Research Article

Visualizing Long-Term Memory Formation in Two Neurons of the Drosophila Brain

See allHide authors and affiliations

Science  10 Feb 2012:
Vol. 335, Issue 6069, pp. 678-685
DOI: 10.1126/science.1212735

Abstract

Long-term memory (LTM) depends on the synthesis of new proteins. Using a temperature-sensitive ribosome-inactivating toxin to acutely inhibit protein synthesis, we screened individual neurons making new proteins after olfactory associative conditioning in Drosophila. Surprisingly, LTM was impaired after inhibiting protein synthesis in two dorsal-anterior-lateral (DAL) neurons but not in the mushroom body (MB), which is considered the adult learning and memory center. Using a photoconvertible fluorescent protein KAEDE to report de novo protein synthesis, we have directly visualized cyclic adenosine monophosphate (cAMP) response element–binding protein (CREB)–dependent transcriptional activation of calcium/calmodulin-dependent protein kinase II and period genes in the DAL neurons after spaced but not massed training. Memory retention was impaired by blocking neural output in DAL during retrieval but not during acquisition or consolidation. These findings suggest an extra-MB memory circuit in Drosophila: LTM consolidation (MB to DAL), storage (DAL), and retrieval (DAL to MB).

In Drosophila, LTM is produced by spaced repetitive training, which induces cyclic adenosine monophosphate (cAMP) response element–binding protein (CREB)–dependent gene transcription followed by de novo protein synthesis (1, 2). A prominent neuroanatomical site involved with memory formation is the mushroom body (MB), which consists of γ, α′β′, and αβ neurons. Calcium-imaging studies have shown that each cell type displays a distinct and altered activity at different time durations after training (3). Moreover, 30 different genetic or pharmacological disruptions have suggested that the MBs are involved in both short-term memory (STM) and long-term memory (LTM) (38). Nevertheless, de novo protein synthesis required for LTM consolidation has never been directly visualized and/or manipulated in any targeted brain structures, including the MBs.

Visualizing de novo protein synthesis in targeted neurons. KAEDE is a green fluorescent protein (GFP), which changes its structure irreversibly to a red fluorescent protein (RFP) upon ultraviolet (UV) irradiation (9, 10). We generated a transgenic upstream activation sequence (UAS)–kaede to monitor de novo transcriptional activities in targeted neurons. We showed that KAEDE faithfully reports the cyclic transcriptional activity of period (per) in circadian pacemaker cells (lateral neurons). Preexisting green KAEDE in the lateral neurons was photoconverted into red KAEDE in living per-Gal4>UAS-kaede flies exposed to UV irradiation (fig. S1, A and B). By photoconverting green KAEDE to red every 4 hours, we showed that KAEDE exhibits a diurnal cycle of de novo synthesis in the lateral neurons (Fig. 1A), parallel to the oscillation of per RNA (11). Newly synthesized green KAEDE was about 10 times as high during the night as during the day, indicated by accumulative measurement (Fig. 1A) and time lapse recording (movie S1). This de novo KAEDE synthesis in lateral neurons was reduced significantly in flies fed the protein synthesis inhibitor, cycloheximide (Fig. 1B). In contrast, red KAEDE remained at a constant level with or without cycloheximide feeding, confirming that photoconverted red KAEDE is irreversible and that spontaneous conversion of green KAEDE to red does not occur.

Fig. 1

Visualizing and blocking de novo protein synthesis in identified neurons. (A) Diurnal cycle of per transcriptional activity in the lateral neurons. To reset preexisting green KAEDE, living flies were UV irradiated every 4 hours (arrowheads) or every 12 hours, i.e., Zeitgeber Time (ZT) 0 or 12. Measurement of relative amount of de novo KAEDE synthesis was estimated within flies by normalizing to preexisting red KAEDE (%ΔF/Embedded Image0). Values are means ± SEM (N = 3 to 7 samples). (B) The effect of cycloheximide (+CXM) feeding. Images of lateral neurons were taken 5 hours after photoconversion at ZT12. Values are means ± SEM (N = 8 to 24 samples; *, P < 0.05). (C) The effect of activated RICINCS. Flies were photoconverted at ZT12 and kept (i) at 18°C (inactive RICINCS), (ii) at 30°C (activated RICINCS) during ZT12 to ZT17, or (iii) at 30°C during ZT7 to ZT12 and at 18°C during ZT12 to ZT17. Images of lateral neurons were taken at ZT17. Values are means ± SEM (N = 6 to 16 samples; *, P < 0.05; **, P < 0.01). (D and E) Effects of activated RICINCS in the MB neurons. After photoconversion, flies were kept at 18°C or 30°C for 24 hours before imaging. Values are means ± SEM (N = 10 to 15 samples; ***, P < 0.001). For all images, scale bar represents 10 μm. See supporting online material for more detailed legends of this and the other figures.

