An accelerated miRNA-based screen implicates Atf-3 in Drosophila odorant receptor expression
The Drosophila olfactory system is highly stereotyped in form and function; olfactory sensory neurons (OSNs) expressing a specific odorant receptor (OR) always appear in the same antennal location and the axons of OSNs expressing the same OR converge on the same antennal lobe glomeruli. Although some transcription factors have been implicated in a combinatorial code specifying OR expression and OSN identity, it is clear other players remain unidentified. To mitigate some of the challenges of genome-wide screening, we propose a two-tiered approach comprising a primary “pooling” screen for miRNAs whose tissue-specific over-expression causes a phenotype of interest followed by a focused secondary screen using gene-specific RNAi. Since miRNAs down-regulate their target mRNAs, miRNA over-expression phenotypes should be attributable to target loss-of-function. Since miRNA-target pairing is sequence-dependent, predicted targets of miRNAs identified in the primary screen are candidates for the secondary screen. Since miRNAs are short, however, miRNA misexpression will likely uncover non-biological miRNA-target relationships. Rather than focusing on miRNA function itself where these non-biological relationships could be misleading, we propose using miRNAs as tools to focus a more traditional RNAi-based screen. Here we describe a proof-of-concept miRNA-based screen that uncovers a role for Atf3 in the expression of the odorant receptor Or47b.
The transgenic RNAi fly stock libraries (e.g., the Vienna Drosophila RNAi library (Dietzl 2007) and the Transgenic RNAi Project (TRiP)) have been a tremendous boon to the Drosophila community because they permit tissue-specific knockdown of almost all genes in the genome. These resources permit genome-wide screens for genes associated with almost any phenotype of interest. Unfortunately, the sheer size of these libraries—more than 22,000 stocks in the case of the Vienna library—means performing such screens remains labor-intensive and tedious. In this paper, we describe our development of a two-tiered screening protocol comprising an initial pooling screen using miRNA over-expression that generates a list of candidate genes involved in a phenotype of interest and a secondary screen using gene-specific RNAi that narrows this list of candidates to the responsible target gene(s). We suggest that this protocol can sometimes accelerate the identification of novel genes involved in a broad range of phenotypes.
MicroRNAs are short, endogenous, single-stranded RNA molecules that act in the context of the miRISC protein complex to either inhibit translation or induce the degradation of target mRNAs (Bartel 2004). Since the miRNA-target mRNA relationship is determined primarily by a short seed sequence at the 5’ end of each miRNA (Lewis 2003, Lai 2002), the complement of which may occur in multiple copies scattered over the genome, many miRNAs are capable of down-regulating multiple targets. The relationship between a miRNA seed sequence and its complements in the open reading frames and 3’-untranslated regions (3’-UTRs) of target mRNAs spurred the development of bioinformatic algorithms that convert mature miRNA sequences into lists of potential mRNA targets (Rajewsky 2006). These lists of candidate targets, however, are plagued by large numbers of false positives because the algorithms that generate them can fully account for neither the precise spatial and temporal patterns of miRNA and target mRNA expression nor target site availability. In other words, a miRNA may be capable of down-regulating a particular target and never actually do so, either because the two are never simultaneously expressed in the same tissue or because RNA-binding proteins or RNA folding render the target site inaccessible. It also follows that miRNA over-expression in arbitrary tissues using the binary GAL4/UAS expression system would likely lead to non-biological miRNA-target mRNA pairings. Rather than seeing these pairings as a potential drawback of using a library of UAS-miRNA stocks, we expect they can be useful as part of a two-tiered screening system.
We previously generated a library of 131 UAS-miRNA fly stocks that permit tissue-specific over-expression of 144 Drosophila miRNAs (Suh 2015a). In this study, we sought to use these UAS-miRNA stocks to validate the concept of a two-tiered miRNA-based screen in the Drosophila olfactory system.
The olfactory sensory neurons (OSNs) of adult Drosophila are housed in porous hair-like sensilla that cover the paired antennae and maxillary palps. Olfactory sensilla are divided into 3 main classes by their shape and 17 subclasses by their odor response profiles (Couto 2005). The odor response profile of an OSN is determined by its expression of the obligatory olfactory co-receptor Orco and one or very few of the adult odor-specific odorant receptors (ORs) (Vosshall 2007). The spatial arrangement of the 17 subclasses of adult olfactory sensilla on the antenna, the arrangement of the OSNs themselves, the precise pattern of OR expression, and the wiring of the antennal OSNs into the appropriate glomeruli of the antennal lobe are all highly stereotyped from fly to fly, indicating well-orchestrated developmental control of every step in the process.
Jafari et al. reported the results of a large-scale RNAi screen that identified seven transcription factors, permutations of which determine the odorant receptor expressed by each population of olfactory neurons in the adult Drosophila antenna. Despite the success of their screen, Jafari et al. extrapolated from the complexity of the fly olfactory system and estimated that at least three more unidentified transcription factors are likely part of the combinatorial code that determines OR expression (Jafari 2012).
In their screen, Jafari et al. combined the Peb-GAL4 driver line, which is strongly expressed in peripheral sensory neurons including the antennal olfactory neurons, with a pair of OR promoter fusions (i.e., Or47b and Or92a) to a membrane-tethered GFP that act as reporters of OR expression. We obtained these lines and by crossing them to our library of UAS-miRNA stocks we were able to identify miRNAs whose over-expression eliminates Or47b expression, Or92a expression, or both. We chose to proceed with the miRNAs that affect Or47b expression (i.e., bantam, miR-2a-2, miR-33, miR-263a, miR-308, miR-973/974, and miR-2491). We then used existing bioinformatic tools to generate lists of their putative mRNA targets, compare the lists for overlap, and define a short list of candidate genes for a small follow-up RNAi screen. In this follow-up screen, we identified a previously unknown role for Activating transcription factor 3 (Atf3) in the expression of Or47b.
Three other collections of UAS-miRNA stocks published recently were presented as tools either for identifying novel miRNA functions (Bejarano 2012, Schertel 2012) or for use in the context of screens for modifiers of existing developmental phenotypes (Szuplewski 2012). Rather than using the UAS-miRNA lines we generated (Suh 2015) to study miRNA biology, we asked whether they could also be used as a tool permitting a “pooling” pre-screen designed to limit the focus and thus accelerate a secondary, but more traditional RNAi-based loss-of-function screen. In such a scheme, after identifying miRNAs whose over-expression induces a phenotype of interest, bioinformatic target prediction provides a list of candidate genes for a follow-up RNAi screen. Since miRNAs inhibit the transl