microRNAs (miRNAs) are small noncoding RNAs that regulate gene expression post-transcriptionally. Prior studies have shown that they regulate numerous physiological processes critical for normal development, cellular growth control, and organismal behavior. Here, we systematically surveyed 134 different miRNAs for roles in olfactory learning and memory formation using “sponge” technology to titrate their activity broadly in the Drosophila melanogaster central nervous system. We identified at least five different miRNAs involved in memory formation or retention from this large screen, including miR-9c, miR-31a, miR-305, miR-974, and miR-980. Surprisingly, the titration of some miRNAs increased memory, while the titration of others decreased memory. We performed more detailed experiments on two miRNAs, miR-974 and miR-31a, by mapping their roles to subpopulations of brain neurons and testing the functional involvement in memory of potential mRNA targets through bioinformatics and a RNA interference knockdown approach. This screen offers an important first step toward the comprehensive identification of all miRNAs and their potential targets that serve in gene regulatory networks important for normal learning and memory.
INVOLVED in post-transcriptional gene regulation, microRNAs (miRNAs) are a class of small noncoding RNAs (Bushati and Cohen 2007). Prior studies have shown that they serve numerous biological processes, ranging from development to tumorigenesis (Esquela-Kerscher and Slack 2006; Kloosterman and Plasterk 2006; Krützfeldt and Stoffel 2006; Chang and Mendell 2007). miRNAs are transcribed as primary miRNAs (pri-miRNAs) from isolated genes or the introns of protein-coding genes (“mirtrons”) (Filipowicz et al. 2008). miRNAs are under regulatory influences similar to protein-coding genes (Krol et al. 2010). The pri-miRNAs are then cleaved into precursor miRNAs (pre-miRNAs) by the microprocessor Drosha/Pasha protein complex and transported into the cytoplasm by Exportin5 where they mature through the Dicer/Loquacious protein complex into ∼21- to 24-nucleotide miRNA hairpins. These hairpins are subsequently assembled with Argonaute-containing protein complexes that bind to specific sequences on target messenger RNAs (mRNAs) using primarily a 2-to 8-nucleotide seed region (Bartel 2009). The small size of the target site allows many mRNAs to be recognized and coregulated by each individual miRNA (Bartel 2009). The miRNA complex, once bound, induces post-transcriptional silencing by translational repression and/or mRNA degradation (Filipowicz et al. 2008; Bazzini et al. 2012; Djuranovic et al. 2012).
One important biological process that is understudied relative to miRNA function is learning and memory formation. Among the several known epigenetic processes that allow the nervous system to adapt to environmental signals, the miRNA system is thought to provide relatively rapid and analog control in both time and space over the expression of genomic content (Kosik 2006; Costa-Mattioli et al. 2009; McNeill and Van Vactor 2012; Wang et al. 2012). To illustrate this point, consider first another fascinating and important epigenetic system: transcriptional control through chromatin changes. Enduring transcriptional regulation through chromatin changes affords a way for neurons to effect long-term, cell-wide, adaptive changes in state (Kramer et al. 2011; Gräff and Tsai 2013; Zovkic et al. 2013). However, this epigenetic system offers limited ability to control the rapid delivery of genomic material in space, since its speed is limited by the requirement for transcription. In addition, newly synthesized mRNAs may be directed to all compartments of the neuron. The miRNA system operates on already synthesized RNAs, potentially altering expression in specific neuronal compartments including axons, dendrites, synapses, and growth cones (McNeill and Van Vactor 2012; Cajigas et al. 2012). miRNAs are thus ideally positioned to dynamically regulate gene expression across the spatial dimensions of the neuron. These considerations stress the importance of gaining a broad appreciation of which miRNAs are involved in the regulation of neural circuit function for memory formation.
An important second way by which miRNAs may alter adult learning and memory is through neurodevelopmental roles (Xu et al. 2010; Im and Kenny 2012; Saab and Mansuy 2014). Dysregulation of miRNA expression has been described in several human neurodevelopmental diseases (Martino et al. 2009; Xu et al. 2010; Feng and Feng 2011; Nowak and Michlewski 2013). For example, Xu et al. (2008) demonstrated that miR-124a ectopic expression decreases dendritic branching, an effect rescued by the loss of dFMRP expression. In addition, miRNAs in circulating blood are candidate biomarkers for early phases of neurological disease (Wang et al. 2008; De Smaele et al. 2010; Absalon et al. 2013). Moreover, miRNAs may offer promise as therapeutics for disease states using multiple approaches (Sayed and Abdellatif 2011; Im and Kenny 2012). These facts illustrate the importance of first identifying all the miRNAs that are involved in learning or memory formation.
