab. Irrelevant Ab (AGO1 Ab -), un-crosslinked (crosslink -), and high RNase (RNase +++) controls are shown. A smear is observed with low RNase (RNase +). RNP complexes are indicated (right).

(D) As in (C) for mosquito lysates.

(E) As in (C) for AGO2 RNPs.

(F) As in (E), for mosquito lysates.

(G) Libraries amplified for standard CLIP isolated from (C) at different PCR cycle numbers (each lane shows 4 additional cycles). No RT (noRT) and irrelevant Ab controls (rIgG) are shown. Excised regions for sequencing (brackets, right) include a strong band corresponding to small RNAs (∼60 bp) and a smear corresponding to target RNAs (∼75–150 bp).

(H) As in (G), for RNA isolated from (D).

(I) As in (G), for RNA isolated from (E).

(J) As in (I), for RNA isolated from (F).

Vertical lines in (C-J) indicate removed lanes. See also Tables S2 and S3.

For AGO1, our tests of available Ae. aegypti and Drosophila antibodies (Abs) showed that none were suitable (Figures S1A-S1E). We therefore elucidated mosquito AGO1 transcripts and protein isoforms (Figures S1F and S1G) and raised polyclonal Abs against a domain present in all isoforms (Figures S1I-S1O). To validate custom anti-AGO1 Abs (Figure 1A), we overexpressed the longest and shortest AGO1 isoforms with N-terminal 3xFLAG tags as positive controls. We confirmed that anti-AGO1 Abs detected 3xFLAG-tagged AGO positive control proteins in immunoblot as well as four distinguishable native AGO1 isoforms (Figure 1B, top, lysate). Anti-AGO1 Ab immunoprecipitated native AGO1, seen as a smear between the shortest and longest AGO1 isoform (Figure 1B, top, IP = immunoprecipitation).

For AGO2, 5′ RACE revealed a single AGO2 isoform in both cells and mosquitoes (Figures S1F and S1H). This isoform was cloned and used to validate a Drosophila anti-AGO2 Ab (Miyoshi et al., 2005) that pulled down mosquito AGO2 (Figure 1C, top). The two Abs were specific to their respective targets and were not pulled down with irrelevant antibody (IgG) controls (Figures 1D, 1E, and S1P; Table S1).

Following each IP of an AGO ribonucleoprotein (RNP), we ligated a radiolabeled RNA linker to enable visualization of AGO-associated RNA (Figure 2A). Autoradiograms of radiolabeled AGO RNPs revealed Ab-specific, crosslink- and RNase-dependent signal (Figures 2C–2F). We generated sequencing libraries from isolated AGO-associated RNA via two different cloning strategies (Figure 2B; Table S2). Standard CLIP relies on RNA linker ligations to introduce adaptor sequences for later amplification of libraries, while bromodeoxyuridine (BrdU) CLIP introduces PCR adaptor sequences during reverse transcription (RT)(König et al., 2010; Ule et al., 2005; Weyn-Vanhentenryck et al., 2014). We confirmed PCR amplification of the expected inserts: strong bands at ∼60nt correspond to cloned AGO-associated small RNAs, and smears at larger sizes correspond to targets (Figures 2G–2J). Signal was observed ∼four cycles earlier in specific Ab samples compared to paired IgG controls (Figure 2G–2J).

We applied these custom reagents and protocols to generate 41 mosquito and 56 mosquito cell line libraries (Table S3). The AGO-CLIP protocols we report were extensively experimentally validated and are the most comprehensive in insects thus far, due to our inclusion of both of the two mosquito AGO subfamily proteins. Furthermore, to our knowledge, AGO2-CLIP in a whole insect has not been reported to-date.

CLIPflexR: a generic R package for CLIP analysis

Lack of automated CLIP bioinformatic pipelines present obstacles for large datasets. Moreover, existing CLIP tools often rely on pre-installed genomes and annotations from model organisms, which was a difficulty in this case. We therefore had to develop custom code based on previous AGO-CLIP workflows (Chi et al., 2009; Moore et al., 2015; Shah et al., 2017) to analyze AGO-bound RNAs (or other RNPs) in any organism with a sequenced genome. To facilitate CLIP analysis for other researchers, we released this code as a streamlined R pipeline, CLIPflexR (https://kathrynrozengagnon.github.io/CLIPflexR/; Figure 3). The CLIPflexR package enables rapid installation and configuration of both the widely-used, perl-based CLIP Tool Kit (CTK) (Shah et al., 2017) and its many external software dependencies in a self-contained, highly reproducible and portable R/Conda environment (https://github.com/RockefellerUniversity/herper). Further, CLIPflexR allows for rapid scaling from small datasets on local computers to large datasets on high performance clusters through the BiocParallel. Most CLIPflexR functionality is easily adapted to CLIP proteins other than AGO and can be used in any organism. Therefore, CLIPflexR facilitates reproducible, parallelized and flexible CLIP analysis within a single software framework.

Figure 3. Overview of the CLIPflexR package.

Flowchart showing the advances made for CLIP analysis by the CLIPflexR package, and the relationship between CLIPflexR and the CTK (Shah et al., 2017). CLIPflexR enables easy installation and configuration of dependencies, in a self-contained package. Installation of the CTK via CLIPflexR confers the same benefits. CLIPflexR integrates the CTK functionality with generic peak calling and annotation tools suitable for CLIP analysis in any organism (unlike in the CTK, which is compatible with human, mouse and fly CLIP data)(Heinz et al., 2010; Yu et al., 2015). CLIPflexR also adds R-based read processing and counting functions for analysis of small RNAs and chimeras (Moore et al., 2015). Together CLIPflexR enables streamlined, flexible CLIP analysis using CLIPflexR or the CTK functionality, carried out within a reproducible, self-contained R/Conda environment. See also Table S5.

Expanding the small RNA repertoire in Ae. aegypti

Each AGO-CLIP experiment produces two datasets, namely AGO-loaded, small RNA sequences and their transcript targets. Because only 164 small RNAs were previously annotated in mosquitoes (known miRNAs), we first leveraged our datasets to annotate additional small RNAs. In mosquitoes and other insects, miRNAs are loaded onto AGO1 and esiRNAs are loaded onto AGO2 (Czech et al., 2009; Förstemann et al., 2007; Ghildiyal et al., 2010; Okamura et al., 2004). We applied miRDeep2 to AGO-associated sequences to identify previously un-annotated miRNAs and a subset of esiRNAs that arise from miRNA-like RNA hairpins (Friedlander et al., 2012)(Figure S2). Compared to applying miRDeep2 to unselected small RNA-seq data, applying it to AGO-bound small RNAs adds biochemical support and yields more reliable annotations. We henceforth use the term novel small RNAs to refer to small RNAs described in this study that are not present in miRBase; however, we cannot exclude the possibility that they have been described elsewhere but are not present in miRBase. Of these novel small RNAs, we refer to AGO1-associated small RNAs as novel miRNAs and AGO2-associated small RNAs as novel esiRNAs. Nevertheless, pending deeper investigation of small RNA processing, this classification remains putative. In total, miRDeep2 predicted 230 high-confidence, novel small RNAs (Table S4). Overwhelmingly, small RNAs were preferentially loaded onto either AGO1 or AGO2 (Figures 4A and 4B). As expected, the vast majority of known miRNAs were AGO1-associated, although miR-11894/miR-11895 were notable exceptions (Akbari et al., 2013; Miesen et al., 2016). These results enumerate a greatly expanded repertoire of small RNAs in Ae. aegypti and confirm global small RNA sorting between AGO proteins.

Figure 4. Transcriptome-wide bifurcation of AGO1 and AGO2.

(A) Abundances of high-confidence AGO1- or AGO2-loaded small RNAs in cells. 81 known miRNAs are shown. Novel AGO1-associated small RNAs (137 putative miRNAs, open circles) or AGO2-associated small RNAs (81 putative esiRNAs, open diamonds) are shown and small RNAs of interest are labeled; * indicates the small RNA is obscured; dashed line = equal abundance; RPM = reads per million mapped.

(B) As in (A), showing abundances of 95 known miRNAs, 27 novel AGO1-associated small RNAs (putative miRNAs), and 34 novel AGO2-associated small RNAs (putative esiRNAs) from whole mosquitoes.

(C-F) Percentages of high-confidence AGO1 or AGO2 peaks by genic region in cells or mosquitoes; n = number of peaks.

(G) Principal component analysis (PCA) of CLIP experiments for high-confidence AGO1 and AGO2 target peaks in cells. Irrelevant Ab controls are shown (rIgG for AGO1; mIgG for AGO2).

(H) The 8 most significantly enriched GO terms for AGO2 (red) and AGO1 (blue) for targets contributing to separation along PC2 based on fast pre-ranked enrichment analysis (fgsea). The ordering of GO terms is based on significance; color indicates normalized enrichment score (NES). White numbers indicate the number of target genes present for each GO term. Note that “host/virus part” refers to endogenous, integrated viral elements and is not due to active infection.

(I) Example peaks on target genes that drive PC2 for the most significant AGO1- or AGO2-enriched GO terms; mIgG, rlgG are control tracks; genome strand (+, -) is indicated.

(J-L) As in (G-I), for mosquitoes.

Novel = previously un-annotated. See also Figures S2 and S3; Tables S4, S5 and S6.

CLIPflexR delineates functionally bifurcated AGO1 and AGO2 target networks

We analyzed AGO-CLIP peaks using CLIPflexR to identify high-confidence AGO targets (Figures S3A-S3C; Table S5). ∼75% of AGO1 binding occurred in exons, with notable enrichment in 3′UTRs (Figures 4C and 4D). These results are consistent with the canonical mode of AGO1 repression in the miRNA pathway via 3′UTR binding and with comparable AGO-CLIP studies in related species (Bartel, 2009; Fu et al., 2020; Wessels et al., 2019). In contrast, AGO2 in cells exhibited dramatically different binding preferences than AGO1, and the majority of AGO2 binding occurred in introns and intergenic regions (Figure 4E). Surprisingly, AGO2 exhibited different target preferences in whole mosquitoes than in cells, and preferentially bound 3′UTRs at least as specifically as AGO1 (∼43% of all peaks; Figure 4F).

Consistent with differing roles for AGO1 and AGO2, principal component (PC) 2 in principal component analysis (PCA) reflected their distinct target preferences (Figures 4G–4L, S3D and S3E). As expected, PC1 separated the AGO libraries from IgG controls. To gain functional insight into targets driving separation of AGO1 and AGO2, we performed fast pre-ranked enrichment analysis (fgsea) on peaks ranked by PC2 value (Korotkevich, 2019). In both cells and whole mosquitoes, AGO1 bound target networks important for intra- and inter-cellular signaling, development, and morphogenesis (Figures 4G–4L; Table S6)(Fu et al., 2020; Wessels et al., 2019; Zhang et al., 2017). In cells, peaks preferentially bound by AGO2 occurred on transcripts derived from integrated endogenous viral elements (EVEs)(Figures 4H and 4I; Table S6). Thus, embryo-derived, cultured Aag2 cells may somewhat resemble embryo-derived, cultured Drosophila cells, in which AGO2 suppresses transposons and repeat features (Czech et al., 2008; Ghildiyal et al., 2008; Kawamura et al., 2008; Lan and Fallon, 1990; Rehwinkel et al., 2006). In mosquitoes, the top pathways for AGO2 were involved in basic cellular functions (Figures 4K and 4L; Table S6). For example, AGO2 preferentially bound a peak on a ribosomal protein mRNA, which contributed to AGO2 clustering by PCA and enrichment of ribosome biogenesis pathways. Although some AGO2 target pathways differed, enrichment of pathways involved in translation and cellular organization was shared between cells and whole mosquitoes (Figures 4H and 4K; Table S6).

A reliable miRNA-target network map in Ae. aegypti

AGO1-CLIP captures both AGO1 transcript targets and AGO1-loaded small RNAs, but it does not directly link miRNAs to their targets. Canonical miRNA target recognition relies on complementarity between the 5′ region of the miRNA (nucleotides 2–7; nt) and the target (6mer) (Bartel, 2009). Therefore, we asked whether AGO1 peaks were enriched for 6mer targets of abundant AGO1-loaded miRNAs in Ae. aegypti. Indeed, 6mers of the most abundant (top) known and novel miRNAs were enriched in AGO1 reads near the centers of target peaks (Figures 5A, 5B, S4A and S4B)(Chi et al., 2009; Luna et al., 2017). 6mer targets of known and novel miRNAs with similar abundances were enriched to a similar extent (Figures S4C and S4D).

Figure 5. AGO1 links abundant miRNAs to targets in vivo.

(A) Frequency of canonical 6mer motifs for the most abundant (top) known miRNA families in AGO1 reads by position in peaks, compared to AGO1 reads in control regions.

(B) As in (A), for novel miRNA families.

(C) Correlation between abundances of miRNA families and their predicted targets. Solid line = linear regression line of best fit for all miRNA families. Dashed line = linear regression line of best fit for miRNA families with abundance > 512 reads per million mapped (RPM; names are shown). The most abundant (top) miRNA families are highlighted in blue. Pie chart shows percentage abundance of miRNAs by family class. r = Pearson’s correlation coefficient.

(D) Correlation between abundance of AGO1-loaded miRNAs estimated by miRNA-target chimeras (y-axis) or direct counts (x-axis). Solid line = linear regression line of best fit; dashed line = equal abundance; miRNAs of interest are indicated; * indicates the small RNA is obscured; r = Pearson’s correlation coefficient.

(E) AGO1 read coverage (RPM) relative to 3′UTR metagene regions (aggregated 3′UTRs normalized by length) compared to control regions.

(F) Frequency of AGO1 chimeras in 3′UTR metagene regions compared to AGO1 chimeras over control regions. Frequency of all miRNA 6mer predicted targets is shown (green).

