Tissue-Specific Expression of a Type I Adenylyl Cyclase Rescues the rutabaga Mutant Memory Defect: In Search of the Engram

Materials and Methods

EXPERIMENTAL SETUP

The chamber (Fig. 1A) is a box (length, 40 mm; height, 2 mm; breadth, 5 mm) with floor and ceiling consisting of peltier elements that control the temperature in the chamber (built in the Genetics workshops, University of Wuerzburg). Clear plastic sides allow the transmission and detection of infrared light from an LED 100 mm away, and a light gate array (bar code reader). Thus, when a fly walks along the length of the chamber, it casts a shadow on the light gate array and this signal is sent to a computer. The fly's position is monitored at a frequency of 10 Hz and when a fly is in the predetermined “punished” half of the chamber, the entire chamber is heated to a maximum of 37°C under control of a thermistor in the chamber (custom software by R. Wolf). When the fly moves to the “unpunished” half, the chamber returns to 24°C. To test learning and memory, flies are subjected to a series of training/test phases. During a 30-sec pretest the fly can explore the chamber while the temperature stays at 24°C. The following training lasts 4 min, binned in 1-min blocks for analysis. Memory of the dangerous side is tested in the next 3 min (also binned) in which the heat is permanently turned off. This corresponds to minutes 5 to 7 in Figure 2. A single training session is a half-experiment, and the side the flies are trained to avoid is alternated between them. Therefore, half of the flies are trained to avoid one side of the chamber and half to avoid the other. This removes spontaneous side preferences in the chamber. A performance index (PI) is determined for different training/test phases as a measure of heat avoidance (during training) and memory (after training). It is calculated as the time spent in the unpunished half of the chamber minus the time spent in the punished half of the chamber, divided by the total time. The PI can have values ranging from −1 to 1, with flies that are perfect heat avoiders having a value of 1. A value of zero indicates no side preference. Pretest side preference (30 sec) was measured for individual flies. The mean pretest side preference was usually <0.05 and this was subtracted from the PIs in Figure 2 and Table 1.

Thermosensitivity is recorded adjusting the temperature in the front versus back half of the chamber independently and also independent of the flies' actions. It is stepped from 24°C on both sides to 24°C/30°C, 24°C/37°C, and 24°C/45°C for 1-min intervals, sequentially alternating the side with the higher temperature. A single run from low to high temperature difference is again a half-experiment and the side with higher temperature is alternated between half experiments. A fly's time spent on a given side is measured and heat avoidance indices are calculated as above (PI).

GENETIC MANIPULATIONS

The rut²⁰⁸⁰; ry double mutant (in a Canton-S genetic background) and the white-eyed Canton-S line wCS10were used as mutant and wild-type strains in all experiments. Two independent lines were generated by standard techniques in which the wild-type rut ⁺ cDNA under control of the UAS_GAL4 promoter was inserted onto the second (#2) and third (#1) chromosomes of the w CS10 strain. TheUAS_GAL4cDNAs were crossed into therut²⁰⁸⁰ background using a cantonized FM7a; SM1 or, respectively, FM7a; TM2 line (which were themselves generated by chromosome replacement using the w CS10 line). It should be noted that only male flies were used in the behavioral tests and were F₁ progeny of a genetic cross (e.g., arut²⁰⁸⁰; UAS_GAL4–rut ⁺female crossed to a GAL4 enhancer trap line male). Several of the GAL4 enhancer trap lines used to rescue the rutabaga phenotype were crossed for more than six generations to the w CS10 line to effectively replace the genetic background to w CS10. Other GAL4 lines were not subjected to outcrossing but did not affect learning negatively in control experiments (e.g., c522).

The wild-type rutabaga cDNA was subcloned from the Bluescript vector (Levin et al. 1992) with KpnI and XbaI into the pUAST vector (Brand and Perrimon 1993). This step removes 294 nucleotides of the cloned 3′ untranslated sequence but leaves the putative stop codon unchanged.

STATISTICS

A repeated measures ANOVA (Sachs 1992) in the Statistica 4.5 software program was used to determine any differences in performance indices between genotypes with the two rut ⁺ cDNA transgene rescue lines, wild type, and rutabaga mutants during both training and test phases (Fig. 2). To test the 3-min memory scores and heat avoidance, an ANOVA was performed on single genotype pairs, potentially rescued line versus rutabaga mutant with either rescue cDNA but no GAL4 driver or the parentalrut²⁰⁸⁰ (Table 1), and wild-type Canton-S versusrut²⁰⁸⁰ (heat avoidance; Fig. 3). GAL4 lines with PIs for 3-min memory significantly different from therutabaga mutant were determined using theP < 0.05 criterion. Differences between w CS10wild-type and rutabaga mutant lines with either of the tworut ⁺ cDNAs were examined with a repeated measures ANOVA and no significant differences were found (n = 127 for w cs10 with either rut ⁺ cDNA and n = 382 forrut²⁰⁸⁰ with either rut ⁺ cDNA,F = 0.611, P = 0.61). Therefore, data were pooled for these lines.

IMMUNOHISTOCHEMISTRY

GAL4 enhancer trap lines of interest were examined for expression patterns using the UAS_GAL4–Tau reporter (Ito et al. 1997). Approximately 20 flies were fixed (Carnoy) for mass histology as described (Heisenberg and Boehl 1979). Seven-micrometer sections were blocked for 3 hr with normal horse serum (1:50) in PBS (