Here, we present the results of a large screen for miRNA functions in olfactory classical conditioning of Drosophila. Olfactory classical conditioning is a type of learning that has been well-studied in flies (Davis 2011; Kahsai and Zars 2011). In this paradigm, an odor [the conditioned stimulus (CS)] is paired with a series of electric shocks [the unconditioned stimulus (US)], creating an association between the two stimuli that behaviorally leads to strong aversion of the odor CS by the fly. Multiple brain regions and neuron types have been shown to be involved in olfactory learning, including projection neurons (PNs) of the antennal lobe, mushroom body neurons (MBn), the dorsal paired medial neurons (DPMn), dopaminergic neurons (DAn), and others (Davis 2005, 2011; Busto et al. 2010). Thus, the rapid regulation of gene expression by miRNA function is predicted to occur in one or more areas of the fly brain that have been shown to mediate olfactory memory formation.
We tested the potential involvement of 134 miRNAs in intermediate-term memory (ITM) by silencing them individually through development and adulthood using specific complementary oligonucleotides called miRNA sponges (miR-SPs), (Ebert et al. 2007; Loya et al. 2009; Fulga et al. 2015). Our study focused on ITM since changes in performance at this time point capture roles for miRNAs in both short-term memory and ITM as well as acute or developmental roles. Our results identified several different miRNAs important for olfactory memory formation. Surprisingly, competitive inhibition of some miRNAs decreased memory formation, while inhibition of others increased memory formation. Our results offer a broad initial foundation for further analysis of miRNA function in memory formation.
Materials and Methods
Drosophila melanogaster were raised on the standard medium at room temperature. Crosses were kept at 25° with 70% relative humidity and with a 12 hr dark/light cycle. For the first screen, we crossed elavc155-gal4 virgin females in a wCS10 genetic background with either uas-scramble-miR-SP or uas-miR-SP males in a w1118 genetic background (Loya et al. 2009; Fulga et al. 2015). The sponge constructs and derivative fly lines are described in Fulga et al. (2015). The same uas-scramble-miR-SP (TTAGAATTTAAACCTCACCATGA) control (either attP2 or attP40 insertion) was used with all miR-SP lines. For the secondary screen, we crossed uas-miR-SP males with virgin females from either the elavc155-gal4 or wCS10 lines. The gal4 drivers that we used in this study were elavc155-gal4 (Lin and Goodman 1994), GH146-gal4 (Stocker et al. 1997), MZ604-gal4 (Ito et al. 1998; Tanaka et al. 2008), OK107-gal4 (Connolly et al. 1996), c316-gal4 (Waddell et al. 2000), TH-gal4 (Friggi-Grelin et al. 2003), Or83b(orco)-gal4 (Kreher et al. 2005), NP2492-gal4 (Tanaka et al. 2008), Gad-gal4 (Ng et al. 2002), Ddc-gal4 (Li et al. 2000), Cha-gal4 (Kitamoto 2001), n-syb-gal4 (Pauli et al. 2008), and uas-dicer 2 (Dietzl et al. 2007). RNA interference (RNAi) lines were obtained from the KK library of the Vienna Drosophila RNAi Center (VDRC) (Dietzl et al. 2007).
One- to 4-day-old flies were used for the behavioral experiments. Flies were collected ∼24 hr prior to conditioning and transferred into fresh food vials ∼30 min before training to adapt to the conditions of the behavioral test room (dim red light, 25°, ∼75% humidity). Groups of ∼65 flies received the standard aversive olfactory conditioning that was previously described (Beck et al. 2000). For conditioning, flies were placed into a tube containing a copper grid. There, they received a sequence of two odor stimuli separated by 30 sec of air. The first 1 min of odor (CS+) stimulus was paired with 12 pulses of a 90-V electric shock (US). The second 1 min of odor stimulus was not associated with electric shocks and thus constituted a nonconditioned stimulus (CS−). The odorants used were benzaldehyde (BEN) and 3-octanol (OCT). They were diluted in mineral oil at concentrations of ∼0.05 and ∼0.2%, respectively. Odorant concentrations were varied slightly to obtain a neutral distribution between the two odors among naive flies. Odors were delivered to the flies in an air stream (rate of ∼400 ml/min) produced by bubbling pressurized air through mineral oil laced with odorant. After conditioning, flies were tapped back into their food vials and tested 3 hr later for memory retention. Memory retention was tested by allowing flies to choose for 2 min between an arm in a T-maze containing the CS+ odor and an arm with the CS− odor. To avoid naive bias for one odor, two groups were trained and tested simultaneously. A first group was trained using BEN as the CS+ paired with the US and then OCT not paired (CS−) with the unconditioned stimulus. The second group was trained with the alternative combination with BEN as CS− and OCT as CS+. Each group (60–70 flies) tested provided a half-performance index (half-PI): half-PI = [(number of flies in CS− arm) – (number of flies in CS+ arm)/(number of flies in both arms)]. A final PI was calculated by averaging the two half PIs.
