A kinase-dependent feedforward loop affects CREBB stability and long term memory formation

In Drosophila, long-term memory (LTM) requires the cAMP-dependent transcription factor CREBB, expressed in the mushroom bodies (MB) and phosphorylated by PKA. To identify other kinases required for memory formation, we integrated Trojan exons encoding T2A-GAL4 into genes encoding putative kinases and selected for genes expressed in MB. These lines were screened for learning/memory deficits using UAS-RNAi knockdown based on an olfactory aversive conditioning assay. We identified a novel, conserved kinase, Meng-Po (MP, CG11221, SBK1 in human), the loss of which severely affects 3 hr memory and 24 hr LTM, but not learning. Remarkably, memory is lost upon removal of the MP protein in adult MB but restored upon its reintroduction. Overexpression of MP in MB significantly increases LTM in wild-type flies showing that MP is a limiting factor for LTM. We show that PKA phosphorylates MP and that both proteins synergize in a feedforward loop to control CREBB levels and LTM. key words: Drosophila, Mushroom bodies, SBK1, deGradFP, T2A-GAL4, MiMIC


Introduction
Forward genetic screens coupled with classical conditioning paradigms (Tully and Quinn, 1985) have been successful in identifying many molecular determinants of learning and memory in Drosophila (Guven-Ozkan and Davis, 2014). Among the genes identified are several components of the cAMP signaling pathway, including dunce (a cAMP specific phosphdiesterase), rutabaga (a cAMP specific adenylyl cyclase), Dc0 (Protein Kinase A, PKA) and CREBB (a transcription factor) (see Figure 1-figure supplement 1 for these and others) (McGuire et al., 2005). In adult flies, these genes act in cells of the mushroom bodies (MB), higher brain centers responsible for associating conditioned stimuli (CS) and unconditioned stimuli (US), and for storing these associations (Dubnau and Tully, 1998). Associations are stored by the MB Kenyon cells (KC), which are activated by CS-such as odors-via cholinergic transmission from olfactory projection neurons and US-such as electric shock-via dopamine signaling (McGuire et al., 2005). Acetylcholine receptors permit Ca 2+ entry into KCs, while dopamine receptors activate a Ca 2+ /Calmodulin-responsive adenylyl cyclase (rutabaga), which acts as a coincidence detector (Busto et al., 2010). cAMP produced by Rutabaga activates PKA and triggers a downstream MAP kinase cascade, which leads to the phosphorylation and activation of the transcription factor CREBB (Impey et al., 1999). CREBB binds to cAMP response elements (CRE), activating the transcription of genes required for long term memory formation (DeZazzo and Tully, 1995).
CREBB is phosphorylated by several kinases, including PKA and CamKII, two kinases known to be involved in memory formation (Horiuchi et al., 2004;Mayr and Montminy, 2001) (Figure 1-figure  supplement 1). Here we describe the identification of a novel gene Meng-Po (MP), that is required for memory formation but not for learning. The MP protein regulates the stability of CREBB together with PKA. In the absence of MP the protein stability of CREBB is affected, and removing a single copy of MP and PKA leads to a dramatic loss of CREBB and memory formation. In addition, overexpression of MP strongly promotes memory formation, indicating that MP is not only required but that it also plays an instructive role in memory formation.

