γ neurons mediate dopaminergic input during aversive olfactory memory formation in Drosophila

SUMMARY

Mushroom body (MB) dependent olfactory learning in Drosophila provides a powerful model to investigate memory mechanisms. MBs integrate olfactory conditioned stimuli (CS) inputs with neuromodulatory reinforcement (unconditioned stimuli, US) [1, 2], which for aversive learning is thought to rely on dopaminergic (DA) signaling [3–6] to DopR, a D1-like dopamine receptor expressed in MB [7, 8]. A wealth of evidence suggests the conclusion that parallel and independent signaling occurs downstream of DopR within two MB neuron cell types, with each supporting half of memory performance. For instance, expression of the rutabaga adenylyl cyclase (rut) in γ neurons is sufficient to restore normal learning to rut mutants [9] whereas expression of Neurofibromatosis I (NFI) in α/β neurons is sufficient to rescue NF1 mutants [10, 11]. DopR mutations are the only case where memory performance is fully eliminated [7], consistent with the hypothesis that DopR receives the US inputs for both γ and α/β lobe traces. We demonstrate, however, that DopR expression in γ neurons is sufficient to fully support short (STM) and long-term memory (LTM). We argue that DA-mediated CS-US association is formed in γ neurons followed by communication between γ and α/β neurons to drive consolidation.

RESULTS

DopR expression in MB Kenyon cells (KCs) is sufficient to fully support both early and late phases of memory

In order to localize the site of CS-US association within MB, we took a complementary approach to previous attempts at mapping the DA neurons that convey the US inputs to MB [3, 4, 6, 12]. We used restricted expression of DopR to map the subset of MB neurons in which the US information is received. This strategy takes advantage of the fact that mutations in DopR cause a complete loss of STM performance [7]. We used an established panel of Gal4 lines [each of which yield expression either in all MB Kenyon cell types, or specific subsets (Fig S1A)] to drive expression and tested for rescue of the memory defects of two null alleles of DopR: dumb1, which is an inversion line In(3LR)234 whose break points were mapped to 67D and 88A-88B [13], and dumb2, which is caused by a piggyBac insertion, PBac{WH}DopR^f02676 in the first intron of the DopR locus [7]. PiggyBac (WH) carries a UAS enhancer for Gal4-driven expression of the flanking adjacent gene [14]. We first confirmed (Fig S1) the finding that expression from this UAS enhancer using the MB247 Gal4 line is sufficient to fully rescue dumb2 STM deficit [7, 15]. Because all Gal4 lines that express in MB also yield some expression in other cell types, we also performed a series of experiments to corroborate the conclusion that post-development MB expression per se is sufficient to rescue the STM defect of DopR mutants (Figs S1 A–I). We also established that the rescued memory observed with Gal4 driven over-expression in MB shares a genetic signature of normal memory, which is that it can be dissected into a rut-dependent and rut-independent component (Fig. S1E). This supports the contention that the DopR rescued memory shares established features of “normal” memory.

We next examined the relationship between DopR expression and two different forms of consolidated memory. We used 10 repeated training sessions, either massed together with no rest interval (Massed training) or spaced out with a 15 min rest between sessions (Spaced training). Massed training induces a long-lived anesthesia resistant memory (ARM) [16, 17]. Spaced training, in contrast, induces a long-term memory (LTM) that requires CREB-dependent gene expression [16–18]. We first tested memory 24-hours after massed and spaced training and found that both dumb1 and dumb2 mutants are devoid of all memory performance at these time points (Fig 1). We then asked if MB-driven DopR expression is sufficient to support these forms of consolidated memory. For these experiments, we used Gal4 lines ok107 and MB247, which expressed in more than one MB cell subtype. With ok107, we were able to fully restore 24hr memory after either spaced (Fig. 1C) or massed training (Fig. 1D). DopR expression driven by MB247Gal4 also rescued the memory performance after both types of training (Fig. 1C, D). These data indicate that as with STM, DopR expression in MB is sufficient to support both forms of consolidated memory, LTM and ARM (See below for additional evidence).

Figure 1. DopR expression in MB is sufficient to support both ARM and LTM.