To test odor avoidance, naive flies were allowed to freely distribute for 2 min between the two arms of a T-maze with an odor stream on one side and a non-odorized air stream on the other. To test shock avoidance, naive flies were allowed to distribute for 2 min between the two arms of a T-maze containing copper grids (same as used for training above) with only one side being electrified. An avoidance index was computed with the following formula: [(number of flies in neutral arm) – (number of flies in odor/shocked arm)/(total number of flies in both arms)].
For the primary screen, 3-hr memory performance of each elavc155-gal4/+>uas-miR-SP/+ genotype was compared to a control elavc155-gal4/+>uas-scramble-miR-SP/+ genotype. Two successive and independent statistical approaches were utilized to select putative hits. First, the scramble-miR-SP genotype score was compared to the miR-SP genotype using a two-tailed, two-sample Student t-test. We chose to use the Student t-test because there is no minimum for sample size and performance indices within a line are known to follow a normal distribution (Tully et al. 1994). We also found that the performance indices of the control elavc155-gal4/+>uas-scramble-miR-SP/+ flies in this study followed a normal distribution (n = 280, μ = 0.48, σ = 0.13, Kolmogorov–Smirnov test, P > 0.1). The structure of the screen using successive steps would also eliminate false positives selected initially. Second, the average of all the scramble-miR-SP scores across the primary screen was used as a population control value to account for day-to-day variability of scramble-miR-SP scores inherent in any behavioral study. The performance of the miR-SP flies was compared to the population control value using a two-tailed, one-sample Student t-test. Significance was set for P-values <0.05, and, when reached, we assigned a score of 1 to the miR-SP. The P-values <0.1 were treated as a trend to minimize false negatives from using a fewer number of biological replicates (n = 4) compared to what is often needed (n = 6–8) in more focused studies. For trends, we assigned a score of 0.5. Lines exhibiting no effect were deemed “neutral” with a score of 0. The scores obtained using the two statistical approaches were then summed and the lines ranked according to their scores. When two inserts (attP40 and attP2) were available for a given miR-SP, both lines were tested by different experimenters. We later combined the scores obtained for the two miR-SP inserts and ranked the lines according to those scores. The maximum score obtainable was 4. Lines scoring >0.5 were retained for the secondary screen.
For the secondary screen, the same two statistical approaches were used with six biological replicates. Each elavc155-gal4/+>uas-miR-SP/+ genotype performance score was compared to the corresponding +/+>uas-miR-SP/+ genotype score. When available, the attP40 and attP2 inserts were tested by different experimenters. We subsequently combined the scores from the primary and secondary screens. The maximal score obtainable was 8, and we retained lines with a score of ≥4 for the third and last screen.
Excel Stat and Prism were used for data analysis. Performance indices (PIs) followed a normal distribution (Figure 1, B and C), so two-tailed, Student t-tests were used to compare the two groups. To compare one group to the population mean value, a one-sample, two-tailed Student’s t-test was used. For multiple group comparisons, ANOVA followed by Bonferroni post-hoc tests was used. Proportions were compared using a χ2 test. Distributions were compared using the Kolmogorov–Smirnov or the D’Agostino and Pearson omnibus normality test when possible. Correlation was assessed with Pearson or Spearman correlation coefficients depending on normality of the samples. Significance was set at α = 0.05.
