Abstract
Aminergic signaling modulates associative learning and memory. Substantial advance has been made in Drosophila on the dopamine receptors and circuits mediating olfactory learning; however, our knowledge of other aminergic modulation lags behind. To address this knowledge gap, we investigated the role of octopamine in olfactory conditioning. Here, we report that octopamine activity through the β-adrenergic-like receptor Octβ1R drives aversive and appetitive learning: Octβ1R in the mushroom body αβ neurons processes aversive learning, whereas Octβ1R in the projection neurons mediates appetitive learning. Our genetic interaction and imaging studies pinpoint cAMP signaling as a key downstream effector for Octβ1R function. The rutabaga-adenylyl cyclase synthesizes cAMP in a Ca2+/calmodulin-dependent manner, serving as a coincidence detector for associative learning and likely representing a downstream target for Octβ1R. Supporting this notion, the double heterozygous rutabaga/+;octβ1r/+ flies perform poorly in both aversive and appetitive conditioning, while individual heterozygous rutabaga/+ and octβ1r/+ flies behave like the wild-type control. Consistently, the mushroom body and projection neurons in the octβ1r brain exhibit blunted responses to octopamine when cAMP levels are monitored through the cAMP sensor. We previously demonstrated the pivotal functions of the D1 receptor dDA1 in aversive and appetitive learning, and the α1 adrenergic-like receptor OAMB in appetitive learning. As expected, octβ1r genetically interacts with dumb (dDA1 mutant) in aversive and appetitive learning, but it interacts with oamb only in appetitive learning. This study uncovers the indispensable contributions of dopamine and octopamine signaling to aversive and appetitive learning. All experiments were performed on mixed sex unless otherwise noted.
SIGNIFICANCE STATEMENT Animals make flexible behavioral choices that are constantly shaped by experience. This plasticity is vital for animals to appropriately respond to the cues predicting benefit or harm. In Drosophila, dopamine is known to mediate both reward-based and punishment-based learning while octopamine function is important only for reward. Here, we demonstrate that the octopamine–Octβ1R–cAMP pathway processes both aversive and appetitive learning in distinct neural sites of the olfactory circuit. Furthermore, we show that the octopamine–Octβ1R and dopamine–dDA1 signals together drive both aversive and appetitive learning, whereas the octopamine–Octβ1R and octopamine–OAMB pathways jointly facilitate appetitive, but not aversive, learning. This study identifies the cognate actions of octopamine and dopamine signaling as a key neural mechanism for associative learning.
Introduction
Experiential learning facilitates animals to promptly respond to pleasant or noxious cues for approach or avoidance, respectively, for fitness and survival. In Drosophila melanogaster, olfactory classical conditioning is widely used to assess the capacity of the organism to learn and remember the odor associated with electric shock punishment (aversive) or sugar reward (appetitive) for delineating the neural, cellular, and molecular mechanisms. Olfactory learning and memory require modulatory actions of multiple aminergic systems including dopamine (DA), octopamine (OA; an invertebrate counterpart of norepinephrine), and serotonin (5-HT; Schwaerzel et al., 2003; Kim et al., 2007a,b; Johnson et al., 2011; Sitaraman et al., 2012; Aso et al., 2014a,b; Iliadi et al., 2017). The fundamental question of how diverse aminergic signals converge within the learning and memory pathways remains incompletely resolved. The actions of modulatory neurotransmitters like DA and OA are initiated on their binding to one or more of their respective receptors in overlapping or parallel neural circuits, conveying additive, synergistic, or antagonistic interactions via intracellular effectors to acquire, store, or retrieve information. OA and 5-HT, for example in aversive conditioning, exert additive effects on intermediate memory through their actions on distinct mushroom body (MB) substructures (Wu et al., 2013), whereas DA via DAMB in the MB has an antagonistic effect by promoting forgetting (Berry et al., 2012). All three monoamines are important for olfactory learning as well; however, the nature of interactions and circuits that DA, OA, and 5-HT shape the ability of the fly to acquire aversive and appetitive memories are unknown.
The best characterized of all three monoamines in olfactory conditioning is DA (Guven-Ozkan and Davis, 2014; Aso et al., 2014a,b; Cognigni et al., 2018). Ectopic activation of discrete PPL1 or PAM DA neurons projecting to the MB substitutes the aversive or appetitive unconditioned stimulus (US), respectively, when paired with the conditioned stimulus (CS; Claridge-Chang et al., 2009; Liu et al., 2012). The DA signals for aversive and appetitive learning require the D1 receptor dDA1 in the MB, while ongoing DA activity after learning through the D5 receptor DAMB regulates forgetting (Kim et al., 2007a; Berry et al., 2012; Qin et al., 2012). The dual functions of DA for memory acquisition and forgetting are executed through the following divergent intracellular effectors: dDA1 engages Gαs/cAMP signaling, whereas DAMB recruits Gαq/Ca2+ signaling (Himmelreich et al., 2017). These findings posit DA as a major contributor to aversive as well as appetitive learning and memory.
OA is a major neuromodulator for appetitive learning. The tβh mutant lacking tyramine β hydroxylase (TβH), a rate-limiting enzyme for OA biosynthesis, is defective in the acquisition of appetitive memory in olfactory conditioning (Schwaerzel et al., 2003; Kim et al., 2007b). The function of OA in appetitive learning engages the α1-like OA receptor OAMB in the MB αβγ neurons (Han et al., 1998; Kim et al., 2013). The role of OA in aversive olfactory learning, however, is yet unsettled. Two studies report contrasting findings on the tβh mutant: Schwaerzel et al. (2003) demonstrate normal performance, whereas Iliadi et al. (2017) report poor performance in aversive conditioning. It is worth noting that the discrepancy is not due to the nature of mutation in tβh, the mutant genetic background, or conditioning parameters. Intermediate-term aversive memory, nonetheless, is shown to require the OA signal through the β-like OA receptor Octβ2R in the MB α′β′ neurons (Wu et al., 2013). In operant courtship conditioning, a different type of aversive conditioning, OAMB in the MB αβ neurons, mediates the OA signal for short-term memory of courtship suppression (Zhou et al., 2012). While these reports illustrate the functions of OA in aversive memory, the key issue that is yet unresolved but important regarding mechanism is whether the OA signal is involved only in appetitive learning or both appetitive and aversive learning. In this report, we address this issue and further delineate the neural sites and intracellular effectors critical for olfactory conditioning.