(G) Example AGO1 and rIgG control coverage over peaks in targets (black arrows). Predicted miR-2 family 6mers (open green triangles); the target portions of chimeras that contained miR-2 family members (green arrows) and miR-2-novel-a (open gray arrows); (x8) is the number of chimeric reads mapped at this location; genome strand (+, -) is indicated.

(H) As in (G), for a predicted aegypti-novel-1 family 6mer (open gray triangle); the target portion of a chimera that contained aegypti-novel-1a (open gray arrow) is shown.

Novel = previously un-annotated. See also Figure S4; Tables S4 and S5.

Because AGO-CLIP captures both small RNAs and their targets, we expected them to be positively correlated in our data. We considered 3′UTRs that contained 6mers of high-confidence miRNA families to be AGO1 targets. For abundant miRNA families, miRNA and target abundances were positively correlated (Figures 5C and S4E)(Chi et al., 2009; Luna et al., 2015; Scheel et al., 2017). Therefore, highly expressed miRNA families are more likely than rare miRNA families to direct AGO1 target binding in mosquitoes and cells.

Unambiguously linking miRNAs to their targets remains difficult (Grosswendt et al., 2014; Helwak et al., 2013; Kudla et al., 2011). However, during AGO-CLIP, rare ligations occur between miRNAs and their paired targets, creating chimeric molecules that unambiguously link both RNAs together (Table S3)(Luna et al., 2017; Moore et al., 2015; Scheel et al., 2017). In our data, the abundances of miRNAs and of chimeras containing miRNAs were highly correlated; the existence of chimeras is strong evidence that these miRNAs indeed direct AGO1 binding (Figures 5D and S4F). Overall read coverage and chimera coverage confirmed that AGO1 preferentially binds 3′UTRs and that bona fide binding events are enriched in 3′UTRs (Figures 5E, 5F, S4G and S4H).

To establish a robust miRNA-target mRNA network map, we required that high-confidence AGO1 peaks contained the 6mer target sequence of a miRNA or had at least one chimeric read that mapped within 70nt of the peak center (Table S5)(Chi et al., 2009). As an example in which both conditions were satisfied, chimeras with miR-2 family member sequences map to AGO1 target peak regions containing predicted 6mers from this family (Figures 5G and S4I). Mapping of aegypti-novel-1a provides an analogous example for a novel miRNA family (Figure 5H).

Therefore, a subset of novel miRNAs form chimeras, providing direct evidence that they function to target AGO1 to mRNAs. Together, these data are the basis for an expanded atlas of miRNA networks in cells and mosquitoes (Table S5).

AGO2 binding of 3′UTRs, repeat families and EVEs differs between cells and mosquitoes

To further characterize AGO2 target networks, we investigated the divergence in AGO2 genic target preference between cells and mosquitoes noted above (Figure 4). We obtained similar results when we restricted our annotation to targets of abundant, AGO2-loaded small RNAs (Figures S3F and S3G); thus, differential small RNA expression does not explain disparate AGO2 binding preferences. Further, AGO2 exhibited ∼2.4-fold higher coverage over 3′UTR regions in whole mosquitoes compared to cultured cells (Figure 6A), consistent with our prior observation that, ∼40% of AGO2 binding occurs in 3′UTRs in mosquitoes. In contrast, in Aag2 cells we observed nearly half of all AGO2 binding in intergenic regions. Intergenic regions lack annotated genes and often contain unannotated noncoding transcripts, transposons, and repetitive features. Due to AGO2 specificity for intergenic regions in cultured cells and its known role in transposon silencing, we asked whether repeat features contributed to intergenic AGO2 binding (Czech et al., 2008; Ghildiyal et al., 2008; Kawamura et al., 2008; Olmo et al., 2018; Rehwinkel et al., 2006). Although AGO2 bound intergenic repeat features in both cell and mosquitoes, the types of repeats differed (Figure 6B). In cells, AGO2 bound DNA transposon and retrotransposon transcript families, consistent with AGO2-mediated transposon silencing (see example tracks in Figures S5A and S5B). In mosquitoes, AGO2 bound low complexity repeats and ribosomal RNA (rRNA). The disparate AGO2 binding of 3′UTRs and repetitive features in cultured, embryo-derived cells compared to whole adult mosquitoes may explain its regulation of different functional target networks in these two contexts.

Figure 6. AGO2-CLIP captures hallmarks of antiviral RNAi on transposon and virus transcripts and uncovers imperfect AGO2 targets.

(A) AGO2 read coverage (RPM = reads per million mapped) relative to 3′UTR metagene regions for AGO2 reads from cells or mosquitoes (signal over control regions is subtracted for comparison). The mean read depth was calculated by sample type (cells or whole mosquitoes, solid horizontal lines) and the fold-change (dashed vertical line) between the average coverage for each sample type is indicated.

(B) AGO2 targets different repeat families in cells and mosquitoes. The percentage of reads mapped to each repeat family (graph titles) is indicated. LTR ERV1 = endogenous retroviral sequence 1 long terminal repeats; SINE tRNA Deu = tRNA-derived short interspersed nuclear elements; rRNA = ribosomal RNA; low complexity = low complexity repeats. Classes shown are significant in either cells or mosquitoes (Student’s t-test, *p-value < 0.01) compared to paired mIgG controls and have FDRs < 0.02. Gray dots = outliers.

(C) As in (B) for EVEs derived from different virus families. Classes shown are significant in cells (Student’s t-test, *p-value < 0.01) compared to paired mIgG controls and have FDRs < 0.02. Gray dots = outliers.

(D) Lengths and strand bias for short reads mapping to the CFAV (open bars) or the mosquito (Aag2, closed bars) genome from AGO1 or AGO2 libraries from cells. Y-axis: the frequency of short reads of different lengths that mapped to the positive (+) or negative (−), normalized to the total number of short reads.

(E) AGO2 coating of the cell fusing agent virus (CFAV) RNA genome; partial 3′UTR is expanded in shaded region; genome strand (+,-) is indicated.

(F) Frequency of predicted targets for the most abundant novel esiRNA families by position in AGO2 peaks in mosquitoes.

(J) Most significantly enriched auxiliary motif in peak centers with perfect 18mer targets (of esiRNAs in (F); motif itself is outside of perfect targets). This may be a motif recognized by AGO2 itself.

Novel = previously un-annotated. See also Figure S6; Tables S4 and S5.

Another important difference in the functional networks bound by AGO2 was the specific enrichment of EVEs in cells (Figures 4H and 4K). EVEs are often unannotated in the mosquito genome and thus may also contribute to intergenic binding. Interestingly, EVEs have been increasingly implicated as a potential source of memory in mosquito antiviral defense (Blair et al., 2020). However, to our knowledge there has not been a direct link established between AGO2 and EVEs. We examined whether EVEs were bound by AGO2 (Figure 6C). In cells alone, AGO2 significantly bound EVEs derived from various viral families (Flaviviridae, Virgaviridae, Rhabdoviridae, and Phasmaviridae). These results suggest that AGO2 may play a role in regulating invasive genetic elements derived from viruses.

Consistent differences in AGO2 binding preferences likely reflect dissimilarities in target transcriptomes between cultured, embryo-derived Aag2 cells and homogenized adult mosquitoes. We cannot discriminate whether the key distinction in AGO2 genic target preference is related to cell culture versus in vivo models or to embryonic versus somatic cellular origins. Nonetheless, our data suggest that in diverse, primarily somatic cells from mosquitoes, AGO2, like AGO1, may repress targets via 3′UTR binding, potentially via mechanisms other than cleavage. Our examination of AGO2 target preference also highlights prioritization of binding to transposons and EVEs in Aag2 cells.

AGO2-CLIP captures hallmarks of siRNA targeting and immune interactions

Interestingly, in some cases we observed fairly equal coating of AGO2 on both positive and negative strands in transposon-rich regions (Figure S5B). This coverage profile is reminiscent of those observed for confirmed sources of esiRNAs in Drosophila (Czech et al., 2008; Kawamura et al., 2008). Further, near-equal balance of positive- and negative stranded-reads is a key hallmark of viral-derived siRNA (vsiRNA) generation during antiviral RNAi (Blair, 2011; Mongelli and Saleh, 2016). However, it remains unclear whether the equal strand bias of siRNAs leads to targeting of both mRNA strands. To systemically interrogate whether equal strand ratios of esiRNAs are reflected in AGO2 binding, we examined strand bias on cellular target RNAs by quantifying strand bias at overlapping peaks (Figures S5C-S5E). We hypothesized that because esiRNAs can arise from convergent transcription, AGO2 would exhibit less target strand bias than AGO1. Indeed, a smaller fraction of overlapping AGO2 peaks had strand bias compared to overlapping AGO1 peaks. These results indicate that AGO2 utilizes esiRNAs in both senses to bind cellular RNAs.

Because we observed equal strand bias (a key siRNA hallmark) on cellular RNAs, we asked whether AGO2-CLIP would reveal similar signatures during exogenous virus infection. During infection, vsiRNAs are generated from dsRNA replication intermediates and direct AGO2 to bind and silence viral transcripts. Small RNA-seq is commonly applied to characterize vsiRNAs and captures two hallmarks of antiviral RNAi: equal strand bias and enrichment for 21nt vsiRNAs. However, small RNA-seq provides no information on which vsiRNAs are loaded into AGO2 and target viral RNA, and thus functionally mediate immune responses. It is of clear interest to apply AGO2-CLIP to better understand mosquito antiviral defenses.

Aag2 cells are persistently infected with a number of viruses, including cell fusing agent virus (CFAV, a flavivirus) and, in some cases, Culex Y virus (CLY, an entomobirnavirus) and Phasi Charoen-like virus (PCLV, a phasivirus)(Franzke et al., 2018; Maringer et al., 2017; Stollar and Thomas, 1975). We investigated whether vsiRNAs from these viruses were loaded into AGO2 and thus could direct binding to viral RNA genomes. We selected short, virus-mapped reads (18–24nt) as putative vsiRNAs. As expected, for all viruses examined, vsiRNAs were specifically associated with AGO2. For example, CFAV vsiRNAs comprised ∼0.41% of AGO2 libraries versus ∼0.017% of AGO1 libraries (Table S3). Consistently, AGO2 bound the CFAV 3′UTR and AGO2-loaded vsiRNAs exhibited key hallmarks of vsiRNA biogenesis (Figures 6D and 6E). In contrast, AGO1 binding to CFAV was negligible and AGO1-loaded small RNAs derived from CFAV did not exhibit siRNA hallmarks. To directly compare length and strand biases for all vsiRNAs (small RNA-seq) with those of AGO2-loaded vsiRNAs (AGO2-CLIP), we reprocessed published small RNA-seq from three different sources of Aag2 cells (Haac et al., 2015; Ma et al., 2021; Miesen et al., 2016). AGO2-loaded vsiRNAs reflected hallmarks of antiviral RNAi similar to those observed in small RNA-seq for all three viruses (Figures S5F-S5H). These results indicate that AGO2-CLIP detects bona fide vsiRNAs and reveals binding of viral target RNA.

AGO2-CLIP captures cleavage of perfect targets of novel esiRNAs

In the context of antiviral RNAi, AGO2 is currently thought to primarily act by cleaving targets containing perfect reverse complements of siRNAs. We therefore investigated whether AGO2 peaks contained perfect targets of novel esiRNAs. Because the 5′ regions of siRNAs are important for target recognition, and to include novel esiRNAs of varying lengths, we defined 18mer target sequences complementary to first 18nt of novel esiRNAs (Elbashir et al., 2001; Haley and Zamore, 2004). For the most abundant novel esiRNAs, we searched for predicted 18mers with 0 or 1 mismatches or miRNA-like 6mers in AGO2 peak sequences (Tables S4 and S5). Surprisingly, perfect (0 mismatch) targets were depleted at peak centers (Figures 6F and S6A). Conversely, there was strong enrichment for imperfect (1 mismatch) 18mers. We hypothesize that this reflects faster AGO2 cleavage activity compared to AGO1-mediated repression: AGO2 cleaves perfect targets quickly and these are therefore difficult to capture by CLIP. To further support this hypothesis, we asked whether chimeras were captured by AGO2-CLIP. Chimera formation is predicated on capturing both small RNA and target RNA in AGO simultaneously; if AGO2 cleaves fast and then disassembles from target RNAs, those AGO2 complexes would rarely generate chimeras. Consistent with a faster catalytic rate of cleavage, AGO2 formed very few chimeras compared to AGO1, although AGO2 formed more chimeras in mosquitoes than in cells (Figures S6B-S6E; Table S3); no chimeras were observed mapping to perfect targets. However, while AGO2 bound imperfect targets, it preferred regions with high complementary, as miRNA-like 6mer targets were not enriched at peak centers.

We noted that in peak sequences that contained a perfect 18mer, we often found many additional instances of the same 18mer (Figure S6F). To understand how AGO2 discriminates between identical 18mers, we searched for sequence motifs enriched in peak centers compared to flanking regions to uncover auxiliary sequences outside the 18mer that may guide AGO2 specificity. These may be motifs recognized by AGO2 itself or other RISC proteins. Four of the five nt of the most significant motif (of the two identified; Figures 6G and S6G) are shared with an AGO2 binding motif uncovered in mammals (Leung et al., 2011).