A genetic screen targeting 134 miRNAs identified at least five that are involved in the biology of memory formation
We tested the involvement of 134 different miRNAs in the biology of olfactory classical conditioning by attenuating their function with a library of transgenic miRNA sponges (miR-SPs, Supporting Information Figure S1, A–C). We anticipated that this behavioral screen would produce substantial variability in memory scores leading to significant numbers of false positives and negatives in the initial screen for several reasons. First, behavioral assays are inherently variable and require extensive replication and statistical analyses to identify bona fide differences. Second, because of the significant labor involved in a behavioral screen, we limited the testing in the initial screen to four biological replicates for each line. Third, the miR-SP transgenes produce a partial loss of function rather than a complete loss of function, since the mechanism of competitive inhibition requires hybridization of the miR-SPs to their target miRNAs (Figure S1C). The extent of hybridization and inhibition depends on numerous factors, including the relative expression level of the miRNA targets and the miR-SPs, the efficiency of stable hybridization of the miR-SPs to endogenous miRNAs in the cellular context, and the possible existence of compensatory mechanisms. Fourth, miRNAs provide analog modulation of the expression of their target mRNAs (McNeill and Van Vactor 2012) so that even a strong knockdown of an miRNA might have only modest effects on the expression level of its mRNA targets. Thus, the screen was extraordinarily demanding in requiring the identification of authentic hits using a partial loss-of-function approach of a regulatory mechanism that might influence the inherently variable readout of conditioned behavior.
Because of these constraints, we adopted a stepwise and liberal approach for selecting putative hits, initially accepting lines with significant differences from the control but also those that exhibited a trend (0.05 < P-value < 0.1). The miR-SP library contains lines with uas-miR-SP transgenes inserted at the attP40 site on the second chromosome and lines with the same uas-miR-SP transgenes inserted at the attP2 site on the third chromosome (Figure S1A). We tested both the attP40 and the attP2 insertions for each miR-SP. Individual miR-SPs were expressed in the developing and adult nervous system using the neuron-specific gal4 driver, elavc155-gal4. Homozygous elavc155-gal4 flies were crossed with each homozygous uas-miR-SP fly line, and F1 progeny containing both transgenes (elavc155-gal4/+>uas-miR-SP/+) were tested for 3-hr memory. We evaluated the effects of each miR-SP line in two different ways (Figure 1A). First, the average memory performance of each elavc155-gal4/uas-miR-SP group was compared to a control group using a scramble miRNA sponge (uas-scr-miR-SP) tested in parallel using four biological replicates for both the experimental and the control groups. However, given the substantial behavioral variability of both experimental and control groups across days of experimentation using n = 4, we also compared the average performance of each miR-SP group to the scramble control group scores averaged across the screen (Figure 1A). The behavioral performance of each miR-SP in both the attP40 and the attP2 sites, when both were available, was included in our initial selection of putative hits (Figure 1 and Table S1).
Table S1 lists the 134 miRNAs of the 144 high-confidence miRNA sequences described in miRBase (http://www.mirbase.org/) that were tested (Griffiths-Jones 2004; Ruby et al. 2007). One hundred and fifteen attP40 insertions of the 134 were tested for memory because 17 miR-SPs were not available in the collection and 2 crosses failed to produce progeny (Table S1). One hundred and fourteen attP2 insertions were tested; 17 attP2 miR-SPs were not available and 3 failed to produce progeny. Thus, 39 miRNAs were tested using only the attP40 or attP2 insertion that was available.
The 3-hr PIs for the attP40 miR-SP inserts exhibited a normal distribution with an average PI of 0.49 (Figure 1B). Twenty-one of the miR-SPs significantly modulated memory scores compared to the scr-miR-SP controls, with 8 lines decreasing memory and, surprisingly, 13 lines increasing it (Figure 1D, top row, left). Twelve lines exhibited a trend (Figure 1D). Similar numerical results were obtained by comparison to the averaged scr-miR-SP control, with 24 lines significantly modulating memory and 7 lines exhibiting a trend (Figure 1D, top row, center). Among the significantly different group, 19 lines increased memory and 5 decreased it (Figure 1D). For 13 of the lines, significant differences were obtained using both statistical approaches with 8 lines increasing memory and 5 decreasing it (Figure 1D, top row, right).
The PIs for the attP2 miR-SP inserts also followed a normal distribution with an averaged value of 0.45 (Figure 1C). Compared to the scr-miR-SP control, 11 miR-SPs significantly modulated memory and 10 exhibited a trend (Figure 1D, bottom row, left). Two of the significantly different miR-SPs increased memory and 9 decreased it (Figure 1D). When the averaged scr-miR-SP PI score was used in the analysis, 19 lines were determined to significantly modulate memory and 8 lines exhibited a trend (Figure 1D, bottom row, center). Among the significantly different group, 5 miR-SPs increased memory and 14 decreased it (Figure 1D). Four miR-SPs were significant using both statistical approaches, and, among them, 3 miR-SP lines decreased memory and 1 increased it (Figure 1D, bottom row, right).
