Report

Exclusive Consolidated Memory Phases in Drosophila

See allHide authors and affiliations

Science  14 May 2004:
Vol. 304, Issue 5673, pp. 1024-1027
DOI: 10.1126/science.1094932

Abstract

Two types of consolidated memory have been described in Drosophila, anesthesia-resistant memory (ARM), a shorter-lived form, and stabilized long-term memory (LTM). Until now, it has been thought that ARM and LTM coexist. On the contrary, we show that LTM formation leads to the extinction of ARM. Flies devoid of mushroom body vertical lobes cannot form LTM, but spaced conditioning can still erase their ARM, resulting in a remarkable situation: The more these flies are trained, the less they remember. We propose that ARM acts as a gating mechanism that ensures that LTM is formed only after repetitive and spaced training.

Memory is a complex and dynamic process, and the relations between the different memory phases continue to intrigue neuroscientists. Studies of cerebral pathologies or brain lesions show that one form of human memory can be impaired while others remain normal (1). In this context, the formation of long-lasting memory is of particular interest, because it is thought to involve sequential events sustained by metabolic pathways preserved throughout evolution (25).

In Drosophila, a single associative-learning trial (the short protocol) consisting of an odor accompanied by 12 pulses of electric shocks induces three temporally distinct phases of olfactory memory (3): short-term memory (STM) and middle-term memory (MTM), which are labile and rely on the adenosine 3′,5′-monophosphate (cAMP) pathway (6), and ARM, which is a form of consolidated memory. STM is impaired in the dunce (dnc) and rutabaga (rut) mutants. However, these mutants retain a significant level of early memory (7). MTM is affected in the amnesiac (amn) mutant, and ARM is affected in the radish (rsh) mutant (8, 9). STM, MTM, and ARM are also observed after intensive conditioning in which stimuli are presented repeatedly without intervening rest periods (the massed protocol) (10). Another consolidated memory, LTM, appears after multiple spaced training sessions (the long protocol) and is protein synthesis–dependent (10). The current Drosophila model proposes that the short protocol and the massed protocol induce a sequential pathway that begins with learning, passes through STM and MTM, and terminates in ARM, whereas the long protocol generates an additional phase, LTM. ARM and LTM are thought to derive from MTM and to coexist 24 hours after spaced conditioning (3, 10). However, amn mutants present near-normal ARM but defective MTM (8, 9). Thus, the notion that ARM is derived from MTM is questionable. Many issues concerning Drosophila memory remain to be solved. Why are there two forms of consolidated memory? Are they spatially disconnected or do they rely on the same brain structures? And why does LTM form only after spaced conditioning and not after intensive massed conditioning?

The mushroom bodies (MBs) form a bilaterally symmetric structure in the central brain and are composed of different classes of intrinsic neurons that send their axons into vertical and median lobes. To identify the onset of the LTM phase, we studied a subpopulation of alpha-lobe-absent flies (the ala mutant) that lack MB α/α′ vertical lobes. These flies learn normally but show no olfactory LTM 24 hours after spaced conditioning (11). ala memory was measured at several early time points after conditioning with the short or the long protocols (12). Thirty-minute memory performances were similar after both protocols (Fig. 1). However, spaced repetitions of the conditioning regime significantly decreased memory performance at 5 hours, in comparison with what was observed after the short protocol (Fig. 1). Thus, the more intensively flies lacking vertical lobes are trained, the less they seem to remember.

Fig. 1.

Consolidated memory phases are mutually exclusive. The ala mutant was trained with one cycle (1×) (dashed line) or five spaced cycles (5×) (continuous line) and tested at various times after training. ala flies were processed as described (11). Data from flies without vertical lobes are presented here. PI, performance index. At 30 min, the 5× PI is not significantly different from the 1× PI (χ2 P = 0.52). At 5 hours, the 5× PI is significantly lower than the 1× PI (χ2 ***P < 0.0003). The numbers of flies lacking vertical lobes among the total ala flies were for 1×: 30 min, 140/1021; 5 hours: 134/1325; for 5×: 30 min, 229/1591; 3 hours, 158/1054; 5 hours, 106/902. After the long protocol, control ala flies with all MB lobes present did not display decreased memory at 5 hours (1×: 31.4, n = 257; 5×: 43.5, n = 183).

