Main

The MBs form a prominent bilateral structure of the insect brain. They are essential for short-term memory (STM), and many of the proteins involved in the early phases of memory establishment are preferentially expressed in the MBs or in neurons projecting to them5. We showed that MBs are also involved in LTM6,7. Moreover, LTM mutants have been isolated recently8, but MB neuronal plasticity after LTM training remains to be characterized.

To acquire additional insights into LTM mechanisms, we have analysed Drosophila enhancer-trap Gal4 strains showing MB-specific expression. Flies were conditioned by exposure to an odour paired with electric shocks and subsequent exposure to a second odour in the absence of a shock. One of these strains, MZ1180 (refs 9, 10), which we have named crammerP (cerP), displayed a decrease in 24-h LTM after spaced training, but a normal 24-h memory capacity after massed training, a conditioning protocol that does not induce LTM2 (Fig. 1). STM induced after a single training cycle was also normal, showing that cerP is a specific LTM mutant. The LTM defect of cer is recessive (Supplementary Fig. 1). We performed excision experiments to generate revertant strains as well as additional mutant alleles. The excision line cerE29 displayed a specific LTM defect; in contrast, the cerE3 strain exhibited normal memory (Supplementary Fig. 1). Taken together, these results show that we have isolated a new LTM mutant with a P-element insertion.

Figure 1: cerP is a specific LTM mutant.
figure 1

Performance indices (PIs) were measured 1 and 3 h after a single conditioning cycle, and 24 h after spaced or massed training. Dark grey bars, CS; pale grey bars, cerP. Results are means and s.e.m.; n = 8–10 groups. Asterisks indicate significant differences in a t-test (P < 0.01).

Full size image

Polymerase chain reaction (PCR) rescue experiments revealed that the P-element in cerP flies was inserted 211 base pairs (bp) upstream of the gene encoding the succinyl-coenzyme A synthetase flavoprotein subunit (Scs-fp) and 53 bp downstream of the CG10460 gene (Fig. 2a). Quantitative reverse transcriptase PCR (RT–PCR) (Fig. 2b) and northern blot (data not shown) analyses were performed to determine which gene was affected by the P-element insertion. Expression of scs-fp was normal in cerP mutants, but expression of CG10460 was strongly reduced, indicating that this gene corresponds to cer (Fig. 2b). Similar results were obtained by western blot analysis with a polyclonal antibody raised against Cer (Fig. 2c). As expected, a decreased level of Cer was found in the mutant excision cerE29, and a normal level was found in the revertant excision cerE3 (Fig. 2c).

Figure 2: cer encodes a cathepsin inhibitor.
figure 2

a, The Gal4 P-element in cerP flies is inserted between the cer and scs-fp genes. The genomic fragment used for the rescue experiment is shown with a grey bar. X, XbaI; P, PstI; ORF, open reading frame; UTR, untranslated region. b, cer mRNA expression is affected by the P-element insertion, as measured by quantitative RT–PCR. Results are means and s.e.m.; n = 10–12 total RNA extractions. c, Similar results were obtained in a western blot assay. Densitometry analysis yielded Cer/Ciboulot (Cib) ratios of CS = 0.78 ± 0.24, cerP = 0.37 ± 0.11, cerE3 = 1.29 ± 0.27 and cerE29 = 0.32 ± 0.05 (n = 5). d, The mutant phenotype can be rescued by the insertion of one genomic copy of the wild-type cer+ gene. The two upper lines represent significant differences between each score and the scores of CS or cerP. PI, performance index. Results are means and s.e.m.; n = 6–10 groups. Asterisk indicates significant differences in a t-test (P < 0.05).

Full size image

Sequence comparisons showed that Cer is highly similar to the amino-terminal regions of cathepsins L, a subfamily within the papain-like cysteine proteinase family (Supplementary Fig. 2a). Cysteine proteinases, which are generally synthesized as inactive proenzymes with an inhibitory N-terminal propeptide region, range in size from 300 to 350 amino acids. Interestingly, Cer is only 79 amino acids long (observed molecular mass 9.5 kDa), and it lacks the cysteine proteinase catalytic region, suggesting that it might act as a trans-inhibitor of cysteine proteinase(s)11.