RICIN is a potent cytotoxic protein that inactivates eukaryotic ribosomes by hydrolytically cleaving the N-glycosidic bond (A4324) of the 28S ribosomal RNA subunit (12, 13). We obtained an effective transgenic fly carrying a cold-sensitive UAS-ricinCS transgene, by remobilization of a P-element insertion generated previously (14). In OK107-Gal4>UAS-ricinCS flies, high-temperature (30°C) RICINCS inactivated ribosomes, causing a severe MB deformation; at low temperature (18°C), however, RICINCS itself was inactive, thereby allowing normal MB development (fig. S1C).

We also visualized the effect of RICINCS on protein synthesis using KAEDE as a reporter. Using per-Gal4 as a driver, we found that KAEDE synthesis in lateral neurons was not inhibited by RICINCS at 18°C but decreased ~80% by RICINCS at 30°C for at least 5 hours (Fig. 1C). KAEDE synthesis inhibited by activated RICINCS at permissive temperature is quickly restored to normal level after a shift to the toxin’s restrictive temperature, suggesting that working ribosomes are resynthesized (Fig. 1C). Similar effects of RICINCS inhibition of protein synthesis were found in MB neurons (Fig. 1, D and E), using OK107-Gal4, which likely labels all MB neurons, and choline acetyltransferase (Cha) promoter–driving Gal4, which likely labels most acetylcholine-producing neurons.

Behavioral screen for neurons involved in protein synthesis–dependent memory formation. The spatiotemporal precision of UAS-ricinCS for acutely blocking protein synthesis in small subsets of targeted neurons allowed us to identify neurons undergoing protein synthesis during LTM formation. RICINCS activated (30°C) immediately after spaced training in Cha-Gal4 that contains ~60% of total brain neurons (15) impaired 1-day memory (Fig. 2A). The impairment was specific to LTM rather than anesthesia-resistant memory (ARM), because cycloheximide fed to these flies did not further reduce 1-day memory after spaced training and because activated RICINCS did not impair 1-day memory after massed training.

Fig. 2

Behavioral screen for neurons in which protein synthesis is required for LTM formation. (A) Protein synthesis is required in Cha-Gal4 neurons for normal LTM formation. After spaced or massed training, Cha-Gal4>UAS-ricinCS flies fed with (+) or without (–) cycloheximide (CXM) were shifted to 30°C or remained at 18°C. Values are means ± SEM (N = 12 experiments; **, P < 0.01; ***, P < 0.001; N.S., P > 0.05). (B) Behavioral screen for Gal4 expression patterns containing neurons in which protein synthesis is required for LTM formation. All flies were subjected to spaced training, and then RICINCS was activated by keeping flies at 30°C during the 24-hour retention interval. Black bars denote significant impairments of 1-day memory. Gal4 expression patterns containing MB neurons (+) or not (–) are indicated. Values are means ± SEM (N = 8 to 12 experiments; **, P < 0.01; ***, P < 0.001). (C) Protein synthesis is required for LTM formation in Cha-Gal4–expressing neurons outside of MB. MB-Gal80 inhibits Gal4 expression in the MB neurons. Values are means ± SEM (N = 12 experiments; ***, P < 0.001). See table S1 for a summary of Gal4 expression patterns.

In this form of associative learning, olfactory information (the conditioned stimulus) is detected by sensory neurons and then is relayed by projection neurons from the antennal lobe to the MB, where it is modulated by anterior paired lateral (APL) and dorsal paired medial (DPM) neurons. Through uncharacterized interneurons, information processed in the MBs eventually reaches the central complex, including the ellipsoid body. Foot-shock punishment (the unconditioned stimulus) is thought to reach MB through dopaminergic TH-Gal4 neurons (1618). Unexpectedly, 1-day memory retention remained intact when RICINCS was expressed in olfactory sensory neurons (Or83b-Gal4), olfactory projection neurons (GH146-Gal4), MB modulatory neurons (GH146-Gal4, c316-Gal4), all MB neurons (c247-Gal4, c772-Gal4, and OK107-Gal4), αβ neurons (c739-Gal4), α′β′ neurons (c305a-Gal4, E0973-Gal4, and G0050-Gal4), ellipsoid body neurons (c42-Gal4, c217-Gal4, c507-Gal4, Feb170-Gal4, and P0010-Gal4) and dopaminergic neurons (TH-Gal4) and again activated immediately after spaced training (Fig. 2B). Furthermore, by limiting RICINCS expression to neurons outside of MB using a combination of Cha-Gal4 and MB-Gal80 (Gal80 inhibits Gal4), 1-day memory again was impaired when RICINCS was activated immediately after spaced training but not after massed training (Fig. 2C and fig. S2A). Therefore, regardless of numerous studies suggesting LTM storage in the antennal lobes (19), the MBs (3, 20) and the ellipsoid body (21), our results suggest that de novo protein synthesis during LTM formation occurs in Cha-Gal4–expressing neurons outside of the MBs.