Materials and Methods
Drosophila strains and culture.
The wild-type strain used in this study is Canton-S, and the mutant defective in Octβ1R is the piggyBac transgenic line octβ1rf02819 generated by the Gene Disruption Project (Thibault et al., 2004). The octβ1rf02819 in the w1118 genetic background was obtained from the Bloomington Drosophila Stock Center (BDSC; stock #18589) and placed in the Canton-S genetic background by backcrossing with Cantonized w1118 for six generations and then replacing the X chromosome containing w1118 with the Canton-S X chromosome with the normal w gene (hereafter called octβ1r). We previously described oamb, octβ2r, dumb, and elav-GAL4,Gal80ts (Lee et al., 2003; Kim et al., 2007a; Lim et al., 2014). The Deficiency (Df) lines Df(3R)Exel6191 (stock #7670) and Df(3R)BSC685 (stock #26537), tdc2-GAL4 (stock #9313), elav-GAL4 (stock #8765), OK107-GAL4 (stock #854), c739-GAL4 (stock #7362), c305a-GAL4 (stock #30829), and UAS-Epac1-camps (stock #25407 and #25408) were obtained from the BDSC; Tβh RNAi (v51667) were obtained from the Vienna Drosophila Resource Center; pBDP-GAL4 was provided by David Anderson (California Institute of Technology, Pasadena, CA); NP1131-GAL4 and APL-GAL4 (VT43924-GAL4 recombined with UAS-GAL4 on the third chromosome) were provided by Josh Dubnau (Stony Brook University School of Medicine, Stony Brook, NY) and Ann-Shyn Chiang (National Tsing Hua University, Hsinchu City, Taiwan); and NP225-GAL4 was provided by Andreas Thum (University of Konstanz, Konstanz, Germany).
The open reading frame of Octβ1R (Maqueira et al., 2005) was cloned under UAS in the gateway vector pTW (Akbari et al., 2009), and transgenic UAS-Octβ1R lines were generated by germ line transformation in w1118 embryos as previously described (Lim et al., 2014). Germ-line-transformed lines were outcrossed with Cantonized w1118 for at least six generations, and the transgenes were placed in the octβ1r genetic background for rescue experiments. All fly lines were maintained on standard cornmeal/agar/yeast medium. Flies were collected under CO2 within 2 d after eclosion and 65–70 flies representing a group were housed together in a food vial at 25°C with ∼50% relative humidity under the 12 h light/dark cycle before testing. Unless otherwise stated, the 4- to 7-d-old flies of mixed sex were used for all experiments. For the temporal control of GAL4 activity, the control and experimental flies carrying elav-GAL4,GAL80ts were reared at 30°C for the induction of GAL4 activity or at 20°C for no induction. For developmental induction, flies were reared at 30°C until a mid-pupal stage and then maintained at 20°C before conditioning. For adulthood induction, flies were reared at 20°C throughout development and for 1 d after eclosion, and then maintained at 30°C for 3 d before conditioning. For the genetic interaction experiments, the virgin females of octβ1r or Canton-S were crossed with rutabaga (rut), dumb, or oamb males. In the case of rut, which is on the X chromosome, only the female progeny of octβ1r or Canton-S crossed with rut were used for conditioning.
RNA analysis.
The transcript levels of Canton-S and octβ1r flies were analyzed as previously described (Lim et al., 2014). Three independent sets of 50 heads per genotype were homogenized in 10 μl of RLT lysis buffer (Qiagen) with the KONTES Micro Tissue Grinder (Thermo Fisher Scientific), which was then applied to the QIAshredder Spin Column (Qiagen). Total RNA was extracted using the RNeasy Protect Mini Kit (Qiagen), and then cDNA was generated using the Invitrogen SuperScript III First-Strand Synthesis System (for RT-PCR). Quantitative PCR was conducted using the iQ SYBR Green Supermix Kit (Bio-Rad) and the StepOnePlus Real-Time PCR Detection System (Applied Biosystems) per the manufacturer instructions. The 100, 200, and 400 ng of cDNA samples were run in triplicate, and the reactions were performed with the primer sets for octβ1r, the experimental gene, and rp49, a ribosomal protein gene as a control, to quantify relative expression levels. The primer sets were designed to span at least one intron and checked for specificity using the BLAST (FlyBase Consortium, 2003) against the Drosophila genome. The PCR primers used in this study are as follows: for octβ1r, forward—TGT GCA GCC ACT GGA CTA TC, reverse—TAT GGC GTA TGC CTT GTT CA; for rp49, forward—TAC ACGG CCC AAG ATC GTG AA, reverse—GTT CGA TCC GTA ACC GAT GT.
Behavioral tests.
All experiments were performed blindly (i.e., an experimenter does not know the genotype of the fly lines when conducting all experiments). The protocol described by Kim et al. (2007a,b) were used for aversive and appetitive conditioning with slight modifications. Training and tests were conducted under dim red light and 50–60% relative humidity. Odorants used for conditioning were 2% ethyl acetate (EA; catalog #319902–500, Sigma-Aldrich) and 2.2% isoamyl acetate (IAA; catalog #112674–500, Sigma-Aldrich). For appetitive conditioning, flies were starved for 18–20 h in the vials containing double-layered, water-soaked Kimwipes before training. A group of 65–70 flies were exposed to the first odor (CS+) paired with 12 pulses of 90 V electric shock (US) for aversive conditioning or 2 m sucrose (US) for appetitive conditioning for 1 min followed by 30 s of rest (interstimulus interval). After exposure to the second odor (CS−) without electric shock or sucrose for 1 min, flies were tested immediately (3 min), 1 h, or 3 h after training. To measure the capacity of flies to learn and remember, the flies were gently transferred to a T-maze and tested for 2 min to assess their avoidance or preference of the CS+ odor. A second group of flies was simultaneously trained with the odors presented in a reversed order to account for any possible odor bias in conditioning. For the short program single-training trial (1T) in aversive conditioning (Beck et al., 2000), flies were exposed to CS+ odors for 10 s with a single pulse of 90 V electric shock followed by 30 s rest and then CS− odors for 10 s before tests. For the two-training trial (2T), flies received another 1T training program with 15 min intertraining interval (ITI). The performance index (PI) was calculated by subtracting the percentage of flies that made an incorrect choice from the percentage of flies that made a correct choice. The average PI of the two groups of flies conditioned with counterbalanced odors was used as one data point. For 1 and 3 h memory tests, trained flies were gently transferred back into their respective vials (food-containing vials for electric shock-conditioned flies or vials with water-soaked Kimwipes for sugar-conditioned flies) and maintained at room temperature before tests.