The proportion of miR-SP lines that significantly modulated memory appeared to be somewhat different for the attP40 vs. attP2 insertion sites (Figure 1D). To test for an authentic difference, we compared the fraction of lines that increased, decreased, exhibited a trend, or were neutral for attP40 and attP2 inserts using the two statistical approaches (Figure 1D). We found that the two insertion sites were indeed different in their effects. This held true when either the scr-miR-SP control (χ2 test, P = 0.03) or the averaged scr-miR-SP control value (χ2 test, P = 0.006) was considered. This may be due to a different expression level from inserts at the attP40 vs. attP2 sites. To show this in a second way, we compared the PI distribution of the 95 lines for which two inserts were available (Figure 1E). The average of the 95 PI values for all the miR-SP was significantly different between the two insertion sites (Figure 1F).
For seven miR-SPs, a significant effect or trend was observed with both attP40 and attP2 inserts (Table S2 and Figure S2A). To integrate the data collected into a single variable for comparing the miR-SPs, we assigned a score of 1.0 for P-values <0.05 and a score of 0.5 for 0.05 < P-values < 0.1 for the four comparisons made for each miR-SP, yielding lines with scores from 0 to 4 (Figure 1A and Table S1). Seventy-one of the miR-SPs modulated memory (score ≥ 0.5) for either the attP40 or the attP2 insert when compared to the paired scr-miR-SP control or the scr-miR-SP PI averaged across the screen. We passed 53 of the lines with a score of ≥1.0 (Table S1) into a secondary screen.
In the secondary screen, we employed an alternative control to test the phenotypic dependency on the presence of both a uas-miR-SP and the elavc155-gal4 driver (Figure 2, A and B and Table S3). Leakiness of any uas-miR-SP insertion would render it difficult for further study. Each uas-miR-SP line was crossed to either elavc155-gal4 or wCS10 flies (control). The results of this secondary screen were analyzed using the same two statistical approaches described above (Figure 2, A and B).
The 3-hr PIs for the miR-SPs inserted in the attP40 locus followed a normal distribution with a mean of 0.5 (Figure 2C). For the 53 lines of interest, 49 attP40 inserts were tested; 4 attP40 inserts were not available (Table S3 and Figure 2A). Ten of the elavc155-gal4-driven miR-SPs significantly modulated 3-hr memory compared to the corresponding +/+>uas-miR-SP/+control genotype, with 8 increasing and 2 decreasing memory (Figure 2E, left). Six lines exhibited a trend, and 33 lines were neutral. An equivalent proportion (χ2 test, P = 0.09) was obtained when the PI scores were compared with the average PI of all tests of the +/+>uas-miR-SP/+ control, with 11 miR-SPs increasing and 9 decreasing memory (Figure 2E, right). Five miR-SPs gave 0.5 < P-values < 0.1 and 24 were neutral. Finally, 8 of the miR-SPs exhibited significantly increased or decreased memory performance using both statistical approaches (Table S4 and Figure S2B).
The average 3-hr memory performance was 0.47 for the attP2 inserts, and the PIs followed a normal distribution (Figure 2D). Forty-four miR-SPs were tested with nine attP2 inserts being unavailable (Table S3 and Figure 2B). Twelve elavc155-gal4-driven miR-SPs significantly increased memory and one decreased it compared to the corresponding +/+>uas-miR-SP/+ control genotype (Figure 2F, left). These proportions were again similar (χ2 test, P = 0.63) when compared with the average PI score of the control across the secondary screen: 16 miR-SPs significantly increased and none decreased memory (Figure 2F, right). Nine miR-SPs significantly modulated memory using both statistical approaches (Table S4 and Figure S2B).
We combined the results of the two successive screens to rank the miR-SPs according to their performance relative to the scr-miR-SP and the +/+>uas-miR-SP controls, including both the attP40 and attP2 inserts (Table S1 and Table S3). From this we chose 16 lines (Table S5) that were retested in parallel with both gal4 and uas controls. This step was added to validate the most reproducible and strongest lines for further analyses. We compared the 3-hr performance of each miR-SP driven by elavc155-gal4 (elavc155-gal4/+>uas-miR-SP/+) with the scr-miR-SP crossed with elavc155-gal4 (elavc155-gal4/+>uas-scr-miR-SP/+) and uas-miR-SP crossed with wCS10 flies (+/+>uas-miR-SP/+). From the 16 miR-SP retested, 5 miR-SPs significantly modulated 3-hr memory compared to both controls (Figure 3 and Figure S3). The five miRNAs identified were miR-9c, miR-31a, miR-305, miR-974, and miR-980. Four of these miR-SPs reduced memory (miR-9c, miR-31a, miR-305, miR-974) and one (miR-980) increased memory. We present additional characterization below on miR-31a and miR-974.