Results
Genes encoding protein kinases expressed in MB in adult brain and behavioral consequences of RNA interference To identify novel protein kinases involved in learning/memory, we developed a strategy that allows us to determine which kinases are expressed in MBs. We selected 27 putative kinase-encoding genes for which fly lines were available that contained intronic insertions of the Minos Mediated Integration Casette (MiMIC) (Nagarkar-Jaiswal et al., 2015;Venken et al., 2011). These MiMICs contain a swappable cassette, allowing integration of any DNA using recombinase-mediated cassette exchange (RMCE). We replaced these cassettes with a Trojan exon encoding SA-T2A-GAL4 to permit the detection of cells expressing the kinase-encoding genes using UAS-mCD8::GFP (Diao et al., 2015). Of the 27 putative protein kinase genes screened, we found 12 that are expressed in MBs In Drosophila, Pavlovian olfactory aversive learning requires coincidence detection of a conditioned stimulus (CS), an odor, and an unconditioned stimulus (US), an electric shock. Through a single training session using a T-maze assay in which flies are exposed to 12 CS-US pairings in one min, flies can associate CS with US and learn to avoid the odor paired with electric shock (Tully and Quinn, 1985). After training, flies can learn (tested immediately) and form an intermediate-term/3 hr memory (tested after 3 hr) (Margulies et al., 2005). We identified two putative kinases, CG11221 and wallenda (wnd), which when knocked down, cause significant reductions in performance index for 3 hr memory, but not for learning . Overexpression of wnd has been documented to enhance memory in Drosophila (Huang et al., 2012), but CG11221 has not been previously characterized in flies. We therefore tested the effects of knocking down CG11221 expression using a second RNAi expressed specifically in MB neurons to confirm the memory loss phenotype (Figure 1-figure supplement  3b). To provide additional evidence that CG11221 is not required for learning, we used a short CS/ US association protocol (30' CS + US instead of 60') to train flies. Knockdown of MP did not affect learning, further indicating that MP is not required for learning ( Figure 1-figure supplement 3b).
CG11221 is an evolutionarily conserved serine/threonine protein kinase (human SBK1: 37% identity and 75% similarity, Figure 1-figure supplement 4). SBK1 is expressed in the hippocampus, cortex, and cerebellum of adult rodents and loss of SBK1 in mice causes partial embryonic lethality (Nara et al., 2001;Skarnes et al., 2011). The CG11221 MI03008 -T2A-GAL4; UAS-mCherry flies exhibit broad expression of the gene in third instar larvae and adult flies (Figure 1-figure supplement 5). As shown in Figure 1a, the gene is also expressed in the adult brain and is prominent in the MB. In light of its importance in memory, we renamed CG11221, Meng-Po (MP), for the Lady of Forgetfulness, a character in Chinese mythology who ensures that people are ready for reincarnation by providing the 'Tea of Forgetfulness' so they lose the memory associated with their former life.