The main form of memory that persists 5 hours after conditioning with the short protocol is ARM (3, 10). Why do flies without vertical lobes show almost no memory 5 hours after spaced conditioning? First, these flies do not display LTM because they lack the MB neuronal projections required to form LTM (11). Second, our results suggest that ARM is erased (or blocked) during a LTM-specific training protocol. Thus, in contrast to the assumptions of previous models (3), we find that ARM and LTM do not coexist. ARM is formed in ala flies after massed conditioning (11), which indicates that its absence is observed only after spaced conditioning. Those results suggest that, in wild-type flies, the LTM phase is promptly initiated after spaced conditioning and that LTM replaces ARM.

How do memory phases relate to brain structures? MBs are implicated in the elaboration and retrieval of early olfactory memory phases (1318), and MB vertical lobes are necessary for olfactory LTM (11). But it has not been directly shown that MB outputs are required for the retrieval of LTM, and ARM has not been formally linked to the MBs. In order to clarify several aspects of the MB/memory phase relationship, we studied flies expressing a thermosensitive version of the protein Shibire (19). Blocking synaptic transmission and possibly other endocytic processes (20) in subsets of MB cells projecting into all lobes, or only into the α/β lobes, led to a decrease in early memory (Fig. 2A) (1618). We used a Gal4 enhancer-trap line (21) to drive UAS-shits transgene expression in the γ lobes (fig. S1). With this Gal4 driver, we also observed a significant 2-hour decrease in memory, although odor perception and shock sensitivity were normal (table S2). To measure ARM, flies were trained with the short protocol and subjected to a cold shock 1 hour after conditioning to eliminate nonconsolidated memories (9). When all MB lobes were blocked during the experiment, ARM was erased (Fig. 2B). ARM was similarly decreased by blockage of the α/β lobes alone, but not significantly decreased by blockage of the γ lobes (Fig. 2B). Thus, ARM is supported by MBs and appears to rely more heavily on α/β neurons. Because ARM was found to be normal in the absence of either α or β lobes (11), we conclude that it can rely on either of those lobes.

Fig. 2.

Consolidated memories are localized in the MBs. Three Gal4 enhancer-trap lines were used to localize memories to the MBs: the MB247 line (247), which shows expression in a subset of cells in all MBs lobes (15, 18), the Gal1471 line (1471), which shows expression in some γ neurons (supporting online material), and the c739 line (739), which expresses Gal4 in some α/β neurons (28). The performances of Gal4/UAS-shits (Gal4/shi) individuals were compared with those of appropriate genetic controls. The vertical black and white arrows represent time points for conditioning and testing, respectively. The values correspond to the means ± SEMs (n = 6 to 30). The P value shown corresponds to the t test comparison with appropriate genetic controls (*P < 0.05; **P < 0.01; ***P < 0.001; NS indicates no significant difference). (A) Two-hour memory relies on various MB neurons (n = 8 to 14). (B) ARM relies more heavily on α/β neurons. Flies in which neurons from all lobes (247/shi) or only α/β neurons (739/shi) were inactivated reveal a memory decrease in comparison to control flies. Inactivation of γ neurons (1471) does not affect ARM (n = 9 to 26). (C) LTM retrieval requires output from α/β neurons. Flies were trained at the permissive temperature and tested at the restrictive temperature 24 hours after five spaced cycles. Inhibition of neurons from all lobes (247/shi) or of only α/β neurons (739/shi) leads to a memory decrease. Inactivation of γ neurons (1471) does not affect LTM. 1471/+ versus 1471/shi (P = 0.94); shi /+ versus 1471/shi (P = 0.98) (n = 8 to 30).