Next, we performed a more extensive behavioral screen for patterns of RICINCS expression that yielded 1-day memory impairments when RICINCS was activated immediately after spaced training (fig. S2B). LTM was impaired with Gal4 drivers, cer (crammer)-, Ddc (Dopa decarboxylase)-, Trh493 (Tryptophan hydroxylase)-, Trh996-, cry (cryptochrome)-, per-, CaMKII-(X), and CaMKII-Gal4(III), whereas LTM was normal with Gal4 drivers, DVGLUT (Vesicular glutamate transporter)-, Gad (Glutamic acid decarboxylase 1)-, tim (timeless)14-27-, tim14-82-, and repo (reversed polarity)–Gal4, the latter of which labels glial cells (Fig. 2B). LTM impairments by activated RICINCS in targeted Gal4 neurons were confirmed again by repeating the experiment using “Cantonized” Gal4 lines outcrossed to control flies to equilibrate genetic backgrounds (fig. S3).

Identification of individual neurons with protein synthesis during LTM formation. PER protein is necessary for LTM after courtship conditioning (22, 23). We found that PER protein was also necessary for LTM after olfactory conditioning. In per0 flies, 1-day memory retention was impaired after spaced, but not after massed, training (fig. S4A). In contrast, 1-day memory after spaced training was normal in other circadian mutants, including tim03, tim04, dClkJrk, and cyc0 (fig. S4B). By blocking neurotransmission with UAS-shits, a temperature-sensitive dynamin protein (24, 25), we found that neural activity from per neurons was required for retrieval of 1-day memory after spaced training but not 3 hours or immediately after a single training session (fig. S4C). Activation of RICINCS in per neurons impaired 1-day memory after spaced training but not after massed training (fig. S4D). Moreover, by activating RICINCS in per neurons at different time windows after spaced training, we found that protein synthesis was required only during the first 12 hours of LTM formation (fig. S4D). We also evaluated activated RICINCS in MB neurons (OK107-Gal4 and c247-Gal4) using a sliding 12-hour window before and after spaced training (fig. S5). One-day memory retention remained normal in every case.

We identified individual neurons by looking for overlap in expression patterns of the 9 Gal4 driver lines in which activated RICINCS impaired LTM formation. Ddc-Gal4 and per-Gal4 were chosen for the initial analysis because of their distinctly different expression patterns (fig. S2B). Two per-Gal4 neurons located at the dorsal-anterior-lateral (DAL) protocerebrum and most Ddc-Gal4 neurons were immunopositive for DDC antibodies (fig. S6A). The two DAL neurons are good candidates for their participation in protein synthesis–dependent LTM formation because they also express N-methyl-d-aspartate (NMDA) receptors (dNR), which are required for LTM formation (26). DDC-antibody immunostaining, in fact, revealed that the DAL neurons are included in the expression patterns of all nine Gal4 driver lines (Fig. 3A). We also used cry-Gal80 to “subtract” expression of RICINCS in the two DAL neurons from Cha-Gal4 and per-Gal4 expression patterns (Fig. 3, B and C, and fig. S6B). In both cases, activated RICINCS in the remaining neurons did not affect 1-day memory after spaced training (Fig. 3, D and E), suggesting that protein synthesis for LTM formation occurred in neurons within the intersection of the Cha-Gal4, per-Gal4 and cry-Gal80 expression patterns, including the two DAL neurons. Next, we identified three new Gal4 drivers (E0946, G0338, and G0431) with relatively limited patterns of expression, but each of which contained the DAL neurons (validated again by DDC-antibody immunostaining) (Fig. 3F). In all three cases, we found that RICINCS, when activated during the first 12 hours after training, disrupted 1-day memory after spaced training but not after massed training (Fig. 3G).