Loss of MP causes a loss of memory
To determine whether the memory deficits resulting from loss of MP are due to its activity in adult MB neurons or are a result of its developmental expression, we used the deGradFP method (Caussinus et al., 2011;Nagarkar-Jaiswal et al., 2015) to selectively decrease levels of MP protein in MB of 5 day old flies. To do so, we replaced the MiMIC insertion of the CG11221 03008 in the first coding intron with a SA-GFP-SD in-frame with the MP coding sequence using a previously described technique (Nagarkar-Jaiswal et al., 2015). This manipulation insures the expression of an internally-    , learning) Learning is normal after knockdown of MP in MB by deGradFP at 28˚C for 3 days (b'). Flies raised at 18˚C are used as a control (a'). (b, 3 hr memory) 3 hr memory is impaired after knockdown of MP in MB (b'). However, the performance score of ARM is intact in MP knockdown flies which are treated with a cold shock and compared to flies raised at 18˚C (3 hr memory a' and b'+cold shock). In c' condition, the flies show normal performance score of 3 hr memory. The 3 hr memory impairment (in b'+cold shock, right panel) can be fully rescued when the animals are shifted to 18˚C for two days (the groups boxed in dashed line are the exactly same flies). (c), 24 hr memory) 24 hr LTM (10 x ST) is impaired upon knockdown of MP in MB (b', red). After treatment with 35 mM cycloheximide (CXM), the MP knockdown flies (b'+CXM, red) don't exhibit a performance that is worse than control flies. In the c' conditions, the flies exhibit a normal performance score for the 24 hr memory assay. The performance of ARM (10 x MT) is intact. 10 x ST:10 times spaced training. 10 x MT:10 times massed training. The mean ±SEM is plotted for each genotype; n = 8 for each group. **p<0.01. ***p<0.001. DOI: https://doi.org/10.7554/eLife.33007.008 Using this strategy, we reduced the level of MP-GFP in adult MB for three days (Figure 2a,b') and assayed learning and memory performance in these flies. As shown in Figure 2b, flies with MP depleted in the MB (b', red column) learn as well as y w control animals (b', white). However, when tested 3 hr after training, these flies exhibited severe memory deficits (Figure 2b It has been previously shown that the 3 hr memory has two distinct components: an anesthesiaresistant memory (ARM) and an anesthesia-sensitive memory (ASM), only the latter of which can be erased by cold-shock (Lee et al., 2011;McGuire et al., 2005). To determine whether MP functions in ARM, ASM or in both, we subjected trained animals in which MP-GFP had been depleted in the MB to cold-shock and found that the residual memory was unaffected ( Figure 2b, 3-h memory, b' with cold shock). Furthermore, the memory performance in these animals was statistically indistinguishable from that of control animals subjected to cold shock. These results suggest that MP is required for ASM, but not for ARM. Importantly, loss of 3 hr memory can be restored by placing the flies at 18˚C for two days after knockdown ( Figure 2b, 3-h memory, c'). To test whether the transient loss of MP in the MB has long-term consequences to the animal's ability to form memories, we restored expression of MP-GFP after 3 days of depletion at 28˚C by returning animals to 18˚C. Their ability to learn and remember upon restoration of MP-GFP revealed complete recovery of memory function ( Figure 2b, 3-h memory, right panel, column c').
In parallel studies we also tested animals with reduced MP for deficits in 24 hr long-term memory, LTM, and radish dependent 24 hr ARM (Margulies et al., 2005). LTM requires CREBB as well as new protein synthesis Keene and Waddell, 2007) and is typically induced by repetitive (10x) spaced training (ST) at 15 min intervals. In contrast, 24 hr ARM does not require protein synthesis (Dubnau and Tully, 1998;Keene and Waddell, 2007) and is induced by repetitive mass training (MT) without the 15 min resting intervals. We found that MP is dispensable for 24 hr ARM (Figure 2b, 24-h memory, 10x MT), but is required for LTM (Figure 2b, 24-h memory, 10x ST, b', red column). After treatment with 35 mM cycloheximide (CXM), an inhibitor for protein synthesis, the MP knockdown flies (Figure 2b, 24-h memory, 10x ST, b'+CXM, red column) do not perform worse than other control flies. This indicates that the loss of memory is protein synthesis-dependent. Loss of 24 hr LTM memory can be restored when returning the flies to 18˚C for two days after knockdown. (Figure 2b, 24-h memory, 10x ST, c'). Hence, loss of MP affects neither learning nor ARM, but severely impairs 3 hr ASM and 24 hr LTM.