AGO1 RNAi network maps predict expression levels of ovarian miRNA targets

If the reported RNAi network map is correct, it should be able to predict transcript levels. To assess this for AGO1, we examined expression of target transcripts of abundant ovarian miRNAs. We selected known miRNAs that are abundant in ovaries (> 5000 reads per million mapped; RPM) and rare in female carcasses (< 1000 RPM)(Akbari et al., 2013). We used the RNAi network map (Table S5) to identify 3′UTR targets of these ovarian miRNAs and selected the three miRNAs with ≥50 targets for further analysis: miR-989, miR-996, and miR-2946. Target expression levels were obtained from a published mRNA-seq atlas of Ae. aegypti tissues, including the ovary, following different feeding conditions (Matthews et al., 2016). Targets of control miRNAs that were rare in the ovaries and abundant in the carcass exhibited similar expression profiles in the ovaries compared to all other tissues (Figures S7A-S7C). In contrast, for all three abundant ovarian miRNAs, target expression was reduced in the ovaries (Figures 7A-7C and S6C), consistent with AGO1-mediated transcriptional repression. This reduction was most striking for targets of miR-989, which had the highest ovarian abundance among the three miRNAs; miR-989 target levels in the ovary were decreased over the entire range of target expression (Figure 7A). miR-996 and miR-2946 target expression was also reduced for low-abundance, but not high-abundance targets (Figures 7B and 7C). This is likely explained by lower expression of these miRNAs compared to miR-989. Although we cannot exclude the contribution of transcriptional or other regulation, target expression for all miRNAs selected was consistent with predictions. Thus, we demonstrate how AGO1 RNAi network maps can link independently generated datasets and predict expression levels of miRNA targets.

Figure 7. Imperfect AGO1 and AGO2 targets exhibit reduced expression.

(A-C) Empirical cumulative distribution functions (eCDFs) of the average expression of predicted 6mer and chimera supported 3′UTR targets of miR-989, miR-996, and miR-2946. Title shows miRNA and its expression in ovaries; RPM = reads per million mapped. Sugar-fed (SF) or gravid (O) ovaries (Ov) were compared to all tissues. One-sided Mann-Whitney U-test p-values of the abundances of target mRNAs in ovaries versus all tissues are shown. RPKM = reads per kilobase million.

(D) Expression of known miRNAs (known), the most abundant novel esiRNAs (top novel), and all other novel small RNAs (novel) in mosquito tissues. Ov = ovary; PBM = postblood-meal; NBF = nonblood-fed; hr = hour. ***p-value < 1e-5; **p-value < 1e-3; *p-value < 0.01; ns = p-value > 0.05; two-sided Mann-Whitney U-tests compared to carcass.

(E) eCDF of the average log2fold-change (log2FC) of expression for targets of the most abundant novel esiRNAs (top novel; from Figure 6D) in sugar-fed ovaries (Ov SF) compared to all tissues. Predicted targets with 6mers outside 3′UTRs = non 3′UTR targets; 3′UTR 6mers = 6mer; 3′UTR 8mers = 8mer; chimera-supported targets = chimera; 0 mismatch 18mers = perfect. One-sided Mann-Whitney U-test p-values comparing target types to non 3′UTR targets are shown.

(F) As in (E), for gravid ovaries (Ov O).

(G) Luciferase reporter assay measuring repression of perfect and shuffled reporters for each AGO2-associated small RNA. Renilla luminescence (Rluc) was normalized to firefly luminescence (Fluc), and then to an empty, unrepressed Rluc construct containing an unrelated sequence. ***p-value < 1e-4; **p-value < 0.01; *p-value < 0.05; ns = p-value > 0.05; one-way analysis of variance (ANOVA) for each small RNA. Overall ANOVA: p-value < 0.05 for all small RNAs. Perfect and shuffled reporters were compared to the empty reporter using the Dunnett’s post hoc test.

(H) As in (G), for miR-11894 family reporters, with and without co-transfection of locked nucleic acid (LNA) or small RNA mimic. ***p-value < 1e-4; **p-value < 0.01; *p-value < 0.05; ns = p-value > 0.05; one-way ANOVA. Overall ANOVA: p-value < 1e-4. Reporters without any treatment were compared to the empty reporter, and LNA- and mimic-treated reporters were compared to the same reporter without treatment. Comparisons were performed using the Dunnett’s post hoc test.

(I) As in (H) for the novel esiRNA Aag2-novel-2a (sole family member). Overall ANOVA: p-value < 1e-4.

Novel = previously un-annotated. Data in panels G-I are represented as mean ± SEM. See also Figure S7.

AGO2 represses imperfect as well as perfect targets

We investigated whether mosquito AGO2 represses both perfect and imperfect targets, which has been reported in synthetic systems in mammalian cells (Broderick et al., 2011; Doench et al., 2003; Du et al., 2005; Zeng et al., 2003). We reanalyzed small RNA-seq data and confirmed that the most abundant (top) novel esiRNAs we identified were highly expressed in mosquito ovaries, providing independent validation of our esiRNA annotation (Figure 7D)(Akbari et al., 2013). Several types of novel esiRNA targets were present in our AGO2 RNAi network maps, including perfect 18mer, chimera-supported, and miRNA-like 6mer targets (Table S5). In addition, we included miRNA-like 8mer targets to increase target prediction specificity and the degree of complementarity between novel esiRNA and target (Bartel, 2009; Luna et al., 2015). Consistent with expression profiles of novel esiRNAs, their target transcript levels were decreased in the ovary (Figures 7E, 7F and S7D). While, as expected, expression of perfect targets was strongly decreased, we also observed more subtle decreases in imperfect target levels. Imperfect, miRNA-like 8mer target expression was slightly but significantly lower in sugar-fed ovaries (Figure 7E). In addition, chimera-supported, imperfect target levels were significantly reduced in the ovaries generally (Figures 7E and 7F).

Because we observed a global decrease in imperfect, 3′UTR target levels in the ovaries, we asked whether AGO2 could mediate repression of such targets in Aag2 cells. While AGO2 in Aag2 cells did not preferentially bind 3′UTRs, we did find evidence of binding to imperfect targets (Figure S6A). Thus, AGO2 in both contexts appears capable of imperfect target recognition. We selected AGO2-associated, abundant novel esiRNAs (> 500 RPM; novel esiRNAs were generally less abundant than miRNAs). We also included two members of the highly abundant, AGO2-associated miR-11984 family: miR-11894a and miR-11894-novel-a (> 5000 RPM). First, we confirmed that perfect target sequences inserted into the 3′UTR of a Renilla luciferase reporter gene were repressed by AGO2-associated, endogenous small RNAs (Figure 7G). Only perfect, not shuffled targets of all small RNAs were significantly repressed.

Next, we examined whether AGO2 could use small RNAs to repress imperfect, 3′UTR targets. We selected the miR-11894 and Aag2-novel-2a families, as these exhibited the highest degree of repression. We designed bulged targets with a 3nt central mismatch as well as canonical 8mer targets (Doench et al., 2003; Zeng et al., 2003). We compared repression of perfect, bulged, and 8mer targets in cells alone, or in cells treated with locked nucleic acid (LNA) small RNA inhibitors or small RNA mimics (Figures 7H and 7I). For both families, we observed repression of bulged, but not 8mer targets, indicating that the degree of complementarity is an important determinant for the strength of repression. While 8mer targets were not repressed, we note that the decrease in 8mer target levels we observed in the ovaries was subtle and specific to sugar-feeding (Figure 7E). Thus, it is possible that repression of 8mer targets can only be observed for aggregated targets, or only occurs in certain contexts. For all reporters that were significantly repressed, we observed some degree of de-repression upon inhibition of the small RNA via LNA treatment. For miR-11894 family reporters, the addition of miR-11894 mimics resulted in a consistent, but slight and insignificant increase in repression. This likely relates to high abundance of miR-11894a and miR-11894-novel-a, which may mask any additional repressive effect upon the addition of mimics (∼9000 RPM and ∼6600 RPM, respectively; Table S4). However, for Aag2-novel-2a (∼600 RPM), addition of mimic significantly increased repression of both perfect and bulged reporters, confirming specific repression of both perfect and imperfect targets. These data highlight repression of imperfect, 3′UTR targets by mosquito AGO2 in an AGO1-like fashion, a phenomenon to our knowledge previously described only in artificial systems. Further, our results suggest that this AGO1-like mode of repression may function in the ovary, resulting in widespread reduction in imperfect target levels.

Discussion

In this study, we developed robust AGO-CLIP experimental protocols and CLIPflexR, an omnibus software package for CLIP data analysis, and then generated a comprehensive RNAi network map for Ae. aegypti. The results reported here provide an experimental roadmap for establishing CLIP in non-model organisms and the CLIPflexR package will aid investigators from diverse disciplines in their study of RNPs in any organism.

The AGO1 miRNA-target network map reported here supports several advances over previous knowledge of the mosquito miRNA pathway, including results from the only Ae. aegypti AGO1-CLIP study reported to-date (Figure S3H)(Zhang et al., 2017). First, in the previous study only ∼9% of AGO1 peaks were in 3′UTRs, which was inconsistent with prior results from other species. In the present study, as expected, 23% to 42% of AGO1 peaks were in 3′UTRs. Second, we identified chimeras and thereby provided concrete evidence that a subset of AGO1-bound miRNAs indeed direct AGO1 binding. Third, we demonstrated the quality and predictive ability of reported miRNA networks by linking miRNA expression to mRNA transcript levels from two independent RNA sequencing studies (Akbari et al., 2013; Matthews et al., 2016). Thus, we provide an experimentally supported AGO1 network map of global miRNA-target mRNA interactions for researchers in vector and infection biology (Table S5).

The AGO2 siRNA-target network map reported here constitutes a substantial advance, as, to our knowledge, there have been no AGO2-CLIP studies in Ae. aegypti, nor in any insect in vivo. The AGO2-CLIP data led to several important findings. First, contrasting AGO1- and AGO2-CLIP datasets confirmed small RNA sorting between AGO1 and AGO2 and yielded insights into their functional roles. Second, we were able to capture AGO2-mediated antiviral responses to a number of persistent virus infections in Aag2 cells. It would be highly relevant to apply AGO2-CLIP to medically important viruses, to investigate whether our findings apply more generally and to characterize the mosquito’s functional antiviral defense. Third, we found that in mosquitoes, AGO2 widely relies on previously unknown AGO1-like functionality which has several facets, including repression of imperfect, 3′UTR targets.

Unexpectedly, AGO2 showed high specificity for 3′UTRs in whole mosquitoes (similar to AGO1’s specificity). To our knowledge, there have been no reports of 3′UTR preference for any insect AGO2 or in antiviral AGO2 cleavage-mediated repression. This preference for 3′UTRs in mosquitoes was in contrast to AGO2 preference in cultured cells, where AGO2 bound intergenic regions, transposons, and integrated viral elements. Interestingly, AGO2 appeared to be more highly expressed in cells than in whole mosquito homogenate. This heterogeneity may be responsible for disparate target preferences; indeed, variable AGO2 expression and binding preferences across cellular compartments and cell types have been reported other species (Clark et al., 2014; Sarshad et al., 2018). It remains unclear whether distinct binding preferences stems from differences in cell culture compared to in vivo systems, or from embryonic compared to somatic cellular origins. In either case, Aag2 cells may not be appropriate for the study of mosquito immune responses in light of our finding that they do not recapitulate AGO2 3′UTR preference in whole mosquitoes.

AGO2’s unexpected preference for 3′UTRs in whole mosquitoes was linked to a widespread reduction in expression levels of imperfect targets, another facet of AGO1-like functionality (Chendrimada et al., 2005; Meyer et al., 2006). We confirmed that imperfect 3′UTR target levels were decreased in the ovary in an AGO1-like fashion. We speculate that in the ovary, silencing of transposable or viral elements by imperfect complementarity may be particularly important to avoid inherited transmission of mutations caused by these elements (Czech et al., 2008; Fu et al., 2017; Ghildiyal et al., 2008; Kawamura et al., 2008). Consistently, upregulation of transposons in ovaries is associated with DNA damage and arrest of oogenesis in flies (Durdevic et al., 2018; Klattenhoff et al., 2007). Thus, our results suggest that failure to silence transposons may contribute to developmental defects in the embryos of AGO2 null flies (Deshpande et al., 2005; Okamura et al., 2004). While AGO2 is known to repress imperfect targets in synthetic systems, our results suggest that AGO2 may mediate AGO1-like imperfect repression of 3′UTR targets in vivo (Figure 7)(Broderick et al., 2011; Doench et al., 2003; Du et al., 2005; Zeng et al., 2003). Although AGO2 did not prefer 3′UTRs in Aag2 cells, we did observe binding of imperfect targets in cells. Further, we demonstrated that if supplied with appropriate imperfect 3′UTR targets, AGO2 can mediate AGO1-like repression in Aag2 cells. Given these results, in the future this functional mode of AGO2 should be considered in the context of antiviral RNAi.

In conclusion, we developed robust AGO-CLIP protocols and CLIPflexR, a freely-available, generic, and user-friendly software pipeline, and used them to generate comprehensive AGO1 and AGO2 network maps for a major vector of human disease, the Ae. aegypti mosquito.

STAR methods

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Charles M. Rice (ricec@rockefeller.edu).

Materials availability

All unique reagents generated in this study are available from the Lead Contact upon completion of a Materials Transfer Agreement. There is restriction to the availability of anti-mosquito AGO1 polyclonal antibody and purified mosquito AGO1 PAZ domains based on laboratory stock levels, as these are limited and labor-intensive items to generate.

Data and code availability

All full-length western blots and gels for mass spectrometry are available in the Supplementary Information. Experimentally validated AGO transcript/protein isoforms were deposited in GenBank under accessions: MW035627-MW035631. The code used in this study is available at the following link: < https://kathrynrozengagnon.github.io/AGOCLIP_2020/ >. Key components of this code were developed into an R pipeline for CLIP analysis, including read processing, mapping, peak calling, matrix building, pattern searching, small RNA counting, and chimeric RNA analysis: < https://kathrynrozengagnon.github.io/CLIPflexR/ >. In addition to offering R-based alternatives for many analysis steps, we also wrapped the previously published CTK toolkit (Shah et al., 2017) in R for easier installation and analysis. The AGO-CLIP sequencing data generated in this study has been deposited in the GEO under accession: GSE157168. Filtered small RNA abundances by direct counting and chimera counting is included in Table S4. Simplified AGO-CLIP RNAi network map is included in Table S5. An extended RNAi network map including extensive filtering information, annotation, prediction of small RNA targets, and presence of chimeras is available at: < https://kathrynrozengagnon.github.io/AGOCLIP_2020/ >.