Control behaviors were examined as previously reported (Kim et al., 2007a, 2013) and include olfactory acuity, shock avoidance, and sugar preference without conditioning. Briefly, flies were placed in a T-maze and allowed to choose (odor vs air for olfactory acuity; 90 or 30 V vs no shock for shock avoidance; 2 or 0.2 m sucrose vs water for sugar preference). An average PI of the counterbalanced set as noted above was calculated and reported as one data point.
cAMP imaging.
The 4- to 7-d-old flies expressing the cAMP sensor Epac1-camps in the MB or projection neurons (PNs) in the wild-type or octβ1r genetic background were used for imaging. A fly brain was dissected within 3 min in the ice-cold hemolymph-like saline HL3 (70 mm NaCl, 5 mm KCl, CaCl2 1.5 mm, 20 mm MgCl2, 10 mm NaHCO3, 5 mm trehalose, 115 mm sucrose, and 5 mm HEPES, pH 7.1) and placed on the microscope glass containing either HL3 (baseline or vehicle control) or HL3 with 100 μm OA (treated), which was then covered with a coverslip. To create space between a microscope glass and a coverslip so as not to compress the brain, two layers of the scotch tape (3M) were placed on a microscope glass. The cAMP sensor Epac1-camps was excited at 440 nm, and emission was detected at 480 nm for CFP and 540 nm for YFP (yellow fluorescent protein). Images were collected using the 20× objective in the LSM700 confocal microscope (Zeiss) at every micrometer with the 512 × 512 pixel resolution. Each brain was used only once for imaging, thus different brains were used for imaging baseline and OA treatments. For quantification, images in grayscale were used to measure fluorescence intensity per pixel in each optical section using ImageJ software (NIH). Fluorescence intensity per pixel was calculated by [Integrated densitytotal/Areatotal] per region of interest (ROI). We defined ROI in the major brain structures, including MB and PN. The non-ROI was defined in the anterior inferior medial protocerebrum (AIMPR) where the Epac1-camps was not induced. The fluorescence intensity for each CFP or YFP was calculated by [Integrated density/Area]ROI − [Integrated density/Area]non-ROI. The cAMP response was determined by calculating the inverse FRET ratio, which is |ΔR/R0|, where R = CFP/YFP and ΔR = Rtreated − Rnontreated was used for data presentation (Shafer et al., 2008; Tomchik and Davis, 2009). To determine the timeline of cAMP responses, the changes in Epac1-camps FRET were measured at several time points (i.e., 1, 3, 5, 10, or 30 min) after OA or HL3 (control) treatment. For representative images of CFP and YFP emission in a fire scale, see Figures 4, A and B, and 5C, which exemplify relative cAMP levels.
Experimental design and statistical analyses.
The 4- to 7-d-old flies of mixed sex were used for all experiments, with the exception of the genetic interaction experiment involving rut. As noted above, rut is on the X chromosome; thus, only female rut heterozygotes and female rut and octβ1r double heterozygotes were used. All behavioral experiments were performed blindly to the experimenter, and the control and experimental groups were tested in the same session in a randomized order. Multiple independent sets of flies obtained from different crosses and cultures were used for behavioral analyses. Statistical analyses were performed using the JMP (SAS) and Minitab 16 (Minitab). The raw data were analyzed using the Anderson–Darling goodness-of-fit or Shapiro–Wilk normality test for distribution and the Levene's test for equality of variances, and are reported as the mean ± SEM. The normally distributed data were analyzed by a two-tailed Student's t test for two groups or ANOVA followed by post hoc Tukey's multiple-comparison or Dunnett's tests for three or more groups. The non-normally distributed data were analyzed by Kruskal–Wallis and post hoc Mann–Whitney tests. Significant difference among the groups under comparison was determined using an α level of 0.05 in all analyses. All raw data files are available on request.
Results
Octβ1R is essential for aversive learning
To clarify whether OA is involved in aversive learning, we used the RNA interference (RNAi) approach to reduce expression of the OA biosynthetic enzyme TβH. The TβH knock-down flies were then examined in olfactory conditioning to assess their avoidance of the odor associated with electric shock. The GAL4 drivers expressed in a majority of OA neurons (tdc2-GAL4) and a pair of the tdc2-GAL4-negative anterior paired lateral (APL; APL-GAL4) OA neurons (Busch et al., 2009; Wu et al., 2013) were used for TβH RNAi. Compared with the control pBDP-GAL4 (promoterless GAL4)/UAS-tβh-RNAi flies, both tdc2-GAL4/UAS-tβh-RNAi and APL-GAL4/UAS-tβh-RNAi flies performed poorly when tested right after training (ANOVA: F(2,27) = 31.4, p < 0.0001; Fig. 1A). To clarify whether developmental or adulthood TβH knockdown (KD) is responsible for poor performance, we used elav-GAL4,GAL80ts (Kim et al., 2007a) to induce TβH knockdown only during development, only during adulthood, or throughout development and adulthood. The flies with developmental TβH knockdown showed normal performance in olfactory conditioning (t test: t = −0.12, p = 0.9074; Fig. 1B, developmental KD) similar to the flies without knockdown (t test: t = −0.11, p = 0.9168; Fig. 1B, no KD). The flies with adulthood TβH knockdown, however, displayed poor performance (t test: t = −5.55, p < 0.0001; Fig. 1B, adulthood KD) similar to the flies with constant knockdown (t test: t = t test 5.7, p < 0.0001; Fig. 1B, development and adulthood KD). These results together indicate that OA is important for aversive olfactory conditioning and that multiple OA neurons are involved in the process.