microRNA-31a is required in cholinergic neurons for optimal 3-hr memory
The behavioral genetic screen described above utilized the pan-neural gal4 driver, elavc155-gal4. A critical issue stemming from these results concerns the brain regions and cell types that require the normal expression of miR-31a for normal 3-hr memory. To address this, we employed 10 different gal4 lines that drive robust expression in limited regions of the adult brain, including portions of the olfactory nervous system that are involved in olfactory memory along with drivers for cell-type-specific expression (Figure 4A; Davis 2005, 2011). We also retested the gal4 driver, elavc155-gal4, in this spatial and cell-type mapping study. The gal4-driven expression of miR-31a-SP (in the attP40 site) had no effect on 3-hr memory compared to the associated scrambled control in any specific neuronal population tested except for those associated with Cha-gal4, a gal4 line driven by the promoter of choline acetyltransferase (Figure 4A; Kitamoto 2001). We reproduced this result using a miR-31a-SP line containing both attP40 and attP2 inserts (Figure 4B). The odor and shock avoidance of cholinergic neuron- or CNS-expressed miR-31a-SP were not significantly different from the control (Figure 4, C and D). We conclude from these data that normal expression of miR-31a is required in cholinergic neurons outside of those represented by the spatially restricted gal4 drivers (Figure 4A) for normal 3-hr memory and not for normal odor or shock perception. Further studies are required to identify the subpopulation of cholinergic neurons involved.
For each miRNA, dozens to hundreds of potential RNA targets are predicted using in silico approaches (Bartel 2009). Predicted targets are often validated by quantitative RT-PCR, sensor assays, direct pull-down, or target protector experiments (Hsu et al. 2009; Staton and Giraldez 2011; Connelly and Deiters 2014). Such experiments may reveal interactions between a given miRNA and its predicted target, but they fail to reveal whether the interaction is functional in the biological process of interest. Moreover, target mRNAs that are inhibited from translation by their miRNA but not degraded would fail to exhibit changes in expression (Griggs et al. 2013). Consequently, we chose to test the involvement of miR-31a potential targets using an RNAi knockdown approach to obtain insights into the network of genes functionally required for normal 3-hr memory that may be potentially regulated by the microRNA. The miR-SPs are predicted to buffer their corresponding miRNAs, which in principle should lead to overexpression of their direct mRNA targets (Figure S1C). Thus, the phenotypes observed with any miR-SP in theory can be induced by the overexpression of the proteins post-transcriptionally regulated by targeted miRNAs. The knockdown of authentic targets of miR-31a might produce the opposite phenotype as the miR-31a-SP if the behavioral phenotype is linearly related to mRNA and protein expression levels. Alternatively, the inhibition of authentic targets of miR-31a might produce the same phenotype as the miR-31a-SP if the behavioral phenotype emerges due to either reduced or increased expression from optimum level. The RNAi knockdown approach was facilitated by the existence of large libraries of conditional uas-RNAi lines that exist for nearly every Drosophila gene (http://stockcenter.vdrc.at/control/main).
We used TargetScan for in silico prediction of potential mRNA targets of miR-31a (Lewis et al. 2003). The algorithm predicts miRNA targets by searching for sequences complementary between the seed region of the miRNA and the 3′ UTR of selected mRNAs, taking into account the local sequence environment (Bartel 2009). Using this approach, 57 putative Drosophila mRNA targets were predicted for miR-31a (Table S6). We expressed the uas-RNAi’s using n-syb-gal4, a second pan-neural and strong gal4 driver, for 38 of the putative targets. Sixteen targets had no uas-RNAi line available in the VDRC KK library, and three uas-RNAi’s failed to produce progeny when expressed in the CNS. Among the 38 lines screened, 15 were identified as potential hits based on the distribution of PI values, selecting those lines deviating from the mean by >1 SD (Figure 5 and Table S6). Of the 15 lines that were retested with another n = 4, four were identified as potential mRNA targets that decrease 3-hr memory by >1 SD from the mean (Table S7). All four potential targets decrease 3-hr memory, functioning as downstream targets of miR-31a, whose increased or decreased expression perturbs memory, or in the biology of memory formation independently of miR-31a. However, some of these lines exhibit a collapsed-wing phenotype in conjunction with n-syb-gal4 (Table S7), leaving open the possibility that the behavioral deficit is due to poor fitness rather than a specific effect on memory processes.