Experimental model and subject details

Cells and mosquitoes

Drosophila melanogaster S2 cells (Thermo Fisher Scientific) were grown in Schneider’s Drosophila media (Thermo Fisher Scientific), supplemented with 10% heat inactivated fetal bovine serum (FBS, Hyclone, GE Healthcare) and 2 mM L-Glutamine (Thermo Fisher Scientific) at 28°C, 0% CO₂. Ae. aegypti Aag2 cells (Lan and Fallon, 1990) were a kind gift from Dr. Maria Carla Saleh and were cultured in Leibovitz’s L-15 Medium, no phenol red (Thermo Fisher Scientific) supplemented with 10% FBS, 0.1 mM non-essential amino acids (Thermo Fisher Scientific), and ∼0.3g/L tryptose phosphate broth (Sigma-Aldrich) at 28°C, 0% CO₂.

Whole female adult mosquitoes (Ae. aegypti, Orlando) were reared according to standard protocols (Kauffman et al., 2017) on a diet of ground Koi pellets and flash frozen at days 5–7 post-eclosion.

Method details

AGO isoform verification and cloning

Total RNA from whole homogenized mosquitoes or Aag2 cells was obtained using standard TRIzol Reagent (Thermo Fisher Scientific) extraction protocols. RNA was DNase-treated (RQ1 RNase-free DNase, Promega), according to manufacturer’s protocol. RNA was purified by standard phenol:chloroform extraction and 5′ RACE was performed on mosquito or cell total RNA using the FirstChoice RLM-RACE Kit (Thermo Fisher Scientific) according to manufacturer’s instructions. Gene-specific antisense primers were designed in conserved AGO regions (all oligos used in the study are listed in Table S2 and were purchased from IDT) and PCRs were performed using Phusion (New England BioLabs, NEB) according to manufacturer’s protocol. Following nested PCR specific bands were size selected and gel extracted (QIAquick Gel Extraction Kit, Qiagen), dATP-tailed using Taq (Thermo Fisher Scientific), and TOPO-TA cloned into the pCR 2.1-TOPO TA vector (Thermo Fisher Scientific) according to manufacturer’s protocols. To verify AGO coding sequences, reverse transcription (RT) was performed on total RNA using Superscript III (Thermo Fisher Scientific) and oligoDT primer according to manufacturer’s protocol. cDNA was amplified using Phusion and primers designed upstream of the end of 5′ RACE sequences and near the end of AGOs. As with 5′ RACE, products, bands were size selected, TOPO-TA cloned and sequenced. Experimentally validated Ago transcript/protein isoforms were deposited in GenBank under accessions: MW035627-MW035631.

Following cloning of AGO coding sequences into pCR 2.1-TOPO, plasmids were digested to generate overlapping fragments between 5′ RACE and downstream coding sequences. Gel-extracted DNA fragments were assembled by Gibson Assembly (NEB) according to manufacturer’s protocol. Assembled DNA was used as a template to PCR the entire AGO coding sequence, and primers were used to add 5′ homology with the 3xFLAG-tag and 3′ homology to the destination vector. The 3xFLAG tag sequence was amplified by PCR, and primers were used to add a 5′ homology to the destination vector and a 3′ overhang corresponding to the beginning of AGO. PCRs were gel extracted, assembled, and PCR amplified again to generate 3xFLAG-tagged AGOs with overhangs extending past destination vector RE sites. The destination vector, a modified in-house vector (pKRG4) based on pSL1180-HR-PUbECFP (a gift from Leslie Vosshall, Addgene plasmid #47917) was linearized and digested. Vector and 3xFLAG-tagged AGO PCRs were gel extracted, and Gibson assembled. Assembled plasmids were transformed into DH5alpha and plasmid DNA was purified (QIAquick Miniprep kit, Qiagen). The final plasmids utilize the Ae. aegypti polyubiquitin (PUb) promoter to express N-terminal 3xFLAG-tagged AGOs that are polyadenylated via the SV40 polyadenylation sequence (pKRG4-xFLAG-AGO1-short, pKRG4–3xFLAG-AGO1-long, pKRG4–3xFLAG-AGO2).

Transfection of 3xFLAG-tagged AGOs

Aag2 cells were transfected with pKRG4 AGO expression plasmids using Fugene HD Transfection Reagent (Promega) according to the manufacturer’s protocol. Cells were seeded at ∼50% confluency and complexes were formed using a ratio of 3:1 transfection reagent to plasmid DNA.

SDS-PAGE and immunoblot

SDS-PAGE and immunoblots were performed using the Mini Gel Tank and NuPAGE gels and buffers (Thermo Fisher Scientific). For protein gels, equal sample volumes were prepared in 1X LDS Sample Buffer and 50 mM DTT (Sigma-Aldrich). Samples were incubated for 10 minutes at 70°C, loaded on 4–12% Bis-Tris gels and run in 1X MOPS SDS Running Buffer according to manufacturer’s instructions alongside Precision Plus Protein Dual Color Standards (Bio-Rad). Gels were stained with 0.1% w/v Coomassie Brilliant Blue R-250 (Thermo Scientific) then destained. For immunoblot, cell pellets were lysed in ice-cold 1X PXL CLIP lysis buffer (0.1% SDS, 0.5% sodium deoxycholate, 0.5% NP-40 in 1X PBS) plus protease inhibitor (cOmplete, Mini Protease Inhibitor Cocktail Tablets, EDTA free, Roche) on ice for 10 minutes and clarified by centrifugation at 15,000 rpm for 20 minutes at 4°C. Supernatants were collected and total protein was determined by BCA assay (Pierce BCA Protein Assay Kit, Thermo Scientific) using a FLUOstar Omega Microplate Reader (BMG LABTECH). 10 μg of total protein was prepared and run as above. Following electrophoresis, protein was transferred (Blot Module Set, Thermo Fisher Scientific) onto 0.2 μM nitrocellulose (Amersham Protran Premium, GE Healthcare) according to manufacturer’s protocol. Blots were developed using fluorescent detection (LI-COR) with the following primary antibodies: Anti-AGO1 (see below for details), anti-AGO2 (9D6, a kind gift from Dr. Mikiko Siomi, published in Miyoshi et al., 2005), and anti-FLAG (monoclonal M2, Sigma-Aldrich). Following development membranes were imaged (Odyssey CLx, LI-COR).

Protein expression and purification

AGO1 PAZ domain sequences (residues 282–431 in experimentally determined sequences) were PCR amplified from pKRG4 expression vectors with 5′ and 3′ sequences designed for insertion into modified pET28a-His6-Smt3 (the yeast small ubiquitin-like modifier protein)(Gu and Rice, 2016). Products were PCR purified (QIAquick PCR Purification Kit, Qiagen). The pET28a-His6-Smt3 vector was digested and gel extracted. AGO1 PAZ inserts and vector were Gibson assembled, transformed, and plasmid DNA was extracted. For bacterial AGO1 PAZ expression, pET28a-His6-Smt3-AGO1-PAZ was transformed into Rosetta (DE3) competent cells (Novagen) according to the manufacturer’s protocol. Cultures were grown at 37°C to OD₆₀₀ 0.5, before temperature was shifted to 25°C. Protein expression was induced with 0.5 mM IPTG for 4 hours. At OD₆₀₀ ∼1, bacteria were pelleted and resuspended in 20% sucrose and 1.5X PBS pH 7.2. The cells were flash frozen in liquid nitrogen, and stored at −80°C. To purify AGO1 PAZ, cells were lysed by sonicating (five pulses for 20 seconds each) in lysis buffer (500 mM NaCl, 5% glycerol, 1 mM PMSF, 0.2% NP-40, 2 ug/mL DNase I, 2 ug/mL RNase A, 15 mM imidazole, 1 mM βME, 10mM Tris pH 8.5 in 1X PBS) at 4°C. Following lysis, insoluble material was pelleted at 19,000 rpm for 1 hour at 4°C. Supernatant was incubated with nickel resin (Ni Sepharose 6 Fast Flow, GE Healthcare) pre-equilibrated in high salt wash buffer (500 mM NaCl, 5% glycerol, 20 mM imidazole, 1 mM βME, 10 mM Tris pH 8.5) for 1 hour at 4°C. Following incubation, resin was pelleted at 1,000 rpm for 1 minute at 4°C, and supernatant was removed. The resin was washed twice in 5 volumes of high salt buffer before being packed into a column and further washed with ∼15 volumes of low salt wash buffer (250 mM NaCl, 10% glycerol, 20 mM imidazole, 1 mM βME, 10 mM Tris pH 8.5). Purified protein was eluted with 2.5 volumes of low salt elution buffer (250 mM NaCl, 10% glycerol, 240 mM imidazole, 1 mM βME, 10 mM Tris pH 8.5) and 2 mM DTT was supplemented into the eluted protein. 1 μg SUMO protease (Ulp1) was added per mg of His6-Smt3-AGO1-PAZ on ice for two hours to cleave His6-Smt3. Cleaved protein was further purified using ion-exchange chromatography (HiTrap Q HP 5 mL, GE Healthcare) on an Akta Explorer 10 system (GE Healthcare). Protein was eluted using a gradient of low- (125 mM NaCl, 5% glycerol, 2 mM βME, 10 mM Tris pH 8.5) to high-salt (1.5 M NaCl, 5% glycerol, 2 mM βME, 10 mM Tris pH 8.5). Fractions containing AGO1 PAZ were pooled, dialyzed into storage buffer (125 mM NaCl, 40 mM imidazole, 2.8 mM βME, 2–5 mM DTT, 6.7% glycerol, 11.7 mM Tris pH 8.5), concentrated to ∼10 mg/mL, and flash frozen and stored at −80°C.

Ab generation, screening, and purification

To generate polyclonal AGO1 Abs, purified AGO1 PAZ domains were dialyzed into immunization buffer (100 mM NaCl, 5% glycerol, 1–2 mM βME in 1X PBS). Immunizations were performed by the Pocono Rabbit Farm and Laboratory. On day 0, rabbits were bled to obtain pre-immunization sera, and injected with 200 μg AGO1 PAZ domain intradermally in Complete Freund’s Adjuvant. Following the first immunization all boosts were done in Incomplete Freund’s Adjuvant (IFA). On days 14 (intradermal, 100 g), 28 (subcutaneous, 100 ug), 56 (subcutaneous, 50 ug), and day 96 (subcutaneous, 50 ug) rabbits were boosted. Rabbits were bled on day 42, 70 and exsanguinated on day 112.

Sera was screened for AGO1 antibodies by custom ELISA. In brief, flat-bottom 96-well plates (Nunc MaxiSorp, Thermo Fisher Scientific) were coated with either 10 ug/mL purified AGO1 PAZ or 250 ug/mL Aag2 lysate in Coating Buffer (100 mM bicarbonate/carbonate buffer, pH 9.6) overnight at 4°C. Wells were washed with Washing Buffer (0.05% Tween-20 in 1X PBS, 3 times) and blocked in Blocking Buffer (2% BSA, 0.05% sodium azide in 100 mM bicarbonate/carbonate buffer, pH 9.6) for 1 hour at room temperature. Sera or primary antibodies (Abcam Drosophila anti-AGO1, Qiagen anti-penta-his) were serially diluted in Sample Diluent (10% Blocking Buffer, 90% Wash Buffer) and incubated for 1 hour at room temperature, shaking. Plates were washed as above, and HRP-conjugated secondary Abs were diluted in Sample Diluent. Secondaries were incubated for 30 minutes at room temperature rocking then washed as above 4 times. Tetramethylbenzidine (Sigma-Aldrich) was added and following sufficient color development, the reaction was stopped by addition of 2 N sulfuric acid. Plates were read on a FLUOstar OMEGA Microplate Reader.

AGO1-reactive Abs were ammonium sulfate precipitated (Pierce Saturated Ammonium Sulfate Solution, Thermo Fisher Scientific) from rabbit 33241 large bleeds and exsanguination sera, which showed the highest specificity, according to manufacturer’s instructions. Abs were dialyzed into 1X PBS and further affinity purified on AGO1 PAZ resin (UltraLink Biosupport, Thermo Fisher Scientific) according to manufacturer’s protocol. Briefly, AGO1 PAZ was dialyzed into 125 mM NaCl, 2.5 mM DTT in 1X PBS and following dialysis sodium citrate was added to a final concentration of 0.6 M. AGO1 PAZ was coupled to resin for 1 hour at room temperature, rocking. Coupling reaction was quenched in 10 resin volumes 1M Tris, pH 8.5 for 2.5 hours at room temperature, rocking. AGO1 PAZ resin was washed in 1X PBS, then 1.0 M NaCl, then 3 times in 1X PBS. Resin was prepared by washing in binding buffer (125 mM NaCl, 0.01 mM βME in 1X PBS) and Abs were diluted into the same buffer. Abs were bound for 2 hours room temperature, rocking, then washed in binding buffer to remove nonspecific Abs. Anti-AGO1 Abs were eluted with 10 resin volumes elution buffer (125 mM NaCl, 0.01 mM βME, 0.1 M glycine, 2% acetic acid) and immediately neutralized in 1M Tris pH 8.5. Antibodies were buffer exchanged into 15% glycerol in 1X PBS, concentrated to 1 mg/mL, flash frozen and stored at −80°C.