Overexpressing MP enhances 24 hr LTM and increases CREBB activity
To determine whether MP is not only necessary for LTM, but also acts as a limiting factor for LTM, we overexpressed MP in MB and analyzed 24 hr LTM. To avoid saturating LTM, as occurs when the spaced training is repeated 10x, we adopted a paradigm in which flies were subjected to only a 3X training paradigm (Lee et al., 2011) (3x ST). Conditional overexpression of MP in MB under the control of OK107-GAL4 was accomplished using a temperature-sensitive GAL4 inhibitor (Tub-GAL80 ts ),  et al., 2004). Flies grown at 18˚C and placed at 29˚C for two days robustly expressed MP as detected by Western blot (Figure 3a,c').
Flies were subjected to temperature protocols that either did (Figure 3b,b') or did not (Figure 3b,a') induce MP overexpression and we then tested their performance in both learning and LTM using the T-maze assay. The learning scores of Tub-GAL80 ts /+;UAS-MP-HA/+;OK107-GAL4/+ flies temperature-shifted to 29˚C, and therefore overexpressing MP, were the same as those of unshifted flies of the same genotype and those of other control groups (Figure 3b,c'). In contrast, (1) at 18˚C for one day after eclosion.
(2) at 29˚C for three days.
(3) at 29˚C for three days, then shift to 18˚C for one day. (4) at 29˚C for three days then shift to 18˚C for two days. 10 fly heads (five males/five females) were collected for a single assay. RLU: relative luminescence unit. The mean ±SEM is plotted for each point; n = 6. DOI: https://doi.org/10.7554/eLife.33007.011 the performance scores of these flies for 24 hr LTM were significantly enhanced after MP overexpression in MB (3x ST; Figure 3b,d'). The observed memory enhancement can be erased by feeding flies 35 mM CXM, indicating that enhanced memory is protein synthesis-dependent (3x ST +CXM; Figure 3b,d'). Moreover, this enhancement is specific to 24 hr LTM, and not 24 hr ARM induced by 3-times mass training (3x MT; Figure 3b,d'). Thus, MP is not only necessary for LTM, but also is also promoting LTM formation.
Given that CREBB is a central player in LTM formation, we tested its activity in MB during MP overexpression using a previously described luciferase assay. This assay relies on CREBB-mediated transcription of luciferase from the CRE-F-Luc construct, which contains three copies of the CREBBbinding cAMP Response Element (CRE), upstream of an FRT-flanked mCherry-encoding stop cassette and the firefly luciferase gene (Figure 3c,a') (Tanenhaus et al., 2012). Spatial control of luciferase expression can be achieved by using the GAL4/UAS system to drive the cell-type specific expression of a UAS-FLP recombinase (Figure 3c,a'-b'). We created flies of the following genotype:

MP is a kinase, and loss of MP affects CREBB protein levels in MP gene trap animals
Given that MP encodes a putative kinase, we tested its ability to phosphorylate a known substrate of two other kinases, ERK and PKA. We expressed MP-HA in S2 cells, affinity purified the protein, and performed kinase assays on Myelin Basic Protein (MBP) (Martenson et al., 1983). As shown in Figure 4a, ERK, PKA and MP phosphorylate MBP. Given the CRE-luciferase assay results and the established regulation of CREBB by phosphorylation (Horiuchi et al., 2004;Tully et al., 2003), we hypothesized that CREBB is a substrate of MP. To determine if MP can phosphorylate CREBB we purified the CREBB protein from S2 cells and performed kinase assays. Unlike PKA, which is able to phosphorylate CREBB, MP was not able to phosphorylate CREBB ( Figure 4b). Hence, although CREBB activity is dependent on MP levels, CREBB may not be a direct substrate of MP.
To more generally assess how MP upregulates CREBB, we sought to determine how CREBB activity is affected by loss of MP function. The MI03008 MiMIC insertion in the MP gene functions as a gene trap (Figure 4-figure supplement 1a) and RT-PCR fails to amplify a product between these exons (Figure 4-figure supplement 1b). This and other evidence suggest that MI03008 is a severe loss of function allele of MP. To estimate the levels of CREBB in MI03008 mutants, we probed Western blots of protein isolated from the heads of adult MI03008/MI03008 animals using an anti-phospho-CREBB antibody  and compared the signal to that observed in Western blots from heads of control animals. In parallel, we analyzed the signals obtained using a second antibody (Pan-CREBB) that assesses total CREBB protein levels. Both antibodies revealed an approximately 60% reduction in immunoreactivity in samples from MP mutant brains (Figure 4c,a'-b'). Note, that there is a minor reduction in the ratio of phospho-CREBB/total CREBB (Figure 4c,c'). In contrast, CREBA protein levels are not altered (Figure 4d), consistent with the observation that CREBA has no role in memory formation (Abrams and Andrew, 2005). Finally, CREBB mRNA levels are unchanged in MP mutants (Figure 4-figure supplement 1c), indicating that the observed reduction in CREBB protein is a post-transcriptional effect.
We also tested if ATG2-CREBB is affected. This CREBB isoform is encoded by a downstream, inframe initiation codon of one of the transcripts of the CREBB gene, and corresponds to a transcriptional activator. It has been shown to be required for memory enhancement . As shown in Figure 4-figure supplement 1d, we find that both the anti-Pan-CREBB antibody and the antibody recognizing ATG2-CREBB  identify a~30 kDa band that is reduced in MI03008 mutant brains. Hence, these data suggest that MP promotes the translation or stabilization of CREBB protein.
If CREBB mediates the effects of MP on LTM and loss of MP function reduces CREBB protein levels, then reducing CREBB in MB should phenocopy the effects of eliminating MP in MBs. To determine whether this is the case, we knocked down CREBB mRNA in adult MB using OK107-GAL4 and two independent UAS-RNAi lines by shifting 3-5 day old flies kept at 18˚C to 25˚C three days prior to testing. We find that decreasing CREBB in MB causes a reduction of 24 hr LTM with both RNAi knockdowns (10 x ST; Figure 4e Figure 2) indicates that loss of MP or CREBB function results in very similar memory defects. The most parsimonious explanation of these data is that loss of the MP protein kinase leads to reduced levels of CREBB protein, which in turn leads to a severe LTM impairment.