Competitive inhibition of miR-974 in olfactory receptor neurons and MB-V2 neurons increases 3-hr memory performance
We also investigated the spatial disruption of miR-974 that modulates 3-hr memory (Figure 6). We tested seven different gal4 transgenes that drive expression in different subsets of brain neurons and found that expression of miR-974-SP had no significant effect. These gal4 lines drive expression in some projection, central complex, mushroom body, and dopaminergic neurons and in both of the dorsal paired medial neurons (Figure 6A). Surprisingly, miR-974-SP expression significantly increased memory when expressed in olfactory receptor neurons and in mushroom body-V2 neurons, a phenotype opposite of that observed when expressed throughout the CNS. Although surprising, such a result is plausible if memory increase, generated by local expression of miR-974-SP, is masked by global expression in other brain areas that impair memory performance (see Discussion). We subsequently verified the CNS-wide effect of impairing 3-hr memory using elavc155-gal4 (Figure 6B) and tested odor and shock avoidance to ensure that flies expressing uas-miR-974-SP exhibited no significant difference from control in sensory parameters (Figure 6C). The simplest conclusion from these analyses is that a reduction in miR-974 abundance in olfactory receptor neurons and mushroom body-V2 neurons might increase 3-hr memory, but a global reduction of this miRNA impairs memory.
We screened 16 predicted mRNAs of miR-974 downstream targets (Table S8) for their involvement in 3-hr memory. Five of these predicted targets passed the first screen and one was confirmed in a rescreen (Table S7 and Table S8). The confirmed RNAi line is against Fas1, a cell adhesion molecule (Table S7). This line also exhibited the collapsed-wing phenotype.
We report here the first systematic screen for individual miRNA involvement in memory formation using miRNA sponges, a loss-of-function approach. We tested 134 miRNAs using miRNA-SP in two genomic locations in successive screens. At present, 256 putative miRNA sequences are available in the miRBase for D. melanogaster (http://www.mirbase.org/cgi-bin/mirna_summary.pl?org=dme) of which 144 are listed as being of high confidence. Therefore, our screen encompassed >90% of the high-confidence miRNAs described so far. The screen identified five miRNAs—miR-9c, miR-31a, miR-974, miR-305, and miR-980—that reproducibly altered memory by either disrupting the neuronal physiology underlying memory formation or altering the development of the nervous system. Most interestingly, some miRNA-SPs impaired memory while others enhanced it. We partially mapped the effects for two miRNA-SPs to cell types or brain regions using a panel of gal4 drivers, used an in silico approach to predict miRNA gene targets, and tested some of the predicted mRNA targets using an RNAi approach that offered a direct way to interrogate the functional involvement of putative miRNA targets in ITM. All together, our results offer important and broad insights toward understanding the roles of miRNAs in memory storage and offer a valuable resource on which to base further studies.
We believe that our ranking method offers value for genetic screens because it is progressive, additive, simple, and straightforward. This method provided a rank according to several parameters: (1) the genomic localization of the sponge insert to compensate for possible genomic position effects, (2) the use of two different control crosses to ensure specificity of the sponge and eliminate false positives due to leaky expression, and (3) the successive and multiple (×5) tests of the same line; all of which provided confidence for our final selections. Our experimental approach also offers advantages: (1) two different researchers tested the lines, countering possible individual bias; (2) the parameters employed were given equal weights; and (3) sponge lines were ranked rather than eliminated.
Although five miRNAs were identified as important for memory formation measured at 3 hr after conditioning, it is likely that false negatives exist among the 129 that failed at one or more points in our extensive and stringent screening pipeline. Learned behavior is an extraordinarily complex phenotype for a genetic screen, the assay for memory formation is labor-intensive and time-consuming, and we utilized a partial loss-of-function approach such that positives with more subtle effects may have been missed. The participation of some miRNAs was missed because memory formation is not a unitary process. For example, miR-276a has been shown to be necessary for long-term memory formation but dispensable for short-term memory (Li et al. 2013). We focused our study on aversive classical conditioning, being thereby blind to miRNAs that might have a specific involvement in appetitive conditioning. Moreover, the miRNA-SPs function by competitive inhibition, so the effectiveness of any given miRNA-SP will vary according to the expression level of the miRNA-SP and its corresponding miRNA. However, our approach offers significant advantages over a genomic knockout methodology by facilitating the subsequent cell type and spatial mapping of the effects, an important consideration given that the process of memory formation is distributed across multiple cells types within the brain (Davis 2011). Furthermore, conditional inhibition offers protection against the accumulation of genetic modifiers or physiological compensation that can occur across generations with chronic knockouts.