Quantitative mass spectrometry

Aag2 cells with or without ectopic 3xFLAG-AGO1 or -AGO2 expression were lysed in 1X PXL (see SDS-PAGE and immunoblot). IPs were performed as outlined below (AGO-CLIP experimental protocol, IP) and proteins were separated by SDS-PAGE and stained with Coomassie (see SDS-PAGE and immunoblot). Bands were cut and quantitative mass spec was performed by The Rockefeller University Proteomics Resource Center.

Luciferase reporter assays

To assay small RNA-mediated silencing in Aag2 cells, we selected AGO2-associated small RNAs abundant in Aag2 cells (> 500 RPM; miR-11894a, miR-11894-novel-a, Aag2-novel-2a, common-novel-2a, common-novel 3a, and common-novel-4a). Perfect, shuffled, 8mer, and bulged target sequences were cloned into the 3′UTR of Renilla Luciferase (Rluc) expressed from pMT-Ren (a kind gift from Ronald van Rij)(van Rij et al., 2006). One perfect or shuffled target sequence was cloned, or two repeated 8mer or bulged target sites were cloned. Oligos were annealed and ligated into the SacII/PmeI-digested pMT-Ren (Table S2) and confirmed by Sanger sequencing (Genewiz). 3e4 cells/well were plated in 96-well plates and co-transfected with 100 ng of each Rluc reporter and 100 ng pMT-GL3, which expresses firefly luciferase (Fluc; normalization control) using Fugene HD (Promega), according to the manufacturer’s protocol. Where appropriate, LNAs (miRCURY LNA miRNA Custom Power Inhibitor, Qiagen) and mimics (custom miRNA miRIDIAN mimic, Horizon) were also co-transfected at a concentration of 50 nM (Table S2). Day 2 post-transfection, media containing CuSO₄ (0.5 mM final concentration) was added to induce pMT expression. ∼18 hours later, cells were harvested and analyzed using the Dual-Luciferase Reporter Assay System (Promega) and a FLUOstar Omega Microplate Reader (BMG LABTECH), according to the manufacturer’s instructions. RLuc signal in each well was normalized to that well’s firefly luciferase (FLuc) signal. Then the RLuc/Fluc ratio was normalized to the ratio obtained with the empty reporter with the appropriate treatment to measure repression. Transfections were performed in triplicate or quadruplicate in at least two independent experiments. Data were analyzed using one-way analysis of variance (ANOVA) with Dunnett’s post hoc test, compared to the empty reporter.

AGO-CLIP experimental protocol

Experimental design

We include 41 mosquito and 56 Aag2 cell line sequencing libraries in this study, based on 41 mosquito and 48 cell line samples. An Aag2 cell sample refers to an independent dish of cells, and a mosquito sample refers to an independent pool of 10–30 mosquitoes. For experiments 3 through 8, each sample was processed as an individual replicate, alongside replicates from the same experiment. For example, in experiment 6 we plated 8 dishes of Aag2 cells, and we separated mosquitoes into 8 tubes, each containing a random pool of 10 mosquitoes (Table S3). Each of these represents a sample; the 16 samples were processed in parallel in separate tubes. For experiments 1 and 2, the same sample was used, and these are technical replicates of one another separated at the library preparation step by splitting the isolated RNA. A library refers to a separate sequencing dataset separated by indices. For standard CLIP libraries, samples from the same experiment were pooled at the sequencing stage. For BrdU CLIP libraries, samples were indexed during reverse transcription and pooled for library generation and sequencing.

Lysate preparation and IP

AGO-CLIP was adapted to mosquito lysates based on the standard linker ligation or BrdU AGO HITS-CLIP protocol (Moore et al., 2014; Weyn-Vanhentenryck et al., 2014). Briefly, Aag2 cells were seeded at a density of 2×10⁶ cells in 150 cm² dishes. At confluency cells were crosslinked in UV 254 nm light in ice-cold 1X PBS on ice (once at 400 mJ/cm² and once at 200 mJ/cm²; Spectrolinker XL-1500, Thomas Scientific). Cells were scraped, pelleted, washed once with ice-cold 1X PBS, flash frozen and stored at −80°C. Whole frozen mosquitoes were homogenized in an N₂ precooled homogenizer for 30 seconds at 30 beats/second (Mixer Mill MM 400, Retsch). Mosquito homogenate was transferred to prechilled dish and crosslinked on ice 3 times at 400mJ/cm². Ice-cold 1X PXL (lysis buffer, see SDS-PAGE and immunoblot) was added to cells or homogenate and lysate was passed through a 26Gx3/8” needle 5 times and incubated on ice for 10 minutes. Homogenate was clarified by centrifugation at 15,000 revolutions per minute (rpm) for 20 minutes at 4°C. Supernatant was recovered, and protein concentration was determined by BCA. ∼1 mg total protein was allocated per sample (∼one 150 cm² dish/sample or 10–30 mosquitoes/sample). Lysates were treated with 30 units RQ1 DNase for 5 minutes at 37°C with shaking (1,100 rpm). Dilutions of RNase I (Thermo Fisher Scientific) were added (high RNase = 10 u/mL, low RNase cells = 0.001 u/mL, low RNase mosquitoes = no RNase) and lysates were treated as with DNase.

Magnetic beads (Dynabeads Protein A, Thermo Fisher Scientific) were pre-bound for 4 hours at 4°C with specific or control antibodies (50uL beads/sample) in Clearer ClickSeal tubes (National Scientific). For AGO1 IP, beads were washed in Ab binding buffer (0.02% Tween-20 in 1X PBS) then 1–2.5 μg AGO1 or control rabbit anti-mouse IgG (Jackson ImmunoResearch) Abs were diluted in Ab binding buffer. For AGO2 IP, beads were first bridged with rabbit anti-mouse Ab in Ab binding buffer at bead binding capacity for 30 minutes nutating at room temperature. Following pre-binding, unbound bridge antibody was removed by washing with Ab binding buffer, then 2.5 μg mouse control Ab (Jackson ImmunoResearch) or 0.5–1.0 mL 9D6 hybridoma supernatant was added. Following Ab binding, beads were washed in lysis buffer and prepared lysates were added. Lysates were incubated with antibodies 4 hours to overnight at 4°C nutating. Following AGO IP, beads were washed in 1X PXL, 5X PXL (high salt wash buffer, same as lysis buffer in 5X PBS, minus protease inhibitor), and 1X Polynucleotide Kinase (PNK) buffer (10 mM MgCl₂, 0.5% NP-40, 50 mM Tris-Cl pH 7.5). In the last wash, beads were changed to a new tube.

3’ linker ligation and autoradiograms

RNA isolated by AGO IP was treated with Alkaline Phosphatase (AP; Roche) on bead in the presence of Recombinant RNasin Ribonuclease Inhibitor (80 units/reaction; Promega) for 20 minutes at 37°C with interval mixing (1,100 rpm every 2 minutes for 15 seconds). Beads were washed in 1X PNK, 1X PNK supplemented with 20 mM EGTA (Boston Bioproducts) without MgCl₂, then 1X PNK. Radiolabeled L32 3’ RNA linker was prepared by adding ³²P-γ-ATP (∼2 μCi/pmol linker; Perkin Elmer) to unphosphorylated PAGE-purified L32 (L32 (-P)) by T4 PNK treatment (NEB) in the presence of RNasin (12 units/reaction). Labelling reactions were incubated for 30 minutes at 37°C, then chased with cold ATP. Un-incorporated ATP was removed using an Illustra Microspin G-25 column (GE Healthcare) according to manufacturer’s instructions. Labeled L32 was ligated to RNA 3’ ends using T4 RNA Ligase (Thermo Fisher Scientific) in the presence of RNasin (80 units/reaction) and 5% PEG8000 (NEB) for 1 hour at 16°C with interval mixing (as in AP treatment). Ligation reaction was chased with cold phosphorylated L32 (40 pmol/reaction) overnight.

After ligation, reactions were washed with 1X PXL, 5X PXL, then 1X PNK. 5′ ends of RNA were phosphorylated by T4 PNK treatment for 20 minutes at 37°C with interval mixing (as in AP treatment). Reactions were washed with 1X PXL, 5X PXL, then 1X PNK and samples were prepared for SDS-PAGE (8% Bis-Tris gel with Midi-Gel Adapters, Thermo Fisher Scientific) as in SDS-PAGE and immunoblot. Electrophoresis (Criterion Cell, Bio-Rad) and transfer (XCell II Blot Module, Thermo Fisher Scientific) were performed at 4°C. Membranes were exposed to autoradiography film at −80°C for 2 hours to 3 days and film was developed.

RNA isolation

Membrane regions corresponding to AGO-associated RNA were cut with scalpels and transferred to nonstick RNase-free tubes (Thermo Fisher Scientific). Protein was digested by treatment with 0.8 mg proteinase K (Roche) in 1X PK buffer (50 mM NaCl, 10 mM EDTA, 100 mM Tris-Cl pH 7.5) for 20 minutes at 37°C with mixing (1,000 rpm). Samples were then treated with 1X PK buffer with 7M urea under proteinase K treatment conditions and RNA was extracted. Briefly, an equal volume of Acid-Phenol:Chloroform, pH 4.5 (Thermo Fisher Scientific) was added and incubated as with proteinase K treatment. Samples were centrifuged at 15,000 rpm for 5 minutes at 4°C, and aqueous layers were transferred to new tubes containing sodium acetate, pH 5.5 (Thermo Fisher Scientific, final concentration 0.1 M), glycoblue (Thermo Fisher Scientific, final concentration 7.5 ug/mL), and 2.5 sample volumes 1:1 ethanol:isopropanol. RNA was precipitated overnight at −20°C, then pelleted at 15,000 rpm for 20 minutes at 4°C. RNA was washed twice in ice-cold 70% ethanol, then air dried and resuspended. Libraries were then prepared from purified RNA by standard linker ligation or BrdU cloning procedures (Weyn-Vanhentenryck et al., 2014).

Standard linker ligation library preparation

RNA was incubated for 5 minutes at 65°C and RL5D3 PAGE-purified 5’ RNA linker was ligated using T4 RNA ligase (20 pmol linker/reaction) for ∼5 hours at 16°C. Reactions were treated with RQ1 DNase for 20 minutes at 37°C, then Acid-Phenol:Chloroform was added. Samples were vortexed, centrifuged at 15,000 rpm for 5 minutes at 4°C, and RNA was isolated as in RNA isolation following proteinase K digestion, above. RNA was incubated for 5 minutes at 65°C and RT was performed using SuperScript III and the DP3-short primer, which contains the 3′ linker (L32) sequence. The same day, the 1^st PCR was performed using AccuPrime Pfx SuperMix (Thermo Fisher Scientific) and DP5 and DP3 primers, which contain 5’ and 3′ linker sequences, respectively. Samples were amplified for 14–18 cycles, then reactions were split into 3–4 reactions that were amplified for an additional 4–12 cycles. For example, 1 sample was split into 4 reactions at 14 cycles; then reactions were further amplified, giving 4 reactions for 1 sample amplified to 14, 18, 22, or 26 cycles. PCR reactions were run on 10% Criterion TBE-Urea Polyacrylamide Gels (Bio-Rad) in a Criterion Cell apparatus with Low Molecular Weight DNA Ladder (NEB) in 1X TBE. Gels were stained in 1X SYBR Gold Nucleic Acid Gel Stain (Thermo Fisher Scientific) and regions corresponding to target RNAs were excised (75–150nt). DNA was extracted using the QIAquick Gel Extraction Kit following the user-developed protocol for extraction of DNA fragments from polyacrylamide gels.

A 2^nd PCR performed was performed using AccuPrime Pfx SuperMix, 5′ primers (AR001-XXX where each primer contains a separate index) containing the standard Read 1 sequencing primer (Illumina), indices (Illumina TruSeq 6nt indices) and the conserved part of the 5′ linker (RL5D3) sequence, and a 3′ primer (MSFP3) containing the 3′ linker (L32) sequence. Reactions were amplified for 4–13 cycles and run on 2% MetaPhor Agarose (Lonza) gels in 1X TAE alongside Low Molecular Weight DNA Ladder. Gels were stained in 1X SYBR Gold Nucleic Acid Gel Stain and regions corresponding to amplified RNA libraries (∼150–250nt) were excised. DNA was purified using the QIAquick Gel Extraction Kit according to the manufacturer’s protocol.