OA and its receptor Octβ1R are required for aversive olfactory learning. A, TβH knockdown (UAS-tβh RNAi) in either OA/TA (tdc2-GAL4) or APL (APL-GAL4) neurons led to significantly impaired learning compared with the control (pBDP-GAL4) in aversive olfactory conditioning (***p < 0.0001, n = 10). B, Uninduced (no KD) or developmental TβH knockdown had no effect on olfactory learning; however, adulthood or constant (development and adulthood KD) TβH knockdown led to impaired learning (KD; ***p < 0.0001, n = 12). C, Top, A scheme depicting the approximate location of the piggyBac insertion (a gray broken line with the filled triangle; SA, splice acceptor) in the second intron of the octβ1r gene along with the Octβ1R translation start site (gray arrow), coding regions (orange boxes), noncoding regions (gray boxes), and introns (gray lines; not scaled). Bottom left, The RT-PCR analysis on the RNA isolated from the heads of Canton-S and octβ1r flies showed clear reduction in Octβ1R mRNA in the octβ1r flies. Rp-49 was used as an internal control. Bottom right, Quantitative PCR analysis revealed ∼75% reduction of Octβ1R mRNA in the octβ1r flies compared with the control Canton-S (**p < 0.01, n = 3). D, octβ1r flies exhibited defective performance immediately (3 min), 1 h, and 3 h after training in aversive conditioning (**p < 0.001, ***p < 0.0001, n = 6), but octβ2r and Canton-S flies showed comparable performance at all time points examined. E, The homozygous octβ1r mutant or the flies with the transheterozygous octβ1r and Df alleles [Df-E, Df(3R)Exel6191; Df-B, Df(3R)BSC685] displayed significantly reduced learning scores compared with either w control or the heterozygotes octβ1r/+, Df-E/+, and Df-B/+ (***p < 0.0001, n = 6). F, The learning performance of both Canton-S and the homozygous octβ1r mutant flies was significantly improved after the second submaximal training (2T) with 15 min ITI in the short program of olfactory conditioning (***p < 0.0001, n = 12).
Next, we investigated which OA receptor is important. The α1-like OAMB receptor is not crucial for aversive learning (Kim et al., 2013); thus, we focused on the β-like OA receptors Octβ1R and Octβ2R. The octβ1r allele contains a piggyBac transposon in the second intron of the octβ1r gene with a splicing acceptor in the transposon in the right orientation to intercept splicing of octβ1r transcripts and generate a truncated protein (Fig. 1C). Quantitative RT-PCR analysis on the mRNA sequence spanning the exons 4 and 5 showed a reduced level of Octβ1R mRNA in octβ1r flies (one-sample t test: t = −29.533, p = 0.0011; Fig. 1C), indicating that octβ1r is a hypomorphic allele. We have previously shown that the octβ2r mutant is a null allele for Octβ2R (Lim et al., 2014). When tested immediately (3 min), 1 or 3 h after electric-shock olfactory conditioning, the octβ2r flies showed comparable performance and memory decay to the wild-type Canton-S, whereas the octβ1r flies displayed substantially poor performance right after training and subsequent time points (two-way ANOVA: F(8,45) = 50.64, p < 0.0001; genotype effect: F = 118.4, p < 0.0001; time effect: F = 64.5, p < 0.0001; interaction effect: F = 9.8, p = 0.0001; Fig. 1D). These data suggest that aversive olfactory learning requires Octβ1R but not Octβ2R.
To ascertain the phenotype of octβ1r to the octβ1r locus, we tested the octβ1r transheterozygotes over the Df lines Df(3R)Exel6191 and Df(3R)BSC685 containing complete deletion of the octβ1r gene. The heterozygotes octβ1r/+, Df(3R)Exel6191/+, and Df(3R)BSC685/+ had learning scores comparable to that of Canton-S; however, the transheterozygotes octβ1r/Df(3R)Exel6191 and octβ1r/Df(3R)BSC685 performed poorly like the octβ1r homozygotes (ANOVA: F(6,35) = 33.39, p < 0.0001; Fig. 1E). This substantiates that the loss of octβ1r function rather than background genetic loci accounts for the learning phenotype. Two submaximal training trials (2T) with 15 min rest (ITI) in the short program of olfactory conditioning are known to improve the performance of the learning mutant rut (Beck et al., 2000). As in the regular electric shock conditioning, the octβ1r flies displayed impaired performance compared with Canton-S in the 1T short program (Fig. 1F; two-tailed t test: t(22) = −11.69, p < 0.0001). The 2T training with 15 min ITI, however, improved performance in the octβ1r flies (Fig. 1F; two-tailed t test on 1T vs 2T: t(22) = 6.47, p < 0.0001) similar to rut, further supporting the role of Octβ1R in aversive olfactory learning. We next asked whether the learning phenotype could be due to abnormal sensory modalities. When tested with two different odor concentrations or shock intensities, Canton-S and octβ1r flies showed comparable olfactory acuity and shock avoidance (Table 1). Together, the results support that the function of Octβ1R is needed to acquire aversive memory of olfactory information associated with electric shock.
