A Drosophila RNAi library targeting RNA binding domain-containing proteins

The Drosophila RNAi Screening Center (DRSC) and others have developed robust approaches for RNAi reagent and library design (Mohr 2014)). We applied these approaches to production of a focused library targeting genes that encode RNA binding domain-containing proteins. To do this, we first generated a list of known and predicted Drosophila RBPs using a combination of literature and database mining and expert curation, and then generated the reagent library (see Materials and Methods). Features of the dsRNA library include: design of dsRNAs with minimal predicted OTEs using SnapDragon (http://www.flyrnai.org/snapdragon); coverage with two or more unique dsRNAs per gene; tracking of dsRNA production and quality analysis in our database; and inclusion of standard dsRNA controls on each 384-well assay plate. To minimize edge effects (i.e., reduced viability or other data anomalies frequently observed at assay plate edges), dsRNAs were excluded from the outermost two wells of each plate. We included two positive controls: dsRNA targeting thread (Diap1), which results in cell death, and dsRNA targeting Rho1, which leads to the appearance of large, binucleate cells. We also included three negative controls, dsRNAs targeting GFP and LacZ, neither of which is present in wild-type Drosophila cells, and wells with an equal volume of water (no dsRNA). The layout of control dsRNAs on each plate can be viewed at http://www.flyrnai.org/DRSC-LAY.html and the layout of all dsRNAs in the library can be downloaded from http://www.flyrnai.org/DRSC-SUB.html.

Screen for changes in total ATP levels using the RBPs library

We next used Promega Cell Titer Glo to measure total ATP levels after 5 days of incubation of S2R+ cultured cells in three replicates of each unique assay plate in the library. Full raw and analyzed data are available at the DRSC FlyRNAi.org database (Flockhart et al. 2012). The distribution of Z-scores for dsRNAs in the screen is shown in Figure 1. An interactive view of the graph as well as the data are freely available at https://plot.ly/∼semohr/60. Genes that might be considered primary screen "hits" (positive results) from the total ATP levels screen based on commonly applied criteria are summarized in Table 1, Table 2, and Table 3. By Z-score, the strongest hit for "high ATP levels" is Argonaute-1 (AGO1), an Argonaut family protein required for miRNA maturation (Okamura et al. 2004). The strongest hit for "low ATP levels" is hoi-palloi (hoip), which encodes a ribosomal protein and has been associated with neuronal mutant phenotypes in Drosophila (Prokopenko et al. 2000). The opposing results obtained with dsRNAs targeting AGO1 and AGO2 (Table 1) are consistent with the distinct roles of AGO1 and AGO2 in small RNA biogenesis (Okamura et al. 2004; Okamura and Lai 2008; Zhou et al. 2008). The result with AGO1 suggests that one or more microRNAs (miRNAs) might normally dampen growth and/or viability of S2R+ cells, consistent with roles identified for some miRNAs in signaling and cancer (Hagen and Lai 2008; Jackstadt and Hermeking 2015). In the case of AGO2, suppression of the AGO2-mediated endogenous small interfering RNA (siRNA) pathway might result in increased mobilization of transposons (Czech et al. 2008), perhaps accounting for the observed compromise in cell viability, and/or might result in detrimental changes to metabolism or stress responses (Lim et al. 2011).

Table 1 comprises a conservative list of high-confidence hits; Table 2 and Table 3 add additional lower-confidence hits that might be interrogated in follow-up studies and/or compared with other screen data. Each of the reagents in Table 3 has no predicted off-targets and is predicted to target all isoforms of the gene (Hu et al. 2013). Moreover, modENCODE cell expression data for S2R+ cultured cells suggest that most of these genes are normally expressed in S2R+ (Booker et al. 2011; Lee et al. 2014). Thus, we predict that most of these are "single hits" due to a false-negative result with the other dsRNAs targeting the same genes rather than due to false-positive discovery with the positive dsRNA. To further improve library quality, we plan to add additional unique dsRNAs for those cases in which two different dsRNAs targeting the same gene do not give comparable results (i.e., for a subset of genes in Table 2 and all genes in Table 3). The complete set of Z-scores as well as modENCODE results regarding expression in S2R+ cells (Booker et al. 2011; Lee et al. 2014) are available in supporting information, Table S1. Moreover, we have made raw and analyzed data available via a variety of online repositories as summarized at http://www.flyrnai.org/DRSC-RBP_data.php.

High-throughput imaging using the RBPs library

To further characterize the RBP library, we stained cells with DAPI and FITC fluorescence-conjugated Phalloidin to visualize nuclei and filamentous actin, respectively, in paraformaldehyde-fixed S2R+ cells. The cells were then imaged at 20× and 60× using a fluorescence confocal imaging system. As expected, Rho1 serves as an effective positive control for image-based assays, as treatment with Rho1 dsRNA results in a binuclear phenotype readily detectable using a DNA dye and any marker that defines the cell body. These cells also appear larger. Also as expected, few cells are detected in images corresponding to thread (Diap1) dsRNA control wells. Consistent with the total ATP assay results (Table 1), few cells are detected in images corresponding to wells treated with dsRNA targeting AGO2 or hoip.

Availability of RNAi screen data in the form of text or numbers is facilitated by public databases including FlyRNAi, GenomeRNAi, and NCBI PubChem BioAssays (Flockhart et al. 2012; Gilsdorf et al. 2010; Wang et al. 2014). Ideally, researchers will also be provided with access to baseline image data, e.g., to help identify reagents with gross effects on cell morphology. However, high-throughput, high-content image data sets are large both in total size and in terms of the total number of individual images, presenting significant challenges to image data management. As a result, making images easily available for search and view online has been seen as a significant challenge. Although solutions arising from within the biological community such as the Online Microscopy Environment OMERO platform (Allan et al. 2012) hold strong promise for making image data public online, the solutions offered to date are not easy to implement and support, requiring significant expertise and infrastructure. As a result, although it is relatively easy to share complete sets of screen image data, such as for automated analyses (e.g., because an entire image data set can be shared using a file transfer protocol or on an external hard drive), making it possible for researchers to easily view images associated with a specific subset of reagents has remained a barrier. To find a near-term, feasible solution to making a set of baseline image data publically searchable and viewable online, we chose to make images available at flickr.com. Advantages of Flickr include that the site has been around for more than 10 years, it offers 1 terabyte of free image storage, and it allows sharing of images under a Creative Commons license agreement, the same type of agreement used by open access journals. More than 200,000 images have been deposited with the user account drsc_lab at Flickr.com and can be viewed at https://www.flickr.com/photos/132735911@N05/. For quick reference, example images from wells treated with dsRNAs targeting Rho1, thread (Diap1), AGO2, or hoip are organized as “albums” within the drsc_lab collection (https://www.flickr.com/photos/132735911@N05/albums). In addition, a searchable lookup table is available at http://www.flyrnai.org/DRSC-RBP_data.php. For additional information on how to navigate the image data resource see Data availability in Materials and Methods.