BrdU library preparation

RNA was incubated for 5 minutes at 65°C and RT was performed on isolated RNA using SuperScript III with primers containing 5′ (partial Read 1 sequencing primer) and 3′ linker (L32) sequences, indices (TruSeq 6nt indices) and degenerate barcodes (RT-XT primers), with BrdUTP (Sigma-Aldrich) substituted for dTTP. Following RT, RNAse H (2 units/reaction; Thermo Fisher Scientific) was added and reactions were incubated for 20 minutes at 37°C. Appropriate cDNAs were pooled and un-incorporated BrdUTP was removed using an Illustra Microspin G-25 column. Samples were diluted into 1X BrdU IP buffer (0.3X SSPE, 1 mM EDTA, 0.05% Tween-20) plus 5X Denhardt’s Solution and incubated for 5 minutes at 70°C. Protein G Dynabeads (Thermo Fisher Scientific) were prepared (50 μl/rxn) by washing 3 times in Ab binding buffer (see AGO-CLIP experimental protocol, IP). Beads were prepared for cDNA purification by blocking in 5 bead volumes of Ab binding buffer plus 5X Denhardt’s Solution (Thermo Fisher Scientific), nutating for 1 hour at room temperature. Anti-BrdU Ab (5 μg/reaction; Abcam) was bound to beads in Ab binding buffer plus 5X Denhardt’s Solution for 1 hour at room temperature. Beads were washed in 1X BrdU IP buffer then cDNA samples were added to beads. IPs were nutated for 45 minutes at room temperature then washed with 1X BrdU IP buffer plus 5X Denhardt’s Solution, Nelson Low Salt Buffer (5 mM EDTA, 15 mM Tris pH 7.5) plus 1X Denhardt’s Solution, Nelson Stringent Buffer (5 mM EDTA, 2.5 mM EGTA, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 120 mM NaCl, 25 mM KCl, 15 mM Tris pH 7.5) plus 1X Denhardt’s Solution, then 1X BrdU IP buffer. cDNAs were eluted from beads for 1 minute at 98°C, shaking (1,200 rpm) in BrdU Elution buffer (1.1X BrdU IP buffer), then diluted into 1X BrdU IP buffer plus 5X Denhardt’s Solution.

cDNA purification was performed a 2^nd time as above, starting with treatment of cDNAs for 5 minutes at 70°C through the last IP wash, where BrdU 1X IP buffer was substituted with CircLigase wash buffer (33mM Tris-Acetate, 66mM KCl, pH 7.8). Circularization with CircLigase II ssDNA Ligase (Lucigen) was performed according to manufacturer’s protocol on bead at for 1 hour 60°C with interval mixing (1,300 rpm every 30 seconds for 15 seconds). Following circularization, beads were washed in Nelson Low Salt buffer, Nelson Stringent buffer, then PCR wash buffer (50 mM Tris pH 8.0). cDNA was eluted from beads for 1 minute at 98°C in 1.1X KAPA HiFi Fidelity Buffer (Roche). cDNA was then PCR amplified using KAPA HiFi Polymerase and DP5-PE (Read 1 sequencing primer) and SP3-PE (L32) primers with 0.5X SYBR Green I Nucleic Acid Gel Stain (Thermo Fisher Scientific) in PCR tubes with optically clear caps for monitoring amplification by real-time quantitative PCR. Reactions were removed at ∼200–400 relative fluorescence units and cleaned using Ampure XP beads (Beckman Coulter) according to the manufacturer’s protocol.

Library verification, pooling, and sequencing

Standard linker ligation and BrdU CLIP library sizes and concentrations were analyzed by TapeStation (Agilent). Libraries were multiplexed at equimolar concentrations and prepared for sequencing according to standard Illumina protocols, with 10% PhiX Control v3 (Illumina). Sequencing was performed using the standard Read 1 sequencing primer on single-end runs on HiSeq-2000 or MiSeq platforms (Illumina).

AGO-CLIP bioinformatic analyses

Analysis steps were based on previously published AGO-CLIP analysis protocols and tools (Moore et al., 2015; Scheel et al., 2017; Shah et al., 2017; Zhang and Darnell, 2011) where possible, wrapped in R v3.5.2 (R Core Team, 2018) and executed in RStudio v1.1.463 (RStudio Team, 2020). Sequencing libraries obtained for each sample were analyzed individually, and then concatenated for visualization and to generate data tables. In some cases, the mosquito genome precludes us from using available tools or required us to modify parameters. The scripts developed to analyze the mosquito genome in this study were used to develop a CLIP analysis R package, CLIPflexR < https://kathrynrozengagnon.github.io/CLIPflexR/ >, suitable for custom genomes. The CLIPflexR package allows users to pick and choose tools developed for this paper or previously published tools depending on their preference, in one R-based user interface. In addition to the CLIPflexR package, exact implementation of scripts used for this paper is available: < https://kathrynrozengagnon.github.io/AGOCLIP_2020/ >.

Processing, peak matrix, and PCA

Briefly, reads were unzipped using gunzip and then processed using the FastX Toolkit v0.0.14 (Gordon, 2010) and the CLIP Tool Kit v1.0.7 (CTK)(Shah et al., 2017). Fastq reads were required to have a minimum quality of 20 over 80% of the read (FastX fastq_quality_filter, -Q 33, -q 20, -p 80), and duplicate reads were collapsed (FastX fastx_collapser –Q 33). For standard linker ligation libraries, samples were then split on 5 or 6nt indices (FastX fastx_barcode_splitter, -bol, -mismatches 0). Random barcodes were stripped (CTK stripBarcode, -len 27) and 3′ linker sequences were clipped (FastX fastx_clipper, -l 18 -a GTGTCAGTCACTTCCAGCGG). For BrdU libraries, random barcodes were stripped (CTK stripBarcode, -len 7) and samples were then split on 5′ indices (FastX fastx_barcode_splitter, -bol, -mismatches 0). Next, 5′ indices were trimmed (FastX fastx_trimmer, -f 10) and 3′ linker sequences were clipped (FastX fastx_clipper, -l 18 -a GTGTCAGTCACTTCCAGCGG).

The Ae. aegypti genome was obtained from Vectorbase: <https://www.vectorbase.org/> (Giraldo-Calderón et al., 2015), Ae. aegypti LVP_AGWG (Matthews et al., 2018) AaegL5 chromosomes. The CFAV genome, Galveston strain is available from NCBI (NC_001564.2). Processed reads were mapped to genomes using the Rbowtie2 R package v1.4.0 (--threads 4 -f -N 1 -L 18)(Wei et al., 2018). Mapped read coordinates were converted to bed files for both Ae. aegypti and CFAV genomes; for Ae. aegypti bed files, HOMER v4.8.1 (Heinz et al., 2010) was used to make tag directories (makeTagDirectory, -single -format bed) and find peaks (findPeaks, -o auto -style factor -L 2 -localSize 10000 -strand separate -minDist 50 -size 10 -fragLength 25 -gsize 1278731969). Peak ranges were imported for all samples and merged to a non-redundant set of peaks, then beds were imported and reads overlapping with peaks were counted by sample. Peaks were annotated using the Vectorbase LVP_AGWG strain AaegL5.2 gene set and ChIPseeker v1.18.0 (Yu et al., 2015). To produce a set of high-confidence filtered peaks (Figure 4), we required peaks called in a condition to have a mean 10-fold higher normalized coverage in one AGO antibody set compared to its paired IgG control and that at least 2 samples had a raw read count greater than 10. This set of high-confidence peaks was used to create the annotation pies in Figures 4C–4F. Note this analysis does not integrate any information concerning the small RNA that is targeting AGO. PCA analyses were performed using DESeq2 v1.22.2 (Love et al., 2014) on transformed read counts (variance stabilizing transformation) from the same set of high-confidence peaks for Figures 4G and 4J. Raw read counts for unfiltered peaks are available (GEO accession: GSE157168).

fgsea

Gene ontology (GO) terms for the LVP_AGWG strain AaegL5.2 gene set were obtained by homology mapping (eggNOG-mapper v2)(Huerta-Cepas et al., 2017; Huerta-Cepas et al., 2018). Peaks were ranked by their contribution to PC2 value (which separated AGO1 and AGO2 samples) from PCA loadings analysis and summarized to the associated gene. Fgsea v1.8.0 (Korotkevich, 2019) was performed on PC2-ranked genes and GO terms were mapped using GO.db v3.7.0 (Carlson, 2018). Significant pathways (p-value < 0.05) were collapsed if they contained the same leading edge genes.

miRDeep2

To identify small RNAs miRDeep2 v2.0.1.2 (Friedlander et al., 2012) was run on 3 sets of files. Linker artifacts were removed from processed (see above), uncollapsed 18–30nt reads. First, short reads were then mapped using Rbowtie2 (--threads 4 -f -N 0 -L 18). Short mapped sequences were converted to collapsed fasta and arf files (containing genome coordinates) in mirRDeep2 format (bwa_sam_converter). miRDeep2 was run by inputting the collapsed fasta, the corresponding arf file, AaegL5 genome fasta, mature and precursor Ae. aegypti miRNA sequences, and Drosophila melanogaster mature miRNAs sequences (all mature and precursors were downloaded March 11, 2019 from miRbase)(Kozomara et al., 2018). Second, processed uncollapsed reads were concatenated by sample type (antibody/lysate), mapped with Rbowtie2 as above, and miRDeep2 was run as above. Third, processed uncollapsed reads were all concatenated and a conFiguretxt file was generated assigning a unique 3-letter id to each sample. Reads were mapped using miRDeep2 mapper (Bowtie v1.2.2, -c -d -j -m -l 18 -p) to generate collapsed fasta and arf files for miRDeep2, which was run as above.

We retained novel small RNAs with positive miRDeep2 scores and significant RNAfold values (Lorenz et al., 2011) or those that shared seeds in related species (Anopheles gambiae, Culex quinquefasciatus, or Drosophila melanogaster miRNAs)(Table S4). Unique small RNAs discovered exclusively in IgG samples were removed (196 out of 1130). The remaining 934 unique putative small RNA sequences were mapped using Rbowtie2 (--threads 4 -f -L 18 -k 1000000) against short processed reads with linker artifacts removed. Small RNAs present in 1/3 of specific Ab samples with a raw read count of 10 in at least 1 sample were considered high-confidence novel small RNAs and used for subsequent analyses (230 small RNAs). Counts were normalized to input reads and small RNA IDs were assigned as follows: 1) novel small RNAs that shared a 6mer seed with Ae. aegypti known miRNA families were named with the known miRNA family name, with “novel-a”, “novel-b”, etc., assigned to determine different family members, by small RNA length 2) completely novel small RNA families that did not share a 6mer seed with any known Ae. aegypti miRNA families were randomly assigned a family number (i.e “aae-novel-1”, “aae-novel-2”), with “a”, “b”, etc., assigned to determine different family members, by small RNA length; 3) completely novel small RNA families were ranked by abundance (RPM) in each sample type (AGO1/AGO2, Aag2/mosquitoes) and the top 10 most abundant novel families were assigned a family name in the format “lysate-AGO-novel”, with a number appended according to abundance rank (i.e . “Aag2-AGO1-novel-1”, “common-AGO2-novel-6”), and with “a”, “b”, etc., assigned to determine different family members, by small RNA length. “Common” denotes that these small RNA families were in the top 10 most abundant families in both cells and mosquitoes.

Known miRNA abundances were calculated by mapping known miRNAs to short processed reads with linker artifacts removed (as with novel). Counts were normalized to input reads and high-confidence known mRNAs were filtered as for novel small RNAs, and used for subsequent analyses (102 known miRNAs). As with novel, the top 10 families were ranked by abundance.

Metagene analysis

Bigwig coverage (RPM) by sample type (Ab/lysate) was calculated over all Ae. aegypti 3′UTRs or control regions (3′UTRs shifted 4000nt) using soGGi (style = “percent of region”). For AGO2, signal in control regions was subtracted from specific Ab samples to enable comparison between lysates. For chimera metagene plots, beds with genomic coordinates of remapped downstream target sequences were converted to bigwigs (proportion) and input to soGGi as above. The AaegL5 genome was also searched for all filtered sRNA 6mer matches and coordinates were converted to bigwigs (proportion) and input to soGGi as above.

Chimeras

Analysis for chimeras was done as described previously (Moore et al., 2015; Scheel et al., 2017). Briefly, known and novel small RNAs were mapped (--threads 4 -f -L 18 -k 1000000) to reads that did not map to the AaegL5 genome; reads where more than one small RNA mapped were randomly discarded, keeping only one small RNA. For reads that contained small RNA sequences, mapping coordinates were used to obtain downstream sequences (miR-first chimeras)(Moore et al., 2015). Sequences longer than 18nt were retained and remapped to the AaegL5 genome (--threads 4 -f -N 1 -L 18). For comparisons of AGO1-abundance of individual small RNAs or their chimeras, downstream sequences were required to remap in more than 2 libraries at any genomic location. For all other analyses, specific peaks were considered to be supported by chimeras if chimeras from high-confidence small RNAs (see miRDeep2) were observed in that peak location in any library (Scheel et al., 2017). To obtain chimera abundance for comparison with overall miRNA abundance (see above), counts were normalized to input reads. For novel RNAs, the top 10 targeting small RNAs were ranked by chimera abundance in cells or mosquitoes, in AGO1 and AGO2 datasets. These were combined with the top 10 most abundant small RNAs and named accordingly (see miRDeep2 above), keeping unique small RNA families (in total, 22 for AGO1 and 22 for AGO2).

Pattern searching

Small RNA target sequences were searched within genomic sequences under peaks using the Biostrings R package v2.50.2 (Pagès, 2019). For enrichment of small RNA targets in reads by position in peaks, peaks were loosely filtered (support by one specific Ab sample) because small RNAs searched were filtered stringently (see miRDeep2 above). Control sequences were generated by shifting peaks 4000nt. Peaks and control peaks were resized to 400nt. For AGO1, 6mers of the 10 most abundant known filtered miRNA families in cells or mosquitoes were searched (we excluded novel members of previously described known miRNA families in abundance ranking). For novel miRNA families, 6mers for the 10 most abundant and 10 most targeting novel families were searched in each lysate (13 unique AGO1 families in mosquitoes and 15 in cells). Known miRNA families with similar abundances as the average “top” (most abundant) novel miRNA family abundance were also searched (4 or 6 families, for cells or mosquitoes, respectively). The number of patterns at each position relative to the peak center was normalized by coverage (RPM) at each position. Coverage values were generated over the same peak sets and calculated using the soGGi R package v1.16.0 (Dharmalingam, 2019).

For AGO2, 6mers for the 10 most abundant and 10 most targeting novel esiRNA families from aegypti were searched in each lysate (16 unique families) and miRNAs that preferentially associated with AGO2 (3 unique families) in both cells and mosquitoes were searched. Full-length 18mer targets for esiRNAs in loosely filtered AGO2 peaks were required to be perfectly complementary, with 0 mismatches in nt 1–18 of the esiRNA.