Sensory modality tests including olfactory acuity, shock avoidance and sugar preference
The site of the function of Octβ1R for aversive memory formation is the MB αβ neurons
The genomic enhancer-driven reporter expression and single-cell RNA-sequencing studies report Octβ1R expression in many neural structures including the PN and the MB (Pfeiffer et al., 2008; Li et al., 2017; Davie et al., 2018). To identify the functional site of Octβ1R, we performed rescue experiments using the GAL4/UAS system (Brand and Perrimon, 1993). We targeted the transgenic Octβ1R expression in the octβ1r mutant background pan-neuronally or in the second-order neurons (PN carrying olfactory information from the antennal lobe to MB) or third-order neurons (MB) for the olfactory pathway, both of which are key neural structures for olfactory learning and memory (Guven-Ozkan and Davis, 2014). Pan-neuronal Octβ1R expression (elav-GAL4) fully reinstated the learning performance of octβ1r (ANOVA: F(8,45) = 42.21, p < 0.0001; Fig. 2A). When targeted in a subset of neural structures, Octβ1R in all MB neurons (OK107-GAL4), but not in the PN neurons (NP225-GAL4), led to full rescue of the octβ1r learning phenotype (Fig. 2A). The MB has three distinct substructures, namely αβ, α′β′, and γ neurons (Crittenden et al., 1998; Aso et al., 2014a), and they are engaged in distinct functions (Guven-Ozkan and Davis, 2014; Cognigni et al., 2018). The cell type-level transcriptome analysis indicates that Octβ1R is abundant in all MB substructures (Shih et al., 2019). To narrow down the functional site within the MB, we used the substructure-specific GAL4 drivers along with the MB247-GAL4 expressed in αβ and γ for rescue experiments (Fig. 2A). The octβ1r flies with reinstated Octβ1R expression in the αβ neurons (MB247-GAL4 and c739-GAL4) exhibited Canton-S-like performance (Fig. 2A), indicating full rescue. The Octβ1R re-expression in other MB substructures, namely α′β′ (c305a-GAL4) or γ (NP1131-GAL4), had no effect (Fig. 2A). These results pinpoint the MB αβ neurons as the major functional substrate where Octβ1R processes aversive learning.
Octβ1R in the MB αβ neurons promotes acquisition of aversive memory. A, Restored Octβ1R (UAS-Octβ1R) in all (elav-GAL4), pan-MB (OK107-GAL4), MB αβγ (MB247-GAL4), or MB αβ (c739-GAL4) neurons rescued the learning defect of octβ1r (p < 0.0001, n = 6). In contrast, the flies with Octβ1R re-expression in other MB neurons (MB γ, NP1131-GAL4; or MB α′β′, c305a-GAL4) or the PN (NP225-GAL4) performed poorly similar to octβ1r mutants (p > 0.05, n = 6). B, Normal aversive learning was reinstated in octβ1r mutants when Octβ1R was restored during adulthood (induced, temperature shift to 30°C; uninduced, no temperature shift; p < 0.0001, n = 13), but not during development (p > 0.05, n = 12). In this and other figures, letters on the bars denote either a significant difference among groups (different letters, i.e., “a” and “b”) or no difference (same letters).
Multiple lines of evidence show persistent Octβ1R expression from late embryonic to adult stages (Chintapalli et al., 2007; Graveley et al., 2011; Brown et al., 2014). To assess whether the phenotype of the octβ1r is attributable to developmental or physiological anomaly, we restored Octβ1R expression in the octβ1r mutant either during development or adulthood using the TARGET system (McGuire et al., 2003; Kim et al., 2007a, 2013). For induction of Octβ1R expression, the octβ1r flies carrying elaGal80ts/UAS-Octβ1R were reared at 30°C either during development or at the adulthood while the uninduced controls were kept at 20°C throughout development and adulthood (Fig. 2B). Octβ1R induced at the adult stage rescued the impaired learning phenotype of octβ1r in aversive conditioning (ANOVA: F(3,48) = 50.64, p < 0.0001; Fig. 2B), whereas induction only during development failed to do so (ANOVA: F(3,44) = 43.64, p < 0.0001). These data indicate that the function of Octβ1R in aversive learning is physiological as opposed to developmental.
OA–DA interplay and intracellular signaling crucial for aversive learning
Our earlier studies (Kim et al., 2007a, 2013) demonstrate that aversive learning requires the MB DA receptor dDA1 but not the MB OA receptor OAMB. Since the MB is a functional site for Octβ1R as well, we asked whether Octβ1R and dDA1 act in collaboration for aversive learning by using a genetic interaction approach. When subjected to electric shock conditioning, all heterozygous receptor mutants showed normal learning, indicating the recessive nature of octβ1r and dumb in this process. The octβ1r/dumb transheterozygotes, on the contrary, displayed significantly impaired performance (ANOVA: F(5,36) = 13.62, p < 0.0001; Fig. 3A). This was not due to a simple combination of two heterozygous mutations since the octβ1r/oamb transheterozygotes showed normal learning. These data suggest that dDA1 and Octβ1R have parallel functions to process aversive learning.
Interaction between Octβ1R and dDA1 concertedly drives aversive learning via cAMP. A, The flies with the transheterozygous mutations in octβ1r and dumb (a mutant allele for dDA1) showed impaired learning (octβ1r/dumb; p < 0.0001, n = 7), but the octβ1r and oamb transheterozygotes (octβ1r/oamb) were comparable to either Canton-S control or the heterozygotes octβ1r/+, dumb/+, and oamb/+ (p > 0.05, n = 7). B, The rut/+;;octβ1r/+ double heterozygotes displayed impaired acquisition similar to octβ1r mutants (p > 0.05, n = 8–9).
Like mammalian β-adrenergic receptors, Octβ1R stimulates an increase in the intracellular cAMP level when studied in heterologous cells in vitro (Maqueira et al., 2005). The cAMP signaling pathway is an integral cellular mechanism for olfactory learning in flies (Guven-Ozkan and Davis, 2014). We conducted ex vivo imaging to visualize and quantify the changes in cAMP levels in the MB lobes of Canton-S and octβ1r. For the task, we used the pan-MB OK107-GAL4 to drive the cAMP sensor Epac1-camps, where an increase in cAMP causes a decrease in FRET between CFP and YFP (Nikolaev et al., 2004). We first determined the timeline of the cAMP response in the wild-type brain expressing Epac1-camps in the MB on 100 μm OA or HL3 (control) treatment. The maximal increase in cAMP levels was evident within 1 min after OA treatment and sustained at least for 5 min (Fig. 4A); thus, we measured the cAMP response at 3 min after OA treatment in subsequent experiments. Compared with the control, the OA-induced cAMP response was substantially dampened across all MB subregions examined (α3/α′3, α2/α′2, α1/α′1/γ2, β1/2, and γ3–5), except for β′1/2 and γ1/heel (Fig. 4B,C). To clarify whether cAMP serves as a major downstream effector of Octβ1R for olfactory learning, we examined the genetic interaction of octβ1r and rut defective in the adenylyl cyclase (AC) that converts ATP to cAMP. Similar to octβ1r/+, the rut heterozygotes (rut/+) showed normal performance but the rut/+;;octβ1r/+ double heterozygotes exhibited poor learning (Fig. 3B). These observations corroborate that OA signaling through Octβ1R recruits cAMP to promote aversive learning.