Fig. 3

DAL neurons are required for consolidation and retrieval of LTM. (A) The same DAL neurons contained in seven different Gal4, as indicated by DDC-antibody immunostaining (magenta). (B and C) DAL neurons are “subtracted” from the Gal4 expression pattern (green) by cry-Gal80. (D and E) LTM formation required protein synthesis in neurons within the intersected expression between Cha-Gal4 and cry-Gal80 or between per-Gal4 and cry-Gal80. Values are means ± SEM (N = 12 experiments; **, P < 0.01 ***, P < 0.001). (F) Three independent Gal4 lines with more restricted expression patterns containing DAL neurons (arrow). (Inset) The identity of DAL neurons was verified using DDC-antibody immunostaining (magenta). (G) Protein synthesis in DAL neurons is required for LTM formation. Values are means ± SEM (N = 8 to 12 experiments; **, P < 0.01; ***, P < 0.001). (H) Neurotransmission from DAL neurons is required during LTM retrieval. One-day memory after spaced training was impaired when neurotransmission from DAL neurons was blocked (by transferring flies to 30°C for 1 hour) during the test trial (retrieval) but not during acquisition or (i) 0 to 8 hours, (ii) 8 to 16 hours, or (iii) 16 to 24 hours of the 24-hour retention interval (consolidation). Control flies kept continuously at 18°C exhibited normal 1-day memory retention. Values are means ± SEM (N = 8 to 12 experiments; ***, P < 0.001). (I) Polarity analysis of the DAL neuron (magenta). (Left) Putative dendrites labeled by Dscam-GFP (green). (Right) Putative axons labeled by syt-GFP (green). DLP, dorsolateral protocerebrum; IDFP, inferior dorsofrontal protocerebrum; SDFP, superior dorsofrontal protocerebrum. (J) Structural connections between DAL and MB neurons visualized by GRASP labeling. (Left) L5275-LexA expressed specifically in the MB pioneer αβ neurons. (Right) GRASP signal (arrowhead) was visualized in the K5 region. DAL and pioneer αβ neurons were labeled by CD4-antibody immunostaining (magenta). General brain structures were counterstained using DLG-antibody immunostaining [magenta in (F); grayscale in (I) and (J)]. Scale bar, 20 μm.

Using the same three Gal4 drivers (E0946, G0338, and G0431), we then used UAS-shits acutely to block neurotransmission from DAL neurons. One-day memory after spaced training was normal when neurotransmission was blocked (i) from 30 min before training to the first 8 hours after training, (ii) 8 to 16 hours after training, or (iii) 16 to 24 hours after training (Fig. 3H). Instead, retrieval of LTM was disrupted when neurotransmission was blocked during the test trial 1 day after spaced training (Fig. 3H). Blocking neurotransmission from DAL with these three Gal4 drivers did not affect memory retrieval immediately or 3 hours after one training session (fig. S7).

Are there other neurons that undergo protein synthesis during LTM formation? We used cry-Gal80 to subtract expression of RICINCS in DAL neurons from the broader expression patterns of a panel of Gal4 driver lines. When cry-Gal80 was combined with cer-Gal4 and RICINCS was activated for 24 hours immediately after training, 1-day memory after spaced training was defective and 1-day memory after massed training was normal (fig. S8A), indicating that other neurons within the cer expression pattern also undergo de novo protein synthesis along with DAL neurons during LTM formation. In contrast, when cry-Gal80 was combined with Cha-Gal4, Ddc-Gal4, Trh493-Gal4, Trh996-Gal4, cry-Gal4, per-Gal4, CaMKII-Gal4(X) or CaMKII-Gal4(III) and RICINCS was activated for 24 hours immediately after training, 1-day memory after spaced training remained normal (Fig. 3, D and E, and fig. S8B).

DAL axons are structurally connected with MB calyx. Using Dscam-GFP as a dendritic marker, we showed that putative DAL dendrites distributed mainly in the superior dorsofrontal protocerebrum (SDFP) region (Fig. 3I, left). Using synaptotagmin-GFP as an axon marker, we showed that DAL axons distributed widely in three brain regions: SDFP, dorsolateral protocerebrum (DLP), and inferior dorsofrontal protocerebrum (IDFP) (Fig. 3I, right) (15). By a close examination of G0431-Gal4 expression pattern in the brain counterstained with DLG-antibody immunostaining, we noticed that the DAL axons and the MB calyx intersected at K5 region where dendrites belonging to pioneer αβ neurons are aggregated (27). Using GFP reconstitution across synaptic partners (GRASP) labeling (28), we verified that DAL neurons in G0431-Gal4 and the MB pioneer αβ neurons in L5275-LexA (Fig. 3J, left) were structurally interconnected at the K5 region (Fig. 3J, right).