RNAi network map and miRNA-target abundance correlations

For the final RNAi network map (Table S5), all peaks were resized to 70nt (Chi et al., 2009). The high-confidence targets in Table S5 are high-confidence peaks (see Processing, peak matrix, and PCA) in which the 70nt extended peak contained either: 1) a 6mer target sequence of a high-confidence small RNA (see miRDeep2, Pattern searching) or 2) a chimeric read from a high-confidence small RNA (see miRDeep2, Chimeras). These data were also used to link miRNA family abundance to predicted target abundance. Abundances for miRNAs were calculated (see miRDeep2), summed by family, and normalized to input reads. Peaks were loosely filtered (see Pattern searching) and the average abundance of all peaks targeted by each miRNA family was calculated. Only AGO1 associated miRNA families were plotted. The high abundance threshold was set by determining the even log₂ count value (2⁹, 512 RPM) where the sum of the abundance of the miRNAs meeting the cutoff comprised at least 95% of the total miRNA abundance. This cutoff includes 34 Aag2 miRNAs and 48 Ae. aegypti miRNAs.

Repeat features and EVEs

To determine the percentage of reads mapping repeat features, coordinates were obtained from the Vectorbase LVP_AGWG strain AaegL5 repeat features GFF3 and fastas were extracted by repeat class. Fastas for additional EVEs were obtained using published genome coordinates for annotated EVEs from (Aguiar et al., 2020) and the AagL5 genome. Reads from each sample were individually mapped to repeat features or EVEs (Rbowtie2, --threads 4 -f -L 18 -N 1). The percentage of mapped reads was extracted from bams and classes with a p-value < 0.01 and FDR < 2% were considered valid.

Reprocessing of small RNA-seq

Published small RNA-seq from uninfected Aag2 cells was obtained from the following ENA Projects: PRJNA310830 (Miesen et al., 2016); PRJNA272825 (Haac et al., 2015); PRJNA610833 (Ma et al., 2021). Datasets were reprocessed according to standard small-RNA seq processing practices alongside short, uncollapsed CLIP reads (see miRDeep2) for comparison. Reads were mapped to the CFAV genome, Galveston strain (NC_001564.2), the PCLV genome, Aag2-Bristol strain (L-segment: KU936057.1; M-segment: KU936056.1; S-segment: KU936055.1) and the CLY genome, isolate P1-BS2010 (A-segment: JQ659254.1; B-segment: JQ659255.1); all fastas are available from NCBI.

Motif enrichment

De-enrichment of perfect targets was observed in the 40nt region around peak centers. Therefore, we extracted the central 40nt of all perfect AGO2 targets in mosquitoes and cells as fastas. Control fastas were generated by extracting 40nt upstream and downstream regions. DREME v5.1.1 (Bailey, 2011) was used to find significantly enriched motifs and generate a position weight matrix for visualization. Motifs may occur anywhere from 0–80nt from the small RNA target.

goseq GO analysis

GO terms were mapped to the AaegL5.2 gene set as described above (fgsea). Perfect targets were defined as above and imperfect targets were 6mer and chimera supported target genes for the top 21 esiRNA families and the 5 AGO2-associated miRNA families. Perfect and imperfect gene sets were analyzed using the goseq R package v1.34.1 (Young et al., 2010), collapsed by semantic redundancy using REVIGO (Supek et al., 2011) and p-values for all significantly enriched GO-terms (p < 0.05) were -log10 transformed. Transformed p-values < 1.3 were set to 0 and those > 5 were set to 5 to highlight p-values between 0.05 and 0.00001. The top 12 most significantly over-represented unique GO terms are shown for each lysate/target group. GO terms and groups were clustered by hierarchical clustering (distance = “euclidean”, clustering = “complete”).

Target eCDF plots and GSVA

We selected known miRNAs that were highly expressed in the ovaries (> 5000 RPM) but not in other Ae. aegypti tissues (< 1000 RPM in carcasses)(Akbari et al., 2013), and had more than 50 targets. Predicted targets for miRNAs were defined as 3′UTR peaks with predicted 6mers or chimera support. Control gene sets were also defined. First, we selected miRNAs high in the carcass and low in the ovary were selected (< 1000 RPM in the ovaries, > 4000 RPM in the carcass; we reduced this from 5000 RPM because only one miRNA met that criteria). Second, transcripts with high or low expression in the ovary were selected (the top or bottom 50 ovary targets compared to all tissues by fold-change average expression) to ensure that genes were differentially expressed in the ovaries. Paired- and single-end RNA-seq raw counts data was obtained from ENA Project PRJNA236239 (Matthews et al., 2016). and all female tissues were normalized to reads per kilobase million (edgeR v3.24.3)(Robinson et al., 2010) and gene sets were analyzed. For eCDFs, average normalized expression was calculated for sugar-fed or gravid ovaries, or across all tissues. Target sets for each miRNA were plotted and one-sided Mann-Whitney U-tests were performed for significance.

For AGO2, esiRNA expression was quantified as above (see miRDeep2) for the 10 most abundant and 10 highest targeting novel esiRNAs in Ae. aegypti (16 unique families) from published small RNA-seq data (ENA Project PRJNA612346)(Akbari et al., 2013). Targets of these 16 novel esiRNA families and three AGO2-loaded miRNA families were defined by extent of complementarity: non-3′UTR targets were defined as targets with predicted 6mers falling outside 3′UTRs; 6mers were defined as 3′UTR predicted targets; 8mers were defined as predicted 8mer targets in 3′UTRs (we included both “miRNA” type 8mers containing a 3′ adenosine as well as targets fully complementary to the first 8nt of the esiRNA); chimeras were 3′UTRs targets with chimeric reads mapped at target locations; and perfect targets were defined as 0 mismatch, perfect targets (nt 1–18 of the esiRNA). All other genes were used as a “not target” control to determine baseline expression. For eCDFs for AGO2, target expression was calculated as for AGO1, except expression levels of target genes in the ovaries were normalized to their expression levels across all tissues because we compared different gene sets.

For GSVA, RNA-seq raw counts were normalized to counts per million (edgeR) and relative expression was determined on small RNA target or control gene sets (GSVA v1.30.0)(Hänzelmann, 2013).

Visualization

To generate trees to compare AGO isoforms, experimentally determined AGO transcripts were converted to amino acid sequences (< https://web.expasy.org/translate/ >) and aligned to annotated AGO protein sequences using T-Coffee v11.00 (Madeira et al., 2019). Unweighted pair-group method (BLOSUM62) trees were calculated and visualized in Jalview v2.11.10 (Waterhouse et al., 2009). Prism 8 was used for visualizing fgsea results and heatmaps of enrichment of key annotation categories. ImageJ v1.51 was used to prepare immunoblots. RStudio v1.1.463 was used for all other data visualization: scatter, line, bar, and ecdf charts (ggplot2 v3.1.1 and ggrepel v0.8.1)(Slowikowski, 2019; Wickham, 2016), Venn diagrams (eulerr v6.1.0)(Larsson, 2020), consensus motifs (ggseqlogo2 v0.1)(Wagih, 2017), annotation pie charts (ChIPseeker), and heatmaps (pheatmap v1.0.12)(Kolde, 2019). All example tracks were generated by visualizing bedfiles (chimeras, 6mers) or bigwigs (coverage) in IGV v2.9.37 (Thorvaldsdóttir et al., 2012).

Quantification and statistical analysis

Quantification of normalized small RNA and target reads was performed using R as described above. Linear regression and statistical analyses were largely performed in R (fgsea, goseq, GSVA, Mann-Whitney U-tests), with the exception of Student’s t-tests and one-way ANOVAs, which were performed in Prism 8.

Additional Resources

The code used in this study is available at the following link: < https://kathrynrozengagnon.github.io/AGOCLIP_2020/ >. The CLIPflexR package for CLIP analysis: < https://kathrynrozengagnon.github.io/CLIPflexR/ >. Also see the previously published CLIP Tool Kit (CTK)(Shah et al., 2017): < https://github.com/chaolinzhanglab/ctk >. The original HITS-CLIP protocol: < https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4156013/ >. Update of HITS-CLIP including both standard linker ligation and BrdU cloning: < https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3992522/ >.

Supplementary Material

Table S1. Table S1. Mass spectrometry of AGO IPs, related to Figure 1. (provided as an Excel file).

UnitProt accessions and descriptions of filtered proteins identified across all IPs (rIgG, AGO1, 3xFLAG-AGO1, mIgG, AGO2, 3xFLAG-AGO2), their molecular weight in kilodaltons (MW, kDa), #AAs (number of amino acids), and calculated isoelectric point (calc. pI) are also indicated. Average area of the 3 most intense peptides (Area), Mascot protein score (Score), percent protein sequence coverage (Coverage), number of peptide matches (# Peptides) and total number of peptide spectrum matches (# PSM) are indicated for each IP.

Table S4. Table S4. Abundance of novel and known small RNAs, related to Figures 4–7. (provided as an Excel file).

IDs for known and novel high-confidence small RNAs (smallRNA) are indicated, along with their full-length sequences and lengths, 6mer seed (six_mer), and 6mer target sequence (six_mer_target). Normalized counts in reads per million mapped (RPM) by sample type (counts_norm); AGO1 versus AGO2 association in Aag2 cells or whole mosquitoes (log2Aag2AGO1overAGO2, log2aegyptiAGO1overAGO2) are indicated. The same was done for chimeras that contained small RNAs; RPM of chimeras that remapped to the AaegL5 genome (norm_chimera), and AGO1 versus AGO2 association (log2Aag2AGO1overAGO2_chimera, log2aegyptiAGO1overAGO2_chimera) is indicated. Number of libraries where chimeras were present is shown by sample type for each small RNA (BC_chimera). For novel small RNAs, miRDeep2 output parameters are shown (best miRDeep2_score, estimated_probability_smallRNA_is_true_positive, signficant_Randfold, unique precursor_coordinate). Unique estimated probabilities the small RNA is a true positive are shown for each sample in which the small RNA was annotated, separated by “;”. Precursor coordinates are shown in the format “chromosome:start_stop:strand”, and multiple unique precursors are separated by “;”. Individual small RNAs sharing a 6mer seed in Ae. aegypti were grouped into families (aae_smallRNA_family); small RNAs sharing a 6mer with known miRNAs in other species are shown (cqu_related_miRNA_family = Culex quinquefasciatus, aga_related_miRNA_family = Anopheles gambiae, dme_related_miRNA_family = Drosophila melanogaster). Novel = previously un-annotated.

Table S5. Table S5. AGO-CLIP RNAi network map, related to Figures 3–7. (provided as an Excel file).

High-confidence small RNAs linked with high-confidence AGO1 and AGO2 targets in cells and mosquitoes. The small RNA family, a group of small RNAs that share the same 6mer target, is indicated (aae_smallRNA_family), followed by the target type (predicted 6mer, predicted 7mer-A1, predicted 7mer-M8, predicted 8mer, chimera); target peakID in the format “chromosome:start_stop:strand”; target genomic coordinates for the AaegL5 assembly (chromosome, start, end, peak width, strand); target genomic annotation, geneID, and transcriptID; the sample(s) in which this peak was a high-confidence target, separated by “;”; and the sample(s) in which chimeras were observed, separated by “;”, if applicable. Note that peaks may be supported by multiple small RNAs; novel = previously un-annotated.

Table S6. Table S6. All fgsea results for cells and mosquitoes, related to Figure 4. (provided as an Excel file).

Leading edge genes are the genes responsible for observed GO term enrichment (Leading Edge Gene ID). GO terms containing the exact same leading edge gene sets were collapsed (each individual GO term is separated by a comma). The number of genes present in the data for each GO term is indicated (size). The p-value and normalized enrichment score (NES) for each set of GO terms is also indicated, with positive NES values reflecting AGO2-enriched GO terms and negative values indicating AGO1-enriched GO terms.

Table S3 [1]. Table S3. Processing and alignment statistics for all CLIP datasets, related to Figures 3–7. (provided as an Excel file).