MP is regulated by PKA
The MP protein contains the canonical PKA phosphorylation motif RRFS (Huang et al., 2005) at residues 331-334 ( Figure 5-figure supplement 1). To determine whether serine residue 334 is a substrate of PKA we introduced a S334A mutation into MP (MP S334A ) and examined phosphorylation of both the wild-type and mutant proteins by PKA. As shown in Figure 5a, MP is phosphorylated by PKA, but MP S334A is barely phosphorylated. We conclude that S334 is required for proper MP phosphorylation by PKA. To determine whether phosphorylation alters the MP kinase activity, we next assayed the kinase activities of both wild-type MP and MP S334A . As shown in Figure 5b, the kinase activity of MP S334A is significantly lower than that of wild-type MP.
The above data indicate that PKA activates MP in addition to activating CREBB. Loss of function mutations in PKA cause a subtle but significant reduction in CREBB ( Figure 5-figure supplement  2). Hence, PKA and MP may work together in a coherent feedforward loop (Mangan et al., 2003) to upregulate CREBB activity and support LTM formation. To assess possible synergistic interactions between PKA and MP, we simultaneously reduced the protein levels of both MP and the catalytic subunit of PKA by creating MI03008,Pka+/MP+,Pka-C1 B10 heteroallelic flies, and assessed the levels of CREBB protein by Western blot . CREBB immunoreactivity in the heads of these animals is very severely reduced compared to that observed in the heads of either wild-type contols or animals with reduced gene dosage of only MP or Pka-C1 (Figure 5c). Simultaneous reduction of PKA and MP thus substantially potentiates the effects of reducing either protein alone, consistent with a model in which the two kinases act within the same signaling pathway to regulate CREBB activity and LTM. To determine if the MI03008,Pka+/MP+,Pka-C1 B10 animals exhibit memory defects, we assayed their ability to both learn and form 24 hr LTM. We find that these flies exhibit normal learning (Figure 5d,a') and ARM (10 x MT; Figure 5d,c'), but lack 24 hr LTM (10 x ST; Figure 5d,b') consistent with the loss of CREBB function. Interestingly, flies heterozygous for either MP or Pka-C1 display neither learning nor LTM defect, suggesting a potent synergistic interaction and feedforward loop between PKA and MP.