Identification of newly synthesized proteins in DAL neurons during LTM formation. Immunostaining revealed preferential expression of DDC, PER, dNR1, dNR2, CaMKII, TEQ (TEQUILA), and CRY proteins and octopamine in DAL neurons (fig. S9A). Using RNA interference (RNAi) with UAS-perPAS-IR G2 (29) driven by four different Gal4 drivers (per-Gal4, Ddc-Gal4, G0338-Gal4, and G0431-Gal4), we found that constitutive disruption of PER protein expression impaired LTM formation after spaced training but not after massed training (Fig. 4A and fig. S9, B to D). Constitutive UAS-perPAS-IR G2 expression in ellipsoid body neurons (Feb170-Gal4) did not affect 1-day memory after spaced training (fig. S9E). Similarly, expressing with G0431-Gal4, RNAi constructs for dNR1/dNR2, CaMKII, TEQUILA, or DDC/TRH also impaired LTM formation after spaced training but not after massed training (Fig. 4, B to E). To eliminate any developmental contribution to the impairments observed above, we repeated the same set of experiments using a temperature-sensitive tub-Gal80ts protein that suppresses Gal4-induced expression at 18°C but not at 30°C (30). When G0431-Gal4–induced RNAi expression for per, dNR1/dNR2, CaMKII, tequila, and Ddc/Trh genes were suppressed throughout development (18°C) and then allowed only in adults (30°C), we again observed defects in 1-day memory after spaced but not after massed training. Moreover, in each case, further inhibition of protein synthesis from feeding flies cycloheximide did not produce stronger LTM impairments (Fig. 4, F to J).

Fig. 4

RNAi-mediated disruption of specific genes in DAL neurons impairs LTM formation. (A to E) Constitutive RNAi-mediated down-regulation of PER, dNR1/dNR2, CaMKII, TEQUILA, and DDC/TRH in DAL neurons impaired 1-day memory after spaced training but not after massed training. Values are means ± SEM (N = 8 to 12 experiments; **, P < 0.01; ***, P < 0.001; N.S., P > 0.05). (F to J) Induced down-regulation of PER, dNR1/dNR2, CaMKII, TEQUILA, and DDC/TRH in the DAL neurons in adult flies impaired 1-day memory after spaced training but not after massed training. Adult flies raised at 18°C were kept at 30°C for 3 days before training to remove tub-Gal80ts–mediated inhibition of Gal4 activity, thereby allowing RNAi-mediated disruption of the target gene(s). Some groups also were fed with 35 mM cycloheximide (+CXM) 1 day before training and again until test trial. Control flies carrying the same transgenes were kept continuously at 18°C before behavioral evaluations. Values are means ± SEM (N = 8 to 12 experiments; **, P < 0.01; ***, P < 0.001; N.S., P > 0.05).

We also found that down-regulation of CRY or overexpression of CER in DAL neurons impaired 1-day memory after spaced training after constitutive, but not after adult-specific, transgenic manipulations (fig. S9, F and G). Moreover, adult-specific TβhRNAi expression in DAL neurons did not affect 1-day memory after spaced training (fig. S9H).

CREB2 activity in DAL neurons, but not MB neurons, is required for LTM formation. We confirmed an earlier report that 1-day memory after spaced training was impaired, but learning was normal, after constitutive expression in MB neurons by the c739-Gal4 driver of UAS-dcreb2-b, which encodes a CREB repressor protein (4, 7). However, learning was impaired after constitutive expression of UAS-dcreb2-b by two additional MB driver lines, OK107-Gal4 and c247-Gal4 (fig. S10, A to C). These learning defects prompted us to examine whether constitutive expression of UAS-dcreb2-b in MB neurons might produce any developmental defects. For all three Gal4 driver lines, we discovered significant neuroanatomical damage in MB. In OK107-Gal4>UAS-dcreb2-b flies, the α′ lobe was completely missing (fig. S10D); in c247-Gal4>UAS-dcreb2-b or c739-Gal4>UAS-dcreb2-b flies, MBs had significantly fewer GFP axons compared with control flies (fig. S10, E and F), and β lobe axons occasionally crossed the midline (fig. S10F).