An unexpected observation made from our screen that may be important for all genetic experiments that employ site-specific transgene integration was that sponges inserted into the attP40 site appeared more potent than those inserted at the attP2 site. This is despite prior studies showing that the expression level of a luciferase reporter inserted in either the attP40 or attP2 locus was similar in the nervous system of larvae (Markstein et al. 2008). The most likely explanation for this is that the local environment of one of the insertion sites influences expression, at least in the adult brain, despite the presence of insulator sequences in the construct. Our ranking approach considered both insertions without bias.
Little is known about the five miRNAs identified to be involved in memory formation, and additional studies will be necessary to obtain deeper insights. Nevertheless, several of those identified have been studied in other systems and found to be involved in neuroplasticity and brain disorders. miR-9c is part of a conserved family found from flies to humans (Yuva-Aydemir et al. 2011). It is highly expressed in the brain (Delaloy et al. 2010), and an unpublished report quoted by Bredy et al. seems to indicate that increased miR-9 regulates dendritic arborization and cognitive abilities in mice (Bredy et al. 2011). Other reports indicate that miR-9 is elevated after alcohol exposure (Pietrzykowski et al. 2008) and is involved in dementia (Hebert et al. 2008) and polyglutamine disease (Packer et al. 2008). In addition, miR-9 expression is downregulated in patients with Alzheimer’s and Huntington’s disease (Cogswell et al. 2008; Packer et al. 2008; Maciotta et al. 2013).
miR-31a is part of a family with one ortholog in humans (Gerlach et al. 2009). It was first described as a tumor suppressor with expression levels varying according to metastatic state (Valastyan and Weinberg 2010). Our data show that miR-31a is necessary for optimal 3-hr memory performance in the CNS and more specifically in cholinergic neurons. The Cha-gal4 driver covers a large number of neurons in the adult brain. We hypothesize that the population responsible for the phenotype is included within the Cha-gal4 expression domain, but is outside of the neuronal populations included in the battery of the more specific gal4 lines employed. Further studies will be necessary to identify the specific subpopulation of cholinergic neurons in which miR-31a is necessary. We tested 3-hr memory in a large group of potential miR-31a target mRNAs and showed that four modulated memory. This, of course, is indirect evidence for potential mRNA targets but offers a list for further study.
miR-974 has been described in five species of Drosophila (Gerlach et al. 2009). Our data show that reducing its expression in the whole nervous system reduces memory. Intriguingly, the effect was opposite when mir-974 was silenced in ORN and MB-V2 neurons. This unusual observation prompts three possible explanations. First, miR-974-SP may have a negative effect on memory when expressed in a neural compartment that we did not test, nullifying the positive effects of expression in ORN or MB-V2 neurons. Second, the increasing effects on memory when expressed locally may be nullified by expression throughout the CNS through systems interactions. Third, the miR-SP expression level induced by the different gal4 drivers might induce different phenotypes. Further studies will be necessary to choose between those explanations. This miRNA has 16 targets predicted by TargetScan with one (Fas-I) participating in 3-hr memory. Fas1 is a cell-adhesion molecule involved in axonal guidance (Zinn et al. 1988). The fas1 mutant exhibits an increased number of terminal branches and varicosities at the neuromuscular junction (Zhong and Shanley 1995). Structural or functional changes in the adult brain due to abnormal Fas1 expression may underlie the observed changes in memory.
We thank the Vienna Drosophila RNAi Center for furnishing all the RNAi lines and Caitlin DeStefanis, Daniel Richter, and Erica Walkinshaw for their assistance with fly husbandry and the behavioral analysis of RNAi lines. This research was supported by National Institutes of Health grants R37 NS19904 (to R.L.D.) and R01 NS069695 (to D.V.V.).
Communicating editor: M. F. Wolfner
Supporting information is available online at http://www.genetics.org/lookup/suppl/doi:10.1534/genetics.114.169623/-/DC1.
- Received June 24, 2014.
- Accepted August 20, 2014.
- Copyright © 2015 by the Genetics Society of America