We performed experiments in vitro to assess the effect of Cer on the human liver cathepsins L and B and human papain. Cer was able to inhibit cathepsins L and B competitively (Ki = 25 nM and 30 nM, respectively; Supplementary Fig. 2b), but it could not inhibit papain even at 1 µM (data not shown). These Ki values are similar to that described for other cathepsin inhibitors such as CTLA-2β, a mammalian inhibitor of cathepsin L12. Although the target(s) of Cer in vivo remain(s) to be identified, a Drosophila protein–protein interactions map has recently been generated showing that Cer interacts strongly with three proteins13 that are all cathepsins. Two of them correspond to cathepsin B group and seem to be expressed in the adult brain (as seen by northern blot assay; D.C., L. Zinck and T.P., unpublished data). Those two cathepsins B constitutively lack the pro-inhibitor region. Cer is therefore probably needed to control their activity in trans.

To rescue the cerP LTM phenotype, we introduced a genomic construction carrying the cer+ gene with 1.5 kilobases (kb) of upstream sequence and 250 bp of downstream sequence (Fig. 2a). Rescue of the cerP phenotype was observed in the presence of a single copy of the cer+ transgene, indicating that this construct carries cer regulating sequences (Fig. 2d). In contrast, the presence of two copies of cer+ could not rescue the cerP LTM defect. Moreover, cer+ flies expressing two additional genomic copies showed a decrease in LTM (Fig. 2d) but their STM remained normal (data not shown). Thus, cer overexpression affects LTM as severely as does a constitutive decrease in the concentration of cer mRNA. Taken together, these results show that the concentration of Cer must be within a short range of that in the wild type to form a normal LTM.

In which brain cells is cer specifically expressed? When the cer Gal4 enhancer-trap line was crossed with the reporter strain P(UAS-mCD8–GFP), MBs were found to be strongly labelled together with large cells (Fig. 3a). Unfortunately, our anti-Cer antibody was not able to detect the peptide on tissues. Therefore, to identify directly the sites at which cer is expressed, we inserted the green fluorescent protein (GFP) open reading frame before the cer stop codon in the previously described genomic construct (Fig. 2a). Because a fragment carrying the same genomic sequences could rescue the LTM cer phenotype, we postulated that this fusion protein would be expressed in cells required for normal Cer function. Expression was observed in the MBs, and γ lobes were more strongly labelled than α/β lobes (Fig. 3b). MB neuron cell bodies were not strongly labelled. In the cortex and in the neuropile, large cells expressing Cer–GFP were found around the MB calyces and near MB lobes. These cells, which were also detected in cerP/P(UAS-mCD8–GFP) individuals, had no axonal projections, indicating that they might be glial cells. Indeed, they did not express the nuclear neuronal cell marker Elav (ref. 14) (data not shown). In contrast, some of these cells were found to express the nuclear glial cell marker Repo (Fig. 3c–e). Some cer-expressing cells seem Repo-negative and Elav-negative. Because Elav is expressed in all neurons14, whereas the anti-Repo antibody does not label all glial cells, the large Cer-expressing cells most probably correspond to glia. Cer is therefore expressed in both MB neurons and adjacent glial cells.

Figure 3: cer is expressed in MB neurons and in glial cells.
figure 3

a, Expression in a cerP – Gal4/P(UAS-mCD8–GFP) line. b, Brain labelling of the 191 transgenic line carrying the Cer–GFP fusion protein, as viewed by confocal microscopy. c, A 191 brain stained with an antibody against Repo. d, Cer–GFP labelling. e, Colocalization of Cer–GFP and Repo. Regions of overlap are indicated by arrow heads. Scale bar, 40 µm. f, Overexpression of cer with the cerP–Gal4 driver and with the glial driver Repo–Gal4 induces an abnormal 24-h LTM. No effect was observed for overexpression in two different MB Gal4 lines, 238Y and 5122. Ca, calyx; α, alpha vertical lobe; β, beta medial lobe; γ, gamma medial lobe. PI, performance index. Results are means and s.e.m.; n = 6–10 groups. Asterisk indicates significant differences in a t-test (P < 0.05).