To eliminate these developmental defects, we used tub-Gal80ts to limit UAS-dcreb2-b expression by OK107-Gal4, c247-Gal4, or c739-Gal4 to adults (fig. S10, G to I). Under these conditions, memory retention immediately after one training session and 1-day memory after spaced training both were normal (fig. S10, J to L). MB morphology was severely damaged if these flies were kept at 30°C throughout development (fig. S10M). Note that c739-Gal4, but not OK107-Gal4 or c247-Gal4, expressed weakly also in DAL neurons (fig. S10N). In contrast, when UAS-dcreb2-b expression was limited to the adult stage in Cha-Gal4>UAS-dcreb2-b;tub-Gal80ts flies, 1-day memory after spaced training, but not immediate memory after one training session, was impaired (fig. S10, O and P).

Next, we asked whether CREB2 activity is required in DAL neurons during LTM formation. Adult-specific (or constitutive) overexpression of UAS-dcreb2-b or RNAi-mediated down-regulation of CREB2 expression in DAL neurons impaired 1-day memory after spaced training but not after massed training (Fig. 5, A to D), and feeding with cycloheximide did not exaggerate these impairments (Fig. 5, C and D).

Fig. 5

CREB2 activity in DAL neurons is required for LTM formation. (A and B) Constitutive expression of UAS-dcreb2-b or UAS-creb2RNAi in the G0431-Gal4 neurons impaired 1-day memory after spaced training but not after massed training. Values are means ± SEM (N = 8 to 12 experiments; **, P < 0.01; ***, P < 0.001; N.S., P > 0.05). (C and D) Induced expression of UAS-dcreb2-b or UAS-creb2RNAi in G0431-Gal4 neurons impaired 1-day memory after spaced training but not after massed training. Adult flies raised at 18°C were transferred to 30°C for 3 days before training to remove tub-Gal80ts inhibition of Gal4 activity. Some groups also were fed with cycloheximide (+CXM) before training. One-day memory after spaced training also was evaluated for control flies carrying the same transgenes but kept continuously at 18°C. Values are means ± SEM (N = 8 to 12 experiments; *, P < 0.05; **, P < 0.01; ***, P < 0.001; N.S., P > 0.05).

Visualizing transcriptional activity in identified neurons during LTM formation. Thus far, we have shown that LTM formation is impaired by acute (adult-specific) disruptions of eight different genes (i.e., per, dNR1, dNR2, CaMKII, Teq, Ddc, Trh, and dcreb2) in DAL neurons (Figs. 4 and 5). Because these disruptions existed before and after training, we wanted to determine whether spaced training itself induced expression of these genes.

CaMKII-Gal4 expresses in both MB and DAL neurons (fig. S2B). LTM formation requires normal expression of CaMKII at the time of training in CaMKII-Gal4 neurons (19) and in MB neurons (7). When activated immediately after spaced training, RICINCS in CaMKII-Gal4 neurons impairs 1-day memory (Fig. 2B) but, in contrast, RICINCS in MB neurons does not (fig. S4). Is the synthesis of CaMKII induced by spaced training? We used KAEDE to report the transcriptional activity of CaMKII for a 24-hour interval after spaced (or massed) training. After photoconversion of preexisting KAEDE (red), spaced training, but not massed training, specifically induced new KAEDE (green) in DAL neurons when UAS-kaede was driven by CaMKII-Gal4 (Fig. 6A). Quantification of newly synthesized green KAEDE indicated that CaMKII promoter activity in the DAL neurons is induced only by spaced training, because the low baseline level of green KAEDE after massed training remained constant throughout the 24-hour interval. In MB neurons, KAEDE synthesis driven by CaMKII-Gal4 remained at a constant low level at the tip of the α lobe or the soma of the MB after either spaced or massed training (Fig. 6A). In the DAL neurons, KAEDE synthesis driven by per-Gal4 likewise was elevated after spaced training but not after massed training (Fig. 6B). We did not see the spaced training–induced elevation of KAEDE synthesis in any other neurons contained in CaMKII-Gal4 and per-Gal4, whether imaging at the same detection sensitivity maximized for the DAL neurons or at a lower detection sensitivity maximized for the MB or the lateral circadian neurons (fig. S11). This elevation of CaMKII and per transcriptional activities occurred mainly during the first 8 hours and then gradually declined, as indicated by monitoring KAEDE synthesis in the DAL neurons in three time regimens after spaced training (fig. S12). By using tub-Gal80ts to limit Gal4-driven UAS-dcreb2-b expression at the adult stage, we found that spaced training–induced levels of KAEDE synthesis driven by CaMKII-Gal4 or per-Gal4 in DAL neurons were diminished (Fig. 6C).