The octβ1r mutant have the dampened cAMP response in the MB on OA treatment. A, Left, Representative confocal images of CFP and YFP emission in the MB lobes of the control (UAS-Epac1-camps/+;;OK107/+) brains at 3 min after HL3 (vehicle; baseline) or OA treatment. The MB lobes are outlined by broken white lines. Right, The cAMP response was measured at various time points after vehicle (baseline) or OA treatment in the MB lobes of the control (UAS-Epac1-camps/+;;OK107/+) brains (***p < 0.0001, n = 4–6). The letter codes (“a,” “b,” and “c”) denote significant differences (p < 0.0001) among the OA-induced cAMP responses at different time points. Scale bar, 25 μm. B, Representative confocal images depicting the cAMP level (represented by CFP emission shown in a fire scale) between baseline and OA treatment in the brains of the control (UAS-Epac1-camps/+;;OK107/+) or octβ1r mutant (UAS-Epac1-camps/+; octβ1r;OK107/+). OA treatment drastically increased cAMP levels (elevated CFP signal) in the control but not in the octβ1r MB. Shown are subregions (α3/α′3; α2/α′2; β1–2 and β′1–2; γ1/heel; γ2/α1/α′1 and γ3–5) of the MB lobes that are delineated by broken white lines. The images and quantifications were performed on the brains at 3 min after vehicle or OA treatments. Scale bar, 25 μm. C, Quantification of the peak cAMP response revealed that all subregions except for the β′1–2 and γ1/heel in the octβ1r MB had significantly dampened responses to OA (*p < 0.05, **p < 0.01, n = 10).
Octβ1R in the PN plays a major role in appetitive learning
OA is a major transmitter for appetitive conditioning (Schwaerzel et al., 2003; Kim et al., 2007b) and it's activity through α1-like OAMB in the MB is indispensable for appetitive memory formation (Kim et al., 2013). We asked whether β-like OA receptors are also important for appetitive learning. The Canton-S, octβ1r, and octβ2r flies were conditioned with 2 m sucrose as US and the same odorants were used for aversive conditioning as CS. Canton-S and octβ2r showed comparable performance when tested immediately, 1 h, and 3 h after training; however, octβ1r displayed poor performance at all time points examined (two-way ANOVA: F(11,60) = 22.29, p < 0.0001; genotype effect: F = 25.26, p < 0.0001; time effect: F = 58.82, p < 0.0001; interaction effect: F = 3.03, p = 0.0118; Fig. 5A). The impaired performance was not due to anomalous sugar perception (Table 1) or genetic background since the transheterozygotes octβ1r/Df(3R)Exel6191 and octβ1r/Df(3R)BSC685 also performed poorly like the octβ1r mutant (data not shown). These results illustrate the crucial role of Octβ1R for acquisition of appetitive memory as well. The impaired appetitive learning phenotype of octβ1r was fully rescued when Octβ1R was reinstated in all neurons (elav-GAL4) or the PN (NP225-GAL4) in the octβ1r background (ANOVA: F(6,77) = 31.01, p < 0.0001; Fig. 5B). Interestingly, the learning defect of octβ1r was slightly improved when Octβ1R was restored in the MB αβ, α′β′, or γ neurons (Fig. 5B). These data suggest that Octβ1R has redundant functions in multiple neural sites for appetitive learning with quantitatively different contributions, but its major functional site is the PN. Consistently, the OA-induced cAMP levels were significantly dampened in the PN dendrites at the AL and axons at the MB calyx area (Fig. 5C,D; dendrites: t(22) = −8.52, p < 0.0001; axons: t(22) = −13.64, p < 0.0001, two-tailed t test).
The action of Octβ1R in the PN fully supports appetitive learning. A, octβ1r mutants showed defective performance immediately (3 min), 1 h, and 3 h, but not 6 h, after training in appetitive olfactory conditioning (*p < 0.05, **p < 0.001, n = 6). B, Restored Octβ1R (via UAS-Octβ1R) expression in either all neurons (elav-GAL4) or the PN (NP225-GAL4) fully rescued the learning phenotype of octβ1r (the letter code “a”; p > 0.05, n = 12). However, the octβ1r flies with Octβ1R re-expression in individual MB subsets (γ, NP1131-GAL4; αβ, c739-GAL4; α′β′, c305a-GAL4) showed marginal yet significant improvement in olfactory learning (the letter codes “c” and “b”; p < 0.05, n = 12). C, Representative confocal images showing the cAMP levels in the PN dendrites and axons of the baseline versus OA-treated control (NP225-GAL4/+;UAS-Epac1-camps/+) and octβ1r mutant (NP225-GAL4/UAS-Epac1-camps;octβ1r). OA treatment increased cAMP levels (elevated CFP signal) in both PN dendrites and axons to a lesser extent in the octβ1r mutant compared with the control. Scale bar, 25 μm. D, Quantified peak cAMP responses. The PN dendrites and axons of octβ1r had significantly dampened responses to OA (***p < 0.0001; n = 12). E, The flies with the transheterozygous mutations in octβ1r and dumb (octβ1r/dumb) or octβ1r and oamb (octβ1r/oamb), or the double heterozygous mutations in octβ1r and rut (rut/+;;octβ1r/+) showed impaired appetitive learning compared with Canton-S or the heterozygotes octβ1r/+, dumb/+, and oamb/+ (***p < 0.0001, n = 7–9).