In this and the following figures, means and standard errors are shown for all groups; wild type flies were w¹¹¹⁸(isoCJ1), unless otherwise noticed. (A) An anterior view of the Mushroom Body in the left hemisphere is illustrated to show the structural features of the three major neuronal cell types. The three major MB Kenyon Cell subtypes are color coded with red (α/β), yellow (α′/β′) or blue (γ). The Kenyon Cells have their cell bodies in the posterior cortex and send projects anteriorly. The dendrites arborize into the calyx. The axons project into 5 distinct lobes, including two vertical lobes (α and α′) and three medial lobes (β, β′ and γ). (B) Both dumb1 and dumb2 exhibit severely defective LTM (24 hours after 10X spaced training) (p<0.05, n=8 for all groups). In the left panel, the WT control for dumb1 was Canton-S. (C) dumb2 flies with either ok107 or MB247 MB Gal4 drivers, exhibit significantly improved LTM performance (24 hours after 10X spaced training) compared with dumb2 control flies (* p<0.05, n=16 for all groups). (D) Expression with the ok107 or MB247 Gal4 drivers also significantly improved the performance of dumb2 flies 24 hours after 10X massed training (*p<0.05, n=15 for all groups). The performance with ok107 is not significantly different from that of WT (for spaced training, p>0.05, n=16; for massed training, p>0.05, n=15).

Acute expression of DopR in γ KCs alone is sufficient to support all STM

MB KCs can be classified into three major cell subtypes, each of which send axon projections into distinct MB lobes. The axons of γ neurons project horizontally to form γ lobes. The axons of α/β neurons bifurcate to form a vertical α lobe and a horizontal β lobe. The axons of α′/β′ neurons also bifurcate into a vertical and horizontal branch, but with terminals in somewhat distinct spatial domains from α/β neurons [19]. In order to dissect the cell-type requirements for DopR, we used a panel of Gal4 lines that distinguish the three major sub-divisions of the MB (Fig. S1A). For α′/β′ lobe expression, we used c305a and g0050 (Fig S1A; [20–22]), which label about half (c305a) or all (g0050) α′/β′ neurons without labeling the other KC classes. For γ lobe expression, we used NP1131 and 201Y each of which yield expression in most of the γ neurons, although the expression level with 201Y is relatively low (Fig S1A; [20, 23–25]). NP1131, but not 201Y also labels a small portion of α′/β′ neurons. 201Y, but not NP1131 labels a small set of core α/β neurons. For α/β lobe expression, we used c739 and NP3061, each of which label all or most of the α/β lobe without expression in other lobes (Fig. S1A; [20, 24, 25]). These cell-type specific Gal4 drivers generally yield less intense labeling than drivers expressed in multiple MB cell types, such as MB247, ok107, c747 and c309 (Fig. S1A; [9, 20, 25, 26]). Since the rescue is sensitive to DopR expression level (Fig. S1I), we used a UAS-DopR-cDNA transgene [27] to supplement the expression derived from the dumb2 PBac element.

We targeted DopR expression to each MB cell-type with lobe specific Gal4 drivers and tested for STM performance. Remarkably, we find that expression in just γ neurons fully restored STM to dumb2 mutants. When NP1131 alone is used to drive DopR expression in γ neurons, STM is fully restored to wild-type levels (Fig. 2A,B) that also are equivalent to performance seen with the combination of NP1131 (γ) and NP3061 (α/β). Similarly, when 201Y alone is used (Fig. 2B) to drive expression in γ neurons (and core α/β), we observe rescue of STM performance to levels that are roughly similar to that observed with the combination of 201Y (γ) and c739 (α/β) (compare Figs 2B with S1I). In contrast, we do not observe even partial rescue with NP3061 or c739 (α/β) (Fig. 2A, C), or with c305a or g0050 (α′/β′) (Fig. 2B,C). Even the combination of both NP3061 and c739 to yield higher levels of expression in α/β neurons is not sufficient to improve memory of the mutants (data not shown). Although NP1131 and 201Y are relatively specific to γ neurons, both of them have some expression outside of MB. To show that DopR expression in γ neurons, rather than other cells outside of MB, rescue the STM performance, we used MBGal80 [22] to suppress the expression of NP1131 within MB. Flies that carry NP1131, MBGal80 along with UAS-DopR-cDNA and dumb2 exhibit performance indices that are not significantly different from dumb2 mutant controls (Fig. 2D). Taken together, these findings support the surprising conclusion that DopR expression in γ neurons is necessary and sufficient to fully rescue the STM defect of dumb2 mutants. In contrast, neither α/β nor α′/β′ expression is necessary or sufficient to significantly restore STM.