Fig. 6

(A to F)Spaced training–induced transcriptional activities. Living flies were subjected to UV irradiation to convert green KAEDE into red KAEDE just before training. KAEDE levels were quantified 24 hours after spaced or massed training to either 3-octanol (OCT) or 4-methylcyclohexanol (MCH). For each brain, a single optical slice through the cell body of a DAL neuron, MB α lobe tip, or MB cell bodies was taken under the same imaging conditions. KAEDE synthesis was determined as a ratio of new (green, 488 nm) to preexisting (red, 543 nm) KAEDE (%△F/Embedded Image0). Adult specific expression of UAS-dcreb2-b was performed by removing tub-Gal80ts inhibition at 30°C for 3 days before the experiment. Values are means ± SEM (N = 8 to 24 samples; **, P < 0.01; ***, P < 0.001). Scale bar, 10 μm.

Three other genes, cry, Ddc and Trh, were not transcriptionally up-regulated after spaced or massed training, as reported by de novo KAEDE synthesis, even though their normal functions are required in DAL neurons for normal LTM formation (Fig. 6, D to F). We also evaluated an unknown gene, in which the G0431-Gal4 enhancer trap P element is inserted. Homozygous G0431-Gal4/G0431-Gal4 flies exhibited normal 1-day memory after spaced training (fig. S13). In DAL neurons of G0431-Gal4>UAS-kaede flies, constant low levels of KAEDE synthesis were seen throughout the 24-hour interval after spaced or massed training (fig. S13).

Discussion. We used genetically based methods to identify neurons in the Drosophila brain in which protein synthesis is required for LTM formation. The bilaterally paired DAL neurons satisfy several criteria to suggest that they are a site of LTM storage (6). First, de novo protein synthesis in DAL neurons during the first 12 hours after spaced training was required for normal LTM formation (Fig. 3G). Second, several proteins (i.e., dNR1, dNR2, PER, CaMKII, and TEQUILA) that have previously been shown to be necessary for LTM formation (19, 21, 23, 26, 31) were colocalized in DAL neurons (figs. S6A and S9A). Third, disruptions of PER, dNR1, dNR2, CaMKII, TEQUILA, DDC, TRH, and CREB2 in DAL neurons impaired 1-day memory after spaced (but not massed) training (Fig. 4 and Fig. 5, B and D). Fourth, expression of a repressor form of CREB2 protein in DAL neurons was sufficient to disrupt 1-day memory after spaced (but not massed) training (Fig. 5, A and C). Fifth, the transcriptional activities of CaMKII and per were elevated in DAL neurons after spaced (but not massed) training (Fig. 6, A and B). Sixth, the up-regulations of CaMKII and per in DAL neurons were CREB2-dependent (Fig. 6C). Seventh, neurotransmission from DAL neurons was required only for LTM retrieval but not for acquisition (LRN) or consolidation of LTM (Fig. 3H and fig. S7). Together, these data suggest that CaMKII and PER are bona fide “LTM proteins” synthesized in DAL neurons after spaced training.

RICINCS not only allowed us to target inhibition of protein synthesis to individual neurons but also allowed us to investigate the critical window for protein synthesis during LTM formation, because of its rapid temporal control (Fig. 1C). Unexpectedly, activated RICINCS did not affect LTM formation when expressed in MB neurons (Fig. 2B and fig. S5). Previous studies have shown that LTM formation involves the vertical axonal branches of αβ neurons in MBs (20), and LTM formation is blocked by overexpression of CREB repressor in MB [using c739-Gal4 (4)]. Moreover, increases in neural activity in these structures at the time of LTM retrieval appear CREB-dependent (4, 7). We clarified this apparent discrepancy by showing that (i) constitutive expression of CREB2b or RICINCS in the MB neurons (using c739-Gal4, OK107-Gal4, and c247-Gal4) resulted in developmental defects of MB structure, along with defects in LTM (figs. S1 and S10); (ii) adult stage–restricted expression of CREB2b or activated RICINCS in MB neurons yielded no detectable structural defects of MB and no LTM defects (figs. S5 and S10); (iii) adult stage–restricted expression of CREB2b or activated RICINCS in DAL neurons was sufficient to impair LTM formation (Figs. 3 and 5); and (iv) spaced training–induced elevation of CaMKII occurred in DAL neurons but not in MB neurons and was diminished by expression of CREB2b (Fig. 6).