To elucidate the relative contributions of Octβ1R and OAMB to appetitive learning, we examined genetic interaction. Unlike the individual octβ1r/+ and oamb/+ heterozygotes that showed normal appetitive learning, the octβ1r/oamb transheterozygote showed drastically diminished learning, which is comparable to that of the homozygous octβ1r flies (ANOVA: F(8,56) = 68.91, p < 0.0001; Fig. 5E). The DA receptor dDA1 plays a significant role in appetitive memory acquisition (Kim et al., 2007a); thus, we were curious whether Octβ1R and dDA1 work together for appetitive learning. Similar to the octβ1r and oamb genetic interaction, the octβ1r/dumb transheterozygote exhibited poor learning (Fig. 5E). These findings indicate that the concerted actions of Octβ1R, OAMB, and dDA1 mediate appetitive learning.
Discussion
Historically, OA is known to mediate reward-based but not punishment-based learning in Drosophila and other invertebrates including honeybees, crickets, and crabs (Farooqui et al., 2003; Schwaerzel et al., 2003; Kaczer and Maldonado, 2009; Mizunami and Matsumoto, 2017). This selective role of OA was first demonstrated by Schwaerzel et al. (2003) based on the observation that tβh mutants are defective in appetitive learning, but not aversive learning. Recently, Iliadi et al. (2017) challenged this notion and showed that tβh mutants are impaired in aversive learning. Both studies examined the same mutant allele and used the same conditioning parameters including CS odorants and US electric shock intensities; thus, the discrepancy in their findings remains puzzling. In this report, by adopting an RNAi approach, we provide the independent evidence that substantiates the finding of Iliadi et al. (2017) and further demonstrate that more than one type of OA neuron processes aversive learning through Octβ1R in the MB αβ neurons.
OA neurons have widespread projections in the brain and are classified into 28 types based on their projection patterns, as follows: 27 types positive for tdc2-GAL4 and 1 type [i.e., anterior paired lateral (APL; APL-GAL4) neurons] negative for tdc2-GAL4 (Busch et al., 2009; Wu et al., 2013). Our data indicate that both tdc2-GAL4 and APL OA neurons are important for aversive learning. Notably, the MB is the only brain structure innervated by both tdc2-GAL4 and APL neurons. Four types of tdc2-GAL4 OA neurons (namely, VPM-3, VPM-4, VPM-5, and VUMa2) project to the calyx where the MB receives olfactory input and the γ lobes comprising MB axons (Busch et al., 2009), whereas the APL neurons innervate all MB structures where they are both presynaptic and postsynaptic (Liu and Davis, 2009; Wu et al., 2013). The APL neurons use both OA and GABA as neurotransmitters and engage in a feedback loop wherein the MB activates the APL and the APL inhibits the MB, thereby controlling odor discrimination (Liu and Davis, 2009; Wu et al., 2013; Lin et al., 2014). This feedback loop is modulated by DA, which suppresses the APL through dD2R for CS processing (Zhou et al., 2019). It is tempting to speculate that the APL usesOA output for CS processing through Octβ1R in the αβ neurons. The tdc2-GAL4-positive OA neurons mediating olfactory conditioning have not been identified in Drosophila; however, in the honeybee, the OA-VUMmx1 neuron projecting to the calyx and the antennal lobe processes appetitive US (Hammer, 1993). The fly homolog of VUMmx1 is the OA-VUMa2 neuron that is positive for tdc2-GAL4 (Busch et al., 2009), serving as a likely candidate to process appetitive US. There is no information on the OA-VPM neurons or their homologs in other insects; nonetheless, their projection onto the MB calyx and lobes makes them potential candidates to process aversive US. This predicts that discrete OA neurons take part in the CS and US pathways for aversive learning, and this notion is consistent with the finding by Boto et al. (2019) that two distinct types of DA neurons independently regulate the CS and US pathways.
The modes of the OA and GABA transmissions of the APL (i.e., whether they are released together or independently and whether they act on the same or different postsynaptic sites) are unknown. It is thus difficult to assign the effectsof synaptic activity manipulations (e.g., blockade via Shits and activation via TrpA1) to OA or GABA actions. However, the Gad and Rdl RNAi studies indicate that selective suppression of the APL GABA output or the GABA input to the MB enhances aversive learning (Liu et al., 2009; Liu and Davis, 2009), which is contrary to our finding that selective suppression of the APL OA output or the OA input to the MB through Octβ1R impairs aversive learning. The opposite actions of OA and GABA released from the APL onto the MB implicate that their release dynamics or functional sites may be distinct. In addition to aversive learning, the APL OA signal modulates amnesia-resistant memory (ARM) consolidation and expression through Octβ2R in the MB α′β′ neurons (Wu et al., 2013; Yang et al., 2016). We found normal 3 h memory of octβ2r but impaired 3 h memory of octβ1r mutants, which are different from the findings (normal 3 h ARM of octβ1r and impaired 3 h ARM of octβ2r mutants) of the Wu et al. (2013) study. This could be due to several differences that include CS odorants, the nature and genetic backgrounds of octβ1r and octβ2r mutants (RNAi-mediated knockdown causing 40–60% reduction in mRNA levels in the study by Wu et al., 2013), and the methods to measure 3 h memory (ARM measured after cold shock). The follow-up studies incorporating these factors would clarify the differences in 3 h memory. Together, the APL OA signals promote aversive olfactory learning and ARM memory through separate β-like receptors located in the distinct MB subsets, namely Octβ1R in αβ and Octβ2R in α′β′, respectively.