Figure 2. DopR targeted expression in γ KCs alone is sufficient to fully support STM.

Memory was tested at 3 minutes after a single training session. Each γ Gal4 driver, NP1131 (A) or 201Y (B), was sufficient to rescue the STM defect of the dumb2 mutant in combination with the UAS-DopR-cDNA transgene (*p<0.05, n=6 for all groups). For NP1131, performance was not significantly different from that of WT (A, B, p>0.05, n=6) or the combination of NP1131 and NP3061 (A, p>0.05, n=6). In contrast, neither the α/β Gal4 drivers (NP3061 (A) and c739 (C)), nor the α′/β′ Gal4 drivers, (c305a (B) and g0050 (C)), exhibit significant rescue of STM performance (p>0.05, n=6 for all groups). D52H, a Gal4 driver that labels both γ and α/β neurons, provides rescue in combination with the UAS-DopR-cDNA transgene (C, *p<0.05, n=6). Although NP1131 was sufficient to rescue dumb2 STM (D, *p<0.05, n=6), adding MBGal80 to subtract the MB-Kenyon Cell expression from NP1131 completely suppress the NP1131 driven rescue (D, p>0.05, n=6).

DopR expression in γ KCs alone is sufficient to support all phases of memory

The above findings indicate that memory measured 3 minutes after one training session can be fully formed with DopR expression restricted to γ neurons. This is consistent with the observation that rut expression in γ neurons can rescue the STM defect of rut mutants, but unlike rut, DopR mutants fully disrupt memory performance. The full rescue with γ neuron expression thus suggests that all DA-mediated US signaling for STM is initially mediated by γ neurons. However, in the case of rut, which is thought to act downstream of DopR signaling, memory at later time-points after training requires additional expression in α/β lobes [23, 28]. We therefore examined the effects of cell-type specific DopR expression on memory measured 3-hours after 1 session of training as well as 24 hours after massed and spaced training.

dumb2 mutant animals do not form detectable 3hr memory (Fig. 3), as is also true for memory measured 24 hours after massed or spaced training (Figs 1, 3, and 4). As with STM measured right after one training session, memory measured 3-hours after one training session can be fully rescued with expression in γ neurons using NP1131 and the rescue observed with 201Y, which yields lower levels of γ neuron expression, is nearly as high (Figs. 3A, B). Here too, we observe no evidence of rescue when we drive expression with NP3061 (α/β) or c305a (α′/β′). Indeed the levels of rescue we observe with NP1131 are equivalent to that seen with the combination of NP1131 and NP3061 (Fig. 3A). We next conducted similar experiments to probe effects on ARM measured 24 hours after 10 cycles of massed training as well as LTM measured 24 hours after spaced training. Here too, expression with NP1131 (γ) was sufficient to fully rescue these consolidated forms of memory (Fig. 3C, D). In each case, the levels of performance obtained with only NP1131 were equivalent to that observed in wild type animals or animals that contained both the NP1131 and NP3061 Gal4 drivers (Fig. 3A, C, D). And as with 3-min and 3-hour memory, expression of DopR in α/β neurons with NP3061 is not sufficient to provide significant performance (Figs 3C, D). Taken together, the above findings support the striking conclusion that DopR expression in only γ KCs is sufficient to support all memory phases including STM, middle-term memory (MTM), ARM and LTM. We did not detect any rescue with DopR expression in either α/β alone or α′/β′ alone. Indeed the levels of rescue observed with γ neurons alone were as high as that observed with γ and α/β combined and as high as performance of wild type animals.

Figure 3. DopR expression in γ KCs alone is sufficient to support both intermediate and consolidated memory.

Memory was tested either 3 hours after a single training session (A, B) or 24 hours after 10 cycles of massed (C) or spaced (D) training. Compared with dumb2 mutant controls, dumb2 flies with the NP1131 γ Gal4 driver and UAS-DopR-cDNA exhibited full rescue of 3 hour memory (A, B, *p<0.05, n=6 for all groups), memory 24 hours after massed training (C, *p<0.05, n=14 for all groups) and memory 24 hours after spaced training (D, *p<0.05, n=7 for all groups). In each case, the performance observed was equivalent to that of WT (A–D, p>0.05).Similarly dumb2 flies with 201Y γ Gal4 driver and UAS-DopR-cDNA also exhibit significant rescue of 3 hour memory (B, *p<0.05, n=6 for all groups). In contrast, the NP3061 α/β and c305a α′/β′ Gal4 drivers do not show significant rescue compared to dumb2 controls (A–D, p>0.05, n=6.)