Discussion
Using MiMIC technology, we converted 27 genes encoding putative protein kinases with the Trojan T2A-GAL4 exon and performed an image screen for genes expressed in MBs (Diao et al., 2015). This tagging approach is especially useful for genes that are expressed at low levels in the CNS. By tagging the proteins with GFP, a conditional and reversible knockdown can be achieved in almost any tissue or cell (Nagarkar-Jaiswal et al., 2015). This allowed us to identify a novel serine/threonine protein kinase, Meng-Po (MP), that is a critical player in LTM formation in Drosophila. MP is a homologue of SBK1 in mammals (Figure 1-figure supplement 4), a gene that is expressed in the hippocampus and the cortex (Nara et al., 2001;Skarnes et al., 2011). Loss of this gene in mice is associated with embryonic lethality (Skarnes et al., 2011), whereas in flies, loss of MP leads to a reduction in viability as well as sterility.
Our data show that CREBB stability is highly susceptible to loss of MP. CREBB activity is modulated by phosphorylation via PKA and CamKII in Drosophila (Horiuchi et al., 2004). Although our findings indicate that MP kinase activity is critical for maintaining CREBB levels and that MP kinase activity acts in synergy with PKA ( Figure 6), we have not been able to demonstrate that CREBB is a direct target of MP. However, some kinases require a previously phosphorylated residue as part of their recognition sequence and we have not mixed various kinases with MP in our in vitro assays (Horiuchi et al., 2004). Hence, it remains to be established how CREBB is degraded in the absence of MP.
A reduction in CREB levels has been shown to be associated with an age-dependent memory loss in rodents. Interestingly, delivery of CREB protein in the hippocampus using somatic cell transfer attenuated LTM impairement (Mouravlev et al., 2006). However, no gene has so far been shown to affect CREBB stability in vivo and our findings that MP, together with PKA, synergize to dramatically affect CREBB levels via a feedforward loop (Mangan et al., 2003), reveal another mechanism to control CREBB levels during memory formation ( Figure 6). This model is supported by the observation that overexpression of MP increases CREBB activity and promotes memory formation, suggesting that it is a central player in LTM.

Plasmid constructs
The MP-HA and CREBB-6xHis cDNA were synthesized by GenScript. The cDNA fragments were double digested with EcoRI and XhoI and cloned into pUAST-attB. For site-directed mutagenesis of MP, the primer sets below were used: MP S334A -F: 5'-CGCCGCTTCGCCCCCCGCCTGAT-3' MP-R: 5'-CTTCAACAGCCCGTGGAACCACC-3' The PCR reaction was performed with the Q5 Site-Directed Mutagenesis kit (NEB). The PCR product was ligated and transformed into E.coli competent cells, and the colonies were selected on ampicillin/LB agar plates. For transgenic animals, the UAS-MP-HA plasmid was extracted with HiPure Plasmid Midiprep kit (Invitrogen) for microinjection.

Confocal imaging
Image processing was performed as described previously (Lee et al., 2011). Briefly, dissected brains were fixed in PBS with 4% paraformaldehyde at 4˚C overnight, transferred to PBS with 2% Triton X-100 at room temperature, vacuumed for 1 hr and left overnight in the same solution at 4˚C. For immunostaining of GFP, the samples were incubated with anti-GFP antibody conjugated with FITC (1:500) (Abcam) in PBS with 0.5% Triton X-100 overnight. Brains were cleared and mounted in Rapi-Clear (SunJin Lab Co.) and imaged with a Zeiss LSM 880 Confocal Microscope under a 20 x or 40 x C-Apochromat water immersion objective lens.
Overexpression and pulldown of MP, MP S334A and CREBB in S2 cells The Effectene transfection reagent (QIAGEN) was used to deliver DNA to S2 cells. For protein overexpression, 2 Â 10 6 S2 cells were transferred into 5 ml fresh media overnight (Schneider's Drosophila medium, Gibco). The cells were collected by centrifuging them for 2 min. UAS-MP-HA, UAS-MP S334A -HA or UAS-CREBB-6xHis were co-transfected with Act-GAL4 into S2 cells. Transfected cells were placed at room temperature for two days in medium, collected, and lysed with sample lysis buffer (50 mM Tris-Cl pH 7.5, 125 mM NaCl, 5% glycerol, 1% NP40, 1.5 mM MgCl 2 , 0.2 mM DTT) containing a protease inhibitor mix (cOmplete, Roche). The cell lysate was collected and a-HA agarose (EZview Red Anti-HA Affinity Gel, Sigma) or a-His resin (HisPur Ni-NTA Resin, ThermoFisher) was added for protein pulldown.