To assess whether vertical lobes are required for LTM retrieval, MB output synapses were impaired during the test, 24 hours after spaced conditioning. When all lobes were inactivated, a strong decrease in performance was observed (Fig. 2C), indicating that MBs are indeed required for LTM retrieval. Inactivation of the α/β lobes generated a similar drop in performance, whereas γ lobe inactivation did not affect LTM (Fig. 2C). Thus, α/β but not γ neurons are required for LTM retrieval. Because flies without β lobes show normal LTM (11), we conclude that the α lobes are the main center for the LTM retrieval. Altogether, those results suggest that ARM and LTM involve the same group of cells.

What is the link between ARM and the earlier phases of memory? We propose that ARM is not mainly derived from MTM because the amn mutant, which is defective for MTM, has normal ARM (8, 9). The amn gene encodes a neuropeptide that might stimulate the cAMP pathway (22, 23) via Rut-adenylyl cyclase activation. ARM therefore might be at least partially independent of cAMP regulation. To strengthen this hypothesis, ARM was measured in the rut mutant and found to be normal (Fig. 3), which confirms a previous study of other rut alleles (24). Thus the ARM pathway is at least partially independent of the STM/MTM pathway.

Fig. 3.

The rutabaga mutant displays normal ARM. Wild-type CS (+) and rutabaga2080 (rut2080) flies were subjected to 2 min of cold-shock anesthesia 1 hour after single-trial training. The 2-hour memory scores are not statistically different (n = 8) (t test, P = 0.85).

In mammals, competing memory systems have been described that involve different anatomical structures (25). We now reveal a competition between two types of consolidated memory, ARM and LTM, within the same structure, the MBs. Several observations suggest that cAMP-independent learning and memory occur, as originally suggested (24). First, significant levels of ARM are found in the amn and rut mutants, although their STM and MTM are strongly affected. Second, mutants with defects in the catalytic subunit (DCO) of PKA—the kinase normally activated by cAMP—display weak but stable memory for several hours after a single trial conditioning (26), which could be consolidated ARM. Third, null mutations affecting enzymes of the cAMP pathway do not abolish immediate memory (7). This cAMP-independent learning mode must rely on the MBs, because their ablation abolishes immediate olfactory memory (13). We propose that cAMP-independent learning could later give rise to rsh-dependent ARM (Fig. 4B).

Fig. 4.

Model of competitive memory phases. (A) A former model that holds that ARM and LTM coexist after spaced conditioning (3, 10). (B) A new model that postulates mutually exclusive consolidated phases. LTM and ARM are supported by parallel pathways: the former is cAMP-dependent, and the latter is rsh-dependent. After massed conditioning ARM is hypothesized to prevent the formation of LTM. After spaced conditioning, LTM replaces ARM. The dashed line represents a hypothetical pathway. The possibility that unidentified molecular interactions connect the two learning pathways cannot be excluded. (C) Anatomical representation of memory phases. STM and MTM are supported by the α/β and γ lobes, and ARM is supported by the α/β lobes. Three hours after spaced conditioning, ARM is erased and LTM forms in the α lobes.

LTM conditioning leads to the disappearance of ARM, and our data suggest that ARM and LTM involve the same group of neurons. The distinction between the two consolidated memories could thus rely on antagonistic molecular mechanisms within the same cells. Alternatively, different subsets of competing MB neurons might support ARM and LTM. Our results could explain the observation that a truncated and persistently active isoform of an atypical protein kinase C (PKMζ) enhances 4-day memory after massed, but not after spaced, training (27), because ARM is absent after spaced training.

Why are there two forms of consolidated memory, and why are they mutually exclusive? We propose that ARM acts as a gating mechanism for LTM formation (Fig. 4B). After massed training, ARM is formed in α/β neurons (Fig. 4C) and prevents LTM formation in the α lobes. After spaced training, ARM is erased, which releases the constraint on the LTM pathway. Such a mechanism would ensure that only information that has been encountered on independent occasions, and which, therefore, has a high predictive value, is stored in LTM. In contrast, a stimulus encountered only once for a short time (single trial) or for a longer time (massed training) would generate a semistabilized memory (ARM) that does not involve a heavy cascade of gene expression.

Supporting Online Material

www.sciencemag.org/cgi/content/full/304/5673/1024/DC1

Fig. S1

Table S1

References and Notes

References and Notes

View Abstract

Navigate This Article