OA increases cAMP in all MB and PN dendritic and axonal structures (Tomchik and Davis, 2009), which we confirmed in this study. We further demonstrate that the OA-induced increase in cAMP is mediated by Octβ1R in all MB and PN, which is the first to clarify Octβ1R as an upstream signal for the cAMP increase and the first to substantiate the in vitro analysis on Octβ1R (Maqueira et al., 2005) ex vivo. Similar to OA, DA activates the increase in cAMP in all MB structures (Tomchik and Davis, 2009). Notably, both Octβ1R and dDA1 are present in all MB neurons (Kim et al., 2003; Shih et al., 2019), yet their activities only in the αβ and γ neurons, respectively, are required for aversive learning (this study; Kim et al., 2007a; Qin et al., 2012). This could be due to selective OA/DA input that exclusively activates αβ-Octβ1R/γ-dDA1 or selective activity of the MB output neurons (MBONs) in αβ/γ (Aso et al., 2014a) that convey the Octβ1R/dDA1 information. Together, our study indicates that aversive learning is processed by two parallel circuits consisting of the well characterized DA (PPL1γ1pedc)-dDA1 (MB-γ)-MBON (MBON-γ1pedc> α/β, MBON-γ5β′2aβ′2mp; Aso et al., 2014b; Owald et al., 2015; Yamazaki et al., 2018), and the OA-Octβ1R (MB-αβ)-MBON (this study) pathways (Fig. 6). The MBONs with dendrites in αβ (7 MBON types of a total of 22; Aso et al., 2014b) are good candidates to process the OA-Octβ1R information for aversive learning. Supporting this notion, the MBON-β2β′2a activity modulates olfactory conditioning (Owald et al., 2015). Alternatively, the MBONs with dendrites in αβ as well as γ may convey both Octβ1R and dDA1 information for aversive learning. For example, MBON-γ1pedc>α/β has dendrites in γ1pedc and αβs and axons in αβ lobes, crepine, and superior intermediate protocerebrum, and its activity modulates aversive learning (Perisse et al., 2016), thus serving as a good candidate conveying both Octβ1R and dDA1 signals for learned avoidance behavior. An anomaly in either Octβ1R or dDA1 signaling to MBON-γ1pedc>α/β would lead to poor aversive learning.
Working model for the pathways that OA and DA process aversive and appetitive olfactory learning. In aversive conditioning, the OA and DA neurons deliver the aversive US information to Octβ1R in the MB αβ and dDA1 in the MBγ, respectively, which, in concert with the CS input, activate the coincidence detector Rut-AC. The parallel cAMP signaling in the αβ (1, red box) and γ (2, red box) modulates the MB outputs to MBONs promoting avoidance behavior. Abnormal signaling in either OA-Octβ1R in αβ or DA-dDA1 in γ leads to defective aversive learning. In appetitive conditioning, the collective activation of OA-Octβ1R in the PN (3, blue box), which projects to the MB, as well as OA-OAMB and DA-dDA1 in the MB αβγ (4, blue box) recruit both Rut-AC and phospholipase C (PLC) to stimulate the cAMP and Ca2+ signaling, which in turn modulate the MB outputs to the MBONs promoting approach behavior. Thus, the OA and DA systems act together for aversive and appetitive learning through distinct neural sites and intracellular signaling.
Rut-AC is a coincidence detector for associative learning and memory (Guven-Ozkan and Davis, 2014). We show that octβ1r and rut are genetically interacting for aversive learning. The previous findings on the functional site of Rut-AC are somewhat confusing. Blum et al. (2009) identify γ neurons as the site that Rut-AC mediates aversive learning, whereas Zars et al. (2000) and Akalal et al. (2006) demonstrate that Rut-AC in both γ and αβ neurons is important. The major difference is the rut allele: Blum et al. (2009) used rut1 but Akalal et al. (2006) and our study used rut2080, which displays a more severe learning deficit than rut1. Thus, Rut-AC likely serves as a key downstream effector for both Octβ1R and dDA1 in αβ and γ, respectively. Regarding the downstream effectors of cAMP, Octβ1R, and dDA1 seems to involve distinct signaling pathways. Gervasi et al. (2010) demonstrate that OA activates protein kinase A (PKA) in αβ in a Rut-AC-dependent manner but that DA does not induce PKA activation in γ. In vivo functional imaging, on the contrary, reveals that an elevated cAMP level leads to enhanced calcium responses to CS odors in γ but not αβ (Boto et al., 2014). It is tempting to speculate that the γ dDA1–cAMP pathway directly activates cyclic-nucleotide gated ion channels while the αβ Octβ1R-cAMP pathway recruits PKA to modify ion channel activity or other downstream effectors for aversive learning.
OA is a major neuromodulator for appetitive learning, and α1-like OAMB in αβγ is a key receptor conveying the OA signal (Kim et al., 2013). Here, we show that Octβ1R is also important for appetitive learning, and this dual function is similar to dDA1 (Kim et al., 2007a). However, there is a major difference: dDA1 processes both aversive and appetitive learning in MBγ, but the functional sites of Octβ1R are distinct (i.e., MBαβ for aversive learning and PN for appetitive learning). Thum et al. (2007) demonstrate that Rut-AC restored in either MB or PN fully rescues the phenotype of the rut mutant in appetitive conditioning, suggesting multiple yet redundant neural sites for Rut-AC-dependent learning. It is yet unclear whether dDA1 recruits Rut-AC for appetitive memory; however, our study clarifies that the PN Rut-AC is activated by Octβ1R for the acquisition of appetitive memory. Together, OA mediates positively reinforcedlearning through the following two parallel circuits: the OA-Octβ1R–Rut-AC pathway in PN and the OA-OAMB–intracellular calcium pathway in MB αβγ.
Footnotes
The authors declare no competing financial interests.
This work was supported by US Department of Agriculture Grant NIFA2010-65 105–20 625, Brain & Behavior Research Foundation NARSAD grants, National Institutes of Health (NIH) National Institute of Mental Health Grant R21-MH-109953, and National Institute on Minority Health and Health Disparities Neuromodulation Disorders Cluster Grant 2G12MD007592. We thank the Bloomington Stock Center; the Exelixis Collection, and Drs. David Anderson (California Institute of Technology, Pasadena, CA), Josh Dubnau (Stony Brook University School of Medicine, Stony Brook, NY), Ann-Shyn Chiang (National Tsing Hua University, Hsinchu, Taiwan), Andreas Thum (University of Leipzig, Leipzig, Germany) and Scott Waddell (University of Oxford, Oxford, UK) for sharing fly lines; the Cytometry, Screening and Imaging Core at the Border Biomedical Research Center for help on confocal microscopy; the NIH-funded MARC program (Grant 5T34-GM-008048) for supporting J.M.S.; and a Keelung Hong Fellowship for supporting P.R.S. We also thank past and current laboratory members for their discussion and support.
- Correspondence should be addressed to Kyung-An Han at khan{at}utep.edu