Possibly, a more stringent requirement for inhibition of protein synthesis exists in MB neurons rather than in DAL neurons. A 50% reduction of total protein synthesis in fly brains from cycloheximide feeding is sufficient to block LTM formation (1). Here, we show that activated RICINCS in MB (or per-expressing) neurons results in an 80% reduction of KAEDE synthesis (Fig. 1, C to E). Also worth noting is that LTM defects in dFmr1 mutants can be ameliorated somewhat by feeding flies inhibitors of protein synthesis (32). Thus, the inhibition of negative regulators of genes involved in LTM formation (in MB neurons) theoretically could enhance, rather than impair, 1-day memory after spaced training—an outcome we did not monitor in this study.

A functional memory circuit must (i) register (acquire) an experience through a persistent neural activity, (ii) consolidate (store) a lasting memory through (protein synthesis–dependent) structural or functional changes somewhere in that circuit; and (iii) retrieve a long-term memory through reactivation of (some or all) of the circuit. Neural activity in MB neurons contributes to acquisition, consolidation, and retrieval of LTM (3, 8, 20, 21, 33, 34). Indeed, more than 30 different disruptions of LTM formation also diminish the calcium-based neural activity observed in αβ neurons in MB (3, 8). Importantly, expression patterns of many genes involved in LTM formation suggest that other neuroanatomical regions also participate in neural activity essential for LTM formation, including glial cells (35), antenna lobes (19), asymmetrical body (36), ellipsoid body (21), many other unidentified neurons in 17 different LTM mutants (37), and, of course, DAL neurons. The latter are an interesting case, because neurotransmission from DAL neurons appears to be required only for retrieval, but not for acquisition or consolidation, of LTM (Fig. 3 and fig. S7). None of these data are sufficient, however, to identify the neurons in which protein synthesis–dependent memory consolidation (storage) occurs.

We provide evidence of memory consolidation in identified neurons via the combination of direct observation of protein synthesis with disruption of LTM formation through targeted inhibition of protein synthesis by activated RICINCS. We also found a CREB2-dependent up-regulation of CaMKII and PER after spaced but not massed training (Fig. 6, A to C), only the former of which induces LTM formation. These observations support the hypothesis that LTM consolidation occurs, at least in part, through CREB-mediated modulation of gene expression in DAL neurons (2, 38). Further, our results indicate that CaMKII, PER, CREB2, DDC, TRH, dNR1, dNR2 and TEQUILA in DAL neurons are required at the time of training for normal LTM formation. In the case of DDC and TRH, however, spaced training did not up-regulate their transcription (Fig. 6, C to E), so they either are regulated posttranscriptionally or function as “basal” cellular machinery for the consolidation process.

Our data suggest a MB-DAL loop as part of the olfactory memory circuit. An olfactory experience first is communicated through olfactory sensory neurons and antennal lobe and registered in MB-APL-DPM as a neural activity (19, 3942). Neurotransmission from MB to DAL for consolidation (MB to DAL) occurs, and protein synthesis within DAL then yields structural and/or functional changes in DAL neural activity that communicate back to MB during retrieval (DAL to MB). Our observation that activated RICINCS in neurons of the cer-Gal4 expression pattern other than DAL still impairs LTM formation suggests that other extra-MB neurons also participate in the consolidation of LTM.

Supporting Online Material

www.sciencemag.org/cgi/content/full/335/6069/678/DC1

Materials and Methods

Figs. S1 to S13

References (4345)

Movie S1

References and Notes

  1. Acknowledgments: We thank S. Benzer, J. Blau, R. L. Davis, J. Hall, J. Hirsh, C. J. O’Kane, S. Kunes, L. Luo, T. Préat, M. Rosbash, S. Waddell, Y. Takamatsu, M. W. Young, and the Bloomington and Drosophila Genomics Resource Centers for fly stocks. We thank J. Hirsh, L. C. Griffith, and T. Préat for the antibodies. We also thank S. F. Teng for the preliminary test of UAS-kaede. This work was funded by grants from Dart Neurosciences LLC, National Science Council, and Ministry of Education to A.-S.C. C.-C.C., T.T., and A.-S.C. conceptualized the project and wrote the manuscript; C.-C.C. and A.-S.C. developed the methods and analyzed the data; C.-C.C., J.-K.W., T.-P.P., and C.-L.W carried out behavioral experiments; H.-W.L. and J.-K.W. performed imaging experiments; T.-F.F. generated the UAS-kaede transgenic flies; A.-S.C. supervised the project. The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to A.-S.C.
View Abstract

Navigate This Article