Full size image

To further delimit the brain structures in which the concentration of Cer is critical for LTM formation, we overexpressed cer with the Gal4/UAS system. We observed a significant decrease in LTM conferred by the cer–Gal4 driver (Fig. 3f), confirming the dosage effect found with the genomic insert. Interestingly, a similar LTM defect was observed when cer was overexpressed with the repo–Gal4 glial-specific driver15. In contrast, no effect was seen when cer was overexpressed with the MB drivers 238Y (ref. 16) and 5122 (Fig. 3f) (the 5122 expression pattern is shown in Supplementary Fig. 3). These results indicate that the glial cells expressing cer might be involved in LTM formation. Indeed, neuron-like roles have recently been found for vertebrate glial cells. Some glial cells are able to integrate neuronal inputs, modulate synaptic activity, and process signals related to learning and memory17. Cer might be involved in the induction of neuronal plasticity by modulating crosstalk between neurons and glial cells.

Because the concentration of Cer seems to be a key factor in the establishment of LTM, we asked whether the cer transcript is regulated in the wild-type strain after LTM conditioning. We performed quantitative RT–PCR experiments at different time points after LTM training and found that the cer mRNA level decreased specifically 3 h after the end of training, in comparison with its level in untrained flies (Fig. 4). However, no significant change was observed 2 or 4 h after training (Fig. 4), indicating that the decrease in cer expression must occur in a narrow window of time to permit LTM formation. Similar results were obtained in western blot analysis: a 30% decrease in Cer level was observed between 3 and 4 h after conditioning (data not shown). Thus, the protein decay occurs slightly after the mRNA decay. Expression of scs-fp mRNA, the cer-adjacent gene, was not affected by LTM training (data not shown). Moreover, wild-type flies subjected to unpaired and repetitive odours and shocks, a regimen that does not induce learning, showed no variation in cer expression (Fig. 4). The sharp modulation of cer mRNA is therefore correlated with LTM formation.

Figure 4: The level of cer mRNA is regulated after LTM conditioning.
figure 4

The level of cer mRNA is downregulated 3 h after LTM conditioning in the CS wild-type strain. An unpaired training does not affect the level of cer mRNA. Dark grey bars, untrained; pale grey bars, trained. Results are means and s.e.m. Asterisks indicate significant differences in a t-test (P < 0.01).

Full size image

What is the physiological meaning of the decrease in cer expression 3 h after LTM training? In the mouse, two waves of hippocampal mRNA synthesis are required for LTM formation after contextual fear conditioning18. Similar waves of gene expression have been described in Hermissenda crassicornis after associative conditioning19. Moreover, the simultaneous inhibition of multiple caspases (a family of cysteine proteinases) in the hippocampus blocks long-term, but not short-term, spatial memory20. Our results extend these observations: we show that the expression of cer, which encodes an inhibitor of cysteine proteinases, is reduced for a short interval 3 h after training, thus probably leading to a transient activation of its cysteine proteinase(s) target(s). Significantly, prolonged cysteine proteinase activation is associated with neuronal degeneration in Alzheimer's disease: regions that contain maximal amounts of amyloid precursor protein also express maximal concentrations of cathepsin B and cathepsin L mRNA, indicating that the memory deficit of these patients might be linked to the deregulation of biochemical pathways involved in neuronal brain plasticity4,21. We speculate that the memory deficit observed in flies with constitutively decreased concentrations of Cer might parallel this situation.