Figure 4. γ neuron DopR expression is sufficient to support aversive memory irrespective of odor choice.

Ethyl Butyrate (EB) was paired with Amyl Acetate (AA) as CS odors for 3 min-STM in (A). Benzaldehyde (BA) was paired with Octanol (OCT) as CS odors for 3 min-STM in (B) and 24 hr-LTM (10X spaced training) in (C). In all cases, dumb2 flies with the NP1131 γ Gal4 driver and UAS-DopR-cDNA exhibit rescue of performance compared with dumb2 controls (A–C, *p<0.05, n=8 for all groups). In contrast, c305a α′/β′ Gal4 driver does not yield significant rescue in all cases (A–C, p>0.05, n=8 for all groups). NP3061 α/β driven expression yields minimal levels of STM performance (A, B, *p<0.05, n=8 for all groups), but LTM performance is not rescued (C, p>0.05, n=8).

γ neuron DopR expression is a gateway for US input to form aversive memory irrespective of odor choice

The Pavlovian olfactory learning paradigm employed here is a discriminative assay in which the animals are trained to associate one odor, the CS+, with electric shock. A second unpaired CS- odor is used as a control. In each experiment, two groups of flies are reciprocally trained to the two odors and the PI is calculated as an average of the two ½-PIs. All experiments in Figures 1–3 utilized 3-Octanol (OCT) and 4-methylcyclohexanol (MCH), the two odors that are most commonly used for this assay. Because perception of pure chemical odors is thought to be represented as sparse responses in populations of KCs [29, 30], we wondered whether the sufficiency of γ lobes as a site of DopR input could derive from odor choice if OCT and MCH by chance triggered mostly γ lobes responses. To rule out this sort of explanation, we tested two additional odor combinations among three additional odors that are chemically dissimilar from OCT and MCH. For this series of experiments, we tested the combination of Benzaldehyde (BA) paired with OCT and Ethyl Butyrate (EB) paired with Amyl Acetate/Pentyl Acetate (AA). These odors have been successfully used as CSs to induce aversive memory in previous studies [7, 31, 32]. We first tested 3-min STM with EB paired with AA and BA paired with OCT. With these odor combinations, DopR expression in γ neurons with only NP1131 was sufficient to restore STM performance to levels nearly as high as that of wild type (Figs. 4A, B). In contrast, expression in α/β or α′/β′ using NP3061 or c305a produced performance levels that were modestly improved (NP3061) or not improved (c305a) relative to dumb2 mutants. Finally, we examined memory 24-hours after spaced training using the combination of BA paired with OCT. Here too NP1131 driven γ neuron expression produced normal levels of LTM performance whereas performance with NP3061 or c305a remained near zero (Fig. 4C).

Taken together, these findings demonstrate that DopR signaling for aversive reinforcement of olfactory memory is restricted to the γ neuron subset of MB neurons. This idea is convergent with the established role of MB-MP1 DA neurons, which send fibers into the heel of the MB (which consists largely of γ neurons), and the inner core of the peduncle (which is occupied by α/β neurons) [25, 33]. As an added validation of this model, we used GFP-reconstituted across the synapse (GRASP [34, 35]) to visualize connections between these two cell types (Fig. S2). When the two halves of GRASP are expressed in MB-MP1 and γ neurons, we observe strong labeling in the heel (Fig. S2A) of the MB, consistent with the hypothesis that MB-MP1 forms synapses with γ neurons.