Experiment, antibody, lysate, cloning procedure, and index are indicated by sample for all sequencing libraries included in this study. Uncollapsed and collapsed numbers indicate processed reads. Collapsed reads mapping to the Ae. aegypti genome (AaegL5_mapped), and the numbers (AaegL5_remapped_chimeras) and percentages (AaegL5_chimera_percent, normalized to the number of AaegL5 mapped reads) of chimeric reads containing small RNAs where the target portion of the RNA remapped to the AaegL5 genome are shown. Collapsed total reads mapping to the cell fusing agent virus genome (CFAV_mapped), Phasi Charoen -like virus genome (PCLV_mapped), and Culex Y virus genome (CLY_mapped) are also shown. The percentage of total virus-mapped reads (CFAV_percent, PCLV_percent, CLY_percent) and percentage of likely vsiRNAs (18–24nt; CFAV_vsiRNA_percent, PCLV_vsiRNA_percent, CLY_vsiRNA_percent) for each virus is also shown, normalized to the number of AaegL5 mapped reads. PCLV and CLY columns include reads that mapped to all segments.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
anti-Ae. aegypti Ago1 (western, IP, ELISA)	This study	N/A
anti-Drosophila Ago1	abcam	Cat#: ab5070; RRID:AB_2277644
anti-Drosophila Ago1 (clone 1B8)	(Miyoshi et al., 2005) Mikiko Siomi	N/A
anti-Ae. aegypti Ago1	Abmart	Cat#: X-Q16M62-C
anti-pan Ago (clone 2A8)	Sigma-Aldrich	Cat#: MABE56; RRID:AB_11214388
anti-Drosophila Ago2 (clone 9D6; western, IP)	(Miyoshi, et al., 2005) Mikiko Siomi	N/A
AffiniPure Rabbit Anti-Mouse IgG, Fcγ fragment specific (rabbit control Ab for IP, bridge Ab)	Jackson ImmunoResearch	Cat#: 315-005-008; RRID:AB_2340035
ChromPure Mouse IgG, whole molecule (mouse control Ab for IP)	Jackson ImmunoResearch	Cat#: 015-000-003; RRID: AB_2337188
anti-FLAG (clone M2)	Sigma-Aldrich	Cat#: F3165; RRID:AB_259529
anti-Penta-His	Qiagen	Cat#: 34660; RRID:AB_2619735
Anti-BrdU antibody [IIB5]	abcam	Cat#: ab8955; RRID:AB_306886
IRDye® 680RD Goat anti-Mouse IgG Secondary (H+	LI-COR	Cat#: 926-68070; RRID:AB_10956588
IRDye® 800CW Goat anti-Mouse IgG Secondary	LI-COR	Cat#: 926-32210; RRID:AB_621842
IRDye® 680RD Goat anti-Rabbit IgG Secondary	LI-COR	Cat#: 926-68071; RRID:AB_10956166
IRDye® 800CW Goat anti-Rabbit IgG Secondary	LI-COR	Cat#: 926-32211; RRID:AB_621843
Peroxidase AffiniPure Goat Anti-Mouse IgG (H+L)	Jackson ImmunoResearch	Cat#: 115-035-146; RRID:AB_2307392
Goat anti-Rabbit IgG (H+L) Cross-Adsorbed Secondary HRP Antibody	Thermo Fisher Scientific	Cat#: 31462; RRID:AB_228338
Bacterial and Virus Strains
Rosetta (DE3)	Novagen	Cat#: 70954
DH5alpha	The laboratory	N/A
Chemicals, Peptides, and Recombinant Proteins
Schneider’s Drosophila media	Thermo Fisher Scientific	Cat#: 21720024
Leibovitz’s L-15 Media, no phenol red	Thermo Fisher Scientific	Cat#: 21083027
FBS	HyClone, GE Healthcare	Cat#: SH30396.03
L-Glutamine (200 mM)	Thermo Fisher Scientific	Cat #: 25030081
MEM Non-Essential Amino Acids Solution (100X)	Thermo Fisher Scientific	Cat #: 11140050
Tryptose Phosphate Broth solution (29.5 g/L)	Sigma-Aldrich	Cat#: T8159
TRIzol Reagent	Thermo Fisher Scientific	Cat#: 15596026
RQ1 RNAse-free DNAse	Promega	Cat#: M6101
Phusion® High-Fidelity DNA Polymerase	NEB	Cat#: M0530L
Taq DNA Polymerase, recombinant	Thermo Fisher Scientific	Cat#: 10342053
Gibson Assembly Master Mix	NEB	Cat#: E2611L
Fugene HD Transfection Reagent	Promega	Cat#: E2311
cOmplete Proteinase Inhibitor, Mini, EDTA-free	Roche	Cat#: 11836170001
Precision Plus Protein Dual Color Standards	Bio-Rad
NuPAGE LDS Sample Buffer (4X)	Thermo Fisher Scientific	Cat#: NP0007
PBS Blocking Buffer	LI-COR	Cat#: 927-70001
Ni Sepharose 6Fast Flow	GE Healthcare	Cat#: GE17-5318-02
SUMO protease Ulp1	This laboratory	N/A
HiTrap Q HP 5	GE Healthcare	Cat#: GE17-1154-01
2-Mercaptoethanol (55 mM; βME)	Thermo Fisher Scientific	Cat#: 21985023
PureLink RNase A (20 mg/mL)	Thermo Fisher Scientific	Cat#: 12091021
RNase I (10 U/μL)	Thermo Fisher Scientific	Cat#: EN0601
DNase I Solution (1 unit/μL), RNase-free	Thermo Fisher Scientifics	Cat#: 89836
3,3′,5,5′-Tetramethylbenzidine (TMB) Liquid Substrate System for ELISA	Sigma-Aldrich	Cat#: T0440-1L
Ago1 PAZ	This study	N/A
Pierce Saturated Ammonium Sulfate Solution	Thermo Fisher Scientific	Cat#: 45216
UltraLink Biosupport	Thermo Fisher Scientific	Cat#: 53111
Dynabeads Protein A for Immunoprecipitation	Thermo Fisher Scientific	Cat#: 10002D
Dynabeads Protein G for Immunoprecipitation	Thermo Fisher Scientific	Cat#: 10004D
Alkaline Phosphatase	Roche	Cat#: 10713023001
Recombinant RNasin Ribonuclease Inhibitor	Promega	Cat#: N2515
ATP, [γ−32P]- 3000Ci/mmol 10mCi/ml EasyTide, 250 μCi	Perkin Elmer	Cat#: BLU502A250UC
T4 Polynucleotide Kinase (PNK)	NEB	Cat#: M0201L
T4 RNA Ligase (10 U/μL)	Thermo Fisher Scientific	Cat#: EL0021
PEG8000, 50% (from T4 RNA Ligase 1 (ssRNA Ligase))	NEB	Cat#: M0204S
Proteinase K, recombinant, PCR Grade Solution	Roche	Cat#: 3115828001
Acid-Phenol:Chloroform, pH 4.5, with IAA, 125:24:1	Thermo Fisher Scientific	Cat#: AM9720
Glycoblue	Thermo Fisher Scientific	Cat#: AM9515
AccuPrime Pfx SuperMix	Thermo Fisher Scientific	Cat#: 12344040
Low Molecular Weight DNA Ladder	NEB	Cat#: N3233
SYBR Gold Nucleic Acid Gel Stain,10,000X Concentrate in DMSO	Thermo Fisher Scientific	Cat#: S11494
SYBR Green I Nucleic Acid Gel Stain, 10,000X concentrate in DMSO	Thermo Fisher Scientific	Cat#: S7585
MetaPhor Agarose	Lonza	Cat#: 50185
5-Bromo-2′-deoxyuridine 5′-triphosphate sodium salt (BrdUTP)	Sigma-Aldrich	Cat#: B0631
Denhardt’s Solution (50X)	Thermo Fisher Scientific	Cat#: 750018s
CircLigase II ssDNA Ligase	Lucigen	Cat#: CL9021K
KAPA HiFi PCR Kit (250 U)	Kapa Biosystems	Cat#: KK2102
Agencourt AMPure XP, 60 mL	Beckman Coulter	Cat#: A63881
PhiX Control v3	Illumina	Cat#: FC-110-3001
Critical Commercial Assays
FirstChoice RLM-RACE Kit	Thermo Fisher Scientific	Cat#: AM1700
QIAquick Gel Extraction Kit	Qiagen	Cat#: 28704
QIAquick PCR Purification Kit	Qiagen	Cat#: 28106
QIAprep Spin Miniprep Kit	Qiagen	Cat#: 27106
TA Cloning Kit, with pCR2.1 Vector	Thermo Fisher Scientific	Cat#: K202020
SuperScript III First-Strand Synthesis System	Thermo Fisher Scientific	Cat#: 18080051
Pierce BCA Protein Assay Kit	Thermo Fisher Scientific	Cat#: 23225
Deposited Data
Raw CLIP data (fastqs)	This paper	GEO: GSE157168
processed raw count matrix	This paper	GEO: GSE157168
Ago1 and Ago2 experimentally verified sequences	This paper	GenBank: MW035627-MW035631
Experimental Models: Cell Lines
Ae. aegypti Aag2	(Lan and Fallon, 1990) Carla Saleh	RRID:CVCL_Z617
Drosophila melanogaster Schneider 2 (S2)	Thermo Fisher Scientific	Cat#: R69007; RRID:CVCL_Z232
Experimental Models: Organisms/Strains
Female Ae. aegypti (Orlando)	Laura Kramer	N/A
Oligonucleotides
For 5’RACE and cDNA sequencing oligos, see Table S1	This study	N/A
For CLIP oligos, see Table S1	This study	N/A
Recombinant DNA
pSL1180-HR-PUbECFP	Leslie Vosshall	Addgene plasmid Cat#: 47917; RRID:Addgene_47917
pKRG4-3XFLAG-Aag2-Ago1-short	This study	N/A
pKRG4-3XFLAG-Aag2-Ago1-long	This study	N/A
pKRG4-3XFLAG-Aag2-Ago2	This study	N/A
pET28a-His6-Smt3-Ago1-PAZ	(Gu and Rice, 2016) This study	N/A
Software and Algorithms
R (version 3.5.2)	(R Core Team, 2018)	https://www.r-project.org/
R Studio (version 1.1.463)	(RStudio Team, 2020)	http://www.rstudio.com/
CLIPflexR package (version 0.1.19)	This study	https://kathrynrozengagnon.github.io/CLIPflexR/
Herper (version 0.99.0)	The Rockefeller University Bioinformatics Resource Center	https://github.com/RockefellerUniversity/Herper
Rbowtie2 (version 1.4.0)	(Wei et al., 2018)	https://www.bioconductor.org/packages/release/bioc/html/Rbowtie2.html
DESeq2 (version 1.22.2)	(Love et al., 2014)	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
ChIPseeker (version 1.18.0)	(Yu et al., 2015)	https://bioconductor.org/packages/release/bioc/html/ChIPseeker.html
GO.db (version 3.7.0)	(Carlson, 2018)	https://bioconductor.org/packages/release/data/annotation/html/GO.db.html
Biostrings (version 2.50.2)	(Pagès, 2019)	https://bioconductor.org/packages/release/bioc/html/Biostrings.html
soGGi (version 1.16.0)	(Dharmalingam, 2019)	https://bioconductor.org/packages/release/bioc/html/soGGi.html
goseq (version 1.34.1)	(Young et al., 2010)	https://bioconductor.org/packages/release/bioc/html/goseq.html
GSVA (version 1.30.0)	(Hänzelmann, 2013)	https://bioconductor.org/packages/release/bioc/html/GSVA.html
fgsea (version 1.8.0)	(Korotkevich, 2019)	https://bioconductor.org/packages/release/bioc/html/fgsea.html
edgeR (version 3.24.3)	(Robinson et al., 2010)	https://bioconductor.org/packages/release/bioc/html/edgeR.html
eulerr (version 6.1.0)	(Larsson, 2020)	https://cran.r-project.org/package=eulerr
ggplot2 (version 3.1.1)	(Wickham, 2016)	https://ggplot2.tidyverse.org
ggrepel (version 0.8.1)	(Slowikowski, 2019)	https://CRAN.R-project.org/package=ggrepel
ggseqlogo2 (version 0.1)	(Wagih, 2017)	https://CRAN.R-project.org/package=ggseqlogo
pheatmap (version 1.0.12)	(Kolde, 2019)	https://CRAN.R-project.org/package=pheatmap
IGV (version 2.3.97)	(Thorvaldsdóttir et al., 2012)	http://software.broadinstitute.org/software/igv/
Jalview (version 2.11.10)	(Waterhouse et al., 2009)	https://www.jalview.org/
Prism (version 8)	GraphPad Software	https://www.graphpad.com/scientific-software/prism/
miRDeep2 (version 2.0.1.2)	(Friedlander et al., 2012)	https://github.com/rajewsky-lab/mirdeep2
CLIP Tool Kit (version 1.0.7)	(Shah et al., 2017)	https://github.com/chaolinzhanglab/ctk
HOMER (version 4.8.1)	(Heinz et al., 2010)	http://homer.ucsd.edu/homer/index.html
FastX Toolkit (version 0.0.14)	(Gordon, 2010)	http://hannonlab.cshl.edu/fastx_toolkit/
eggNOG-mapper (version 2)	(Huerta-Cepas et al., 2017; Huerta-Cepas et al., 2018)	http://eggnog-mapper.embl.de/
T-Coffee (version 11.00)	Madeira et al., 2019	https://www.ebi.ac.uk/Tools/msa/tcoffee/
REVIGO	(Supek et al., 2011)	http://revigo.irb.hr/index.jsp
DREME (version 5.1.1)	(Bailey, 2011)	http://meme-suite.org/tools/dreme
Chimera R scripts (version 0.1.18)	(Moore et al., 2015; Scheel et al., 2017) This study	https://kathrynrozengagnon.github.io/CLIPflexR/
ImageJ (version 1.51)	NIH	https://imagej.nih.gov/ij/
Other
Ae. aegypti LVP_AGWG AaegL5 chromosomes	(Giraldo-Calderón et al., 2015; Matthews et al., 2018)	https://www.vectorbase.org/
Ae. aegypti LVP_AGWG strain AaegL5.2 gene set	(Giraldo-Calderón, et al., 2015; Matthews, et al., 2018)	https://www.vectorbase.org/
Ae. aegypti LVP_AGWG strain AaegL5 repeat features	(Giraldo-Calderón, et al., 2015; Matthews, et al., 2018)	https://www.vectorbase.org/
Aedes aegypti small RNAseq	(Akbari et al., 2013)	Table S19;
Aedes aegypti mRNA seq	(Matthews, et al., 2018)	NCBI SRA: PRJNA236239
mature miRNAs (Aedes aegypti, Anopheles gambiae, Culex quinquefasciatus, Drosophila melanogaster)	(Kozomara et al., 2018)	http://www.mirbase.org/
HITS-CLIP protocol	(Moore et al., 2014)	https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4156013/
BrdU and standard linker ligation paper	(Weyn-Vanhentenryck et al., 2014)	https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3992522/
CLEAR-CLIP paper	(Moore, et al., 2015)	https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4674787/
Custom R scripts used in this study	This study	https://kathrynrozengagnon.github.io/AGOCLIP_2020/

Highlights.

Established AGO-CLIP in Ae. aegypti and CLIPflexR, a universal CLIP analysis package

AGO1 and AGO2 RNAi network maps predict tissue-specific target expression

AGO2 binding of 3′UTRs, transposons & endogenous viral elements is context-dependent

Mosquito AGO2 can repress imperfect, 3′UTR targets in an AGO1-like fashion