Methods

Conditioning

The wild-type reference stock was Canton-Special (CS). The cerP and all other strains used for memory experiments were outcrossed to flies of the CS background. Flies were conditioned by exposure to an odour paired with electric shocks and subsequent exposure to a second odour in the absence of shock, as described previously6.

Excisions and PCR rescue

Excision experiments were performed as described previously22. Genomic DNA adjacent to the P-element insertion was isolated by inverse PCR as described in http://www.fruitfly.org/about/methods/inverse.pcr.html.

Quantitative RT–PCR

Total RNA from the heads of conditioned or naive flies (1–2 µg) was used as a template for reverse transcription using an oligo(dT) and SuperScript II reverse transcriptase (Invitrogen). Quantitative PCR was performed in accordance with the protocols provided with the LightCycler thermocycler (Roche) and the LightCycler – FastStart DNA Master SYBR Green I kit. The primers used were designed to amplify 60–80-bp fragments at the exon–intron boundary. Results were analysed in accordance with the instructions of the LightCycler software (version 3.5, Roche Molecular Biochemicals). The levels of cer and scs-fp mRNA were also normalized to the level of α-Tub84B mRNA.

Antibody production and western blot analysis

The Cer–glutathione S-transferase (Cer–GST) fusion protein was produced as described previously23. Rabbits were injected dorsally several times every month for 3 months with 250 µg of the Cer–GST fusion protein mixed with Freund's adjuvant.

SDS–PAGE (with 70 µg of total adult head protein extract) and western blot analysis were performed as described previously24. The membrane was incubated with polyclonal antiserum against the Cer–GST fusion protein (dilution 1:4,000) or with antiserum against Ciboulot (dilution 1:8,000), as described previously23. The intensity of Cer-specific signals was normalized to that of Ciboulot-specific signals.

Cer inhibition assay

Cer activity was assayed as described for Bombyx cysteine proteinase inhibitor (BCPI) or CTLA-2α (refs 12, 25), with a Safas flx spectrofluorimeter, Cer (0–200 nM), benzyloxycarbonyl-Phe-Arg-7amino4methylcoumarin as a substrate (3, 4.5 and 6 µM; Calbiochem, La Jolla, California), human liver cathepsin B and L (25 and 10 nM, respectively; Calbiochem) and papain (100 nM; Sigma, St Louis, Missouri).

Constructions

To create the genomic cer+ construct the clone BACR21D20 was digested with PstI and XbaI. A 2,296-bp fragment (carrying the cer+ gene with 1.5 kb of sequence upstream of the transcription initiation site and 250 bp of downstream sequence) was cloned into pCaSpeR-4. Two transgenic lines, GENX and GEND, were obtained, with the P-element inserted in the second and third chromosome, respectively. Western blots were quantified and we observed that homozygous GENX and GEND had respectively twofold and threefold more Cer than CS (data not shown).

The Cer–GFP fusion protein construct was generated with the pCaSpeR-4 construction described above, by inserting GFP in phase before the cer stop codon. The cer 5′ region until before the stop codon (BstEII/BamHI), the GFP open reading frame (BamHI/XbaI) and the rest of cer (the 3′ region from the stop codon and 250 bp of its downstream sequence (XbaI/PstI) were amplified by PCR by using specific primers with the indicated restriction sites. All three PCR products were digested with appropriate restriction enzymes and the three products were cloned into BstEII/PstI-digested pCaSpeR-4 in a four-way ligation reaction.

To create the UAS-cer+ construct a full-length cer+ cDNA was isolated from the BDGP clone LP06209 and cloned into BglII/XbaI-digested pCaSpeR-UAS.

Immunohistochemical analysis

Drosophila brains were dissected as described previously6, then stained with the anti-Repo monoclonal antibody (dilution 1:100) or with the anti-Elav monoclonal antibody (dilution 1:10,000). Goat anti-mouse secondary antibody conjugated with Alexa-Fluor 568 (Interchim) was used at a dilution of 1:1,000. Expression was examined with a Leica (Wetzlar, Germany) TCS SP2 laser scanning confocal microscope.