Kinase assay
ERK and PKA were purchased from NEB. MP and MP S334A were overexpressed and purified from S2 cells. For kinase assay, the kinases (ERK: 0.01U (NEB), PKA: 0.1U (NEB), MP and mutant MP: 13.5 ml of a-HA-agarose pulldown) were added into a cocktail containing the kinase buffer (0.5 mg/ml BSA, 15 mM Tris-Cl pH7.5 in final) MgCl 2 /ATP (0.2 mM in final), g-32 P-ATP (10 mCi in final), and myelin basic protein (MBP) (1.2 mM in final). For testing if MP or CREBB are substrates of PKA, MP and CREBB were pulled down, and incubated at 30˚C for 30 min. The reactions were terminated by adding 2x SDS sample buffer, boiling for 5 min, and run on SDS gels. After transferring to nitrocellulose membranes, radioactive signals were detected by CL-X posure film (ThermoFisher).

Luciferase activity assay
The luciferase assay system was from Promega. All groups of flies were raised in the same incubator in 12 hr light/12 hr dark conditions, and the CRE-luciferase assay was performed at the same time for all genotypes tested to avoid a circadian effect (Fropf et al., 2014). Briefly, fly heads were collected and homogenized in a reporter lysis buffer (10 heads/20 ml) and placed at À80˚C overnight. The supernatant of lysate was collected and mixed with luciferase assay reagents (20 ml lysate/100 ml luciferase assay reagent) at RT for 1 min in 96-well plate. The luciferase activity was analyzed using an Optima luminescence reader (BMG LABTECH).

Behavioral assay
Behavioral assays were performed with balanced comparative groups which were trained and tested in parallel without blinding. For behavior assays, MB-GAL4 flies were outcrossed with Canton-S w 1118 (iso1CJ) or y w flies for at least five generations, and Tub-Gal80 ts , UAS-RNAi and UAS-MP-HA flies were outcrossed with iso1CJ. For RNAi knockdown, flies were raised at 18˚C until eclosion then transferred to 25˚C for 3 days. Aversive olfactory learning was performed using the T-maze apparatus (Tully and Quinn, 1985). In brief, one training session consists of approximately 100 flies that were electrically shocked while exposed to one of two odors (3-octanol and 4-methylcyclohexanol, Sigma). The shock was alternated between 3-octanol and 4-methylcyclohexanol. Flies trained by one training session and tested immediately are tested for learning, whereas those tested 3 hr later are assessed for 3 hr memory. Flies undergoing ten training sessions and tested 24 hr later are assessed for 24 hr memory or LTM. For cycloheximide (CXM, Sigma) feeding (DeZazzo and Tully, 1995), 35 mM CXM in 5% glucose was added on Whatman 3 MM filter paper in a bottle with fly food to feed flies overnight prior to performing training sessions. After training, the flies are placed in a vial containing CXM filter paper with fly food for another 24 hr prior testing.

Statistics
Statistical analyses were performed by KaleidaGraph 4.1 (Synergy software). Behavioral data were evaluated via one-way ANOVA followed with Tukey's test for multiple comparisons (Lee et al., 2011). Data from two groups are analyzed by t-test. All data are presented as 'mean ±SEM'. *p<0.05, **p<0.01, ***p<0.001.