DISCUSSION

Because DopR is thought to mediate the US information [3–7], identification of the spatial requirements of this receptor pinpoints the initial site of CS-US coincidence detection. To date, most genetic and circuit manipulations suggest that olfactory memory performance at a given retention interval can be dissected into distinct and independently disrupt-able mechanisms acting in parallel in distinct neuronal cell types [1, 36–42]. For example, the STM defects of rut and NF1 can be rescued with expression in γ for rut and α/β neurons for NF1. Experimental dissections of the circuits required for LTM have suggested a major role for α/β neurons [18, 23, 43, 44] as well as for ellipsoid body (eb) [42] and DAL neurons [45]. These kind of findings have been interpreted to support the idea of independent signaling for parallel memory traces [9, 10, 23] as well as sequential action in different cell types to support a single memory mechanism [22]. Our findings demonstrate that DopR expression in MB is sufficient to support both rut-dependent and independent forms of CS-US association leading to STM, as well as to consolidated ARM and LTM. This conclusion also generalizes to 3 different combinations among 5 different odors, providing strong evidence that the functional distinctions between KC classes are not artifacts caused by differences in the population of neurons involved in coding each odor percept. With each of these odor combinations and memory phases, there also was no case where expression in α/β or α′/β′ populations was sufficient or necessary to provide substantial rescue of dumb2 mutants.

Together, this set of findings pinpoints the DopR-mediated inputs for STM, MTM, ARM and LTM to the γ neuron population of MB KCs. This conclusion is consistent with findings from previous attempts to map the subset of DA neurons that convey the US to MB using either inhibition or activation of neural transmission to block or mimic the US signal [3, 4]. In these studies, the largest magnitude effects were seen with stimulation of MB-MP1, a neuron in the PPL1 cluster of DA neurons (although it should be noted that smaller magnitude effects also were seen for several other DA cell types [3]), which is sufficient to substitute for the US. Although inhibition of MB-MP1 neurons has not been demonstrated to block learning [3, 4, 33], these DA neurons likely participate in mediating at least a portion of the US stimulus for aversive conditioning. MB-MP1 neurons project to the base of the peduncle, occupied by the axons of α/β neurons and the heel of the MB, which is comprised largely of γ neurons [25, 33]. As an independent validation of the hypothesis that these MB-MP1 neurons provide direct input to γ neurons, we used the GRASP method [34, 35] to visualize putative synaptic connections in the heel between these two cell types (Fig. S2).

The fact that γ lobe expression of DopR is sufficient to restore not only STM but also both ARM and LTM is noteworthy. Previous attempts to map the neural circuits for olfactory memory have revealed roles for α/β lobes in particular for consolidated memory [18, 23, 28, 40, 43, 44] (but cf. [45]) . Because massed and spaced training experiments consist of repetitive training rather than the single training trial used for STM and MTM, differences in circuit requirements could in principle derive from training paradigm-dependent differences in the CS-US association circuit, as appears to be true for appetitive reinforcement [28]. But this appears not to be the case for DopR function in aversive reinforcement because we observe full rescue of these consolidated forms of memory with γ lobe expression of DopR.

How can this conclusion be reconciled with the requirement for downstream signaling molecules within α/β lobe neurons [10, 18, 23], as well as in downstream eb neurons [42] and DAL neurons [45]? We see three possible explanations that are not mutually exclusive. First, it is possible that US information is deconstructed into more than one pathway, mediated by different receptors. These could include additional DA receptors, or other neurotransmitter systems such as serotonin. It is worth note that DA inputs to MB also have been implicated in hunger/satiety modulation of appetitive memory retrieval [7, 33]and DopR signaling also has been implicated in several forms of arousal [46] that in principle could represent a component of the reinforcement signal that could be separate from a more specific perceptual representation of the shock experience. Our findings nevertheless lead to the conclusion that any additional US information depends critically on DopR mediated DA signaling in the γ lobe population of neurons. A second possibility worth considering stems from the finding that output from α/β lobe, eb, and DAL neurons are each required for retrieval depending on the retention interval measured [22, 43, 45, 47]. Thus we cannot formally rule out a model in which all of the functional impacts of various manipulations of α/β lobe derive from defects in retrieval. This would be difficult to fathom for cases such as NF1 rescue of STM and rut function for LTM, but in principle this interpretation is possible. The third possibility is that consolidation of the γ lobe CS-US association involves signaling within α/β lobe neurons [10, 18, 23], as well as in downstream eb neurons [42] and DAL neurons [45]. Such a model predicts communication between γ lobe and the rest of MB during training and/or afterwards (cf. [22, 47, 48]).

• DopR-mediated US-reinforcement for Drosophila aversive memory maps to Mushroom Bodies.

• DopR function in γ lobe is sufficient to support all temporal phases of memory.

• Coincidence detection occurs in γ neurons and is relayed to α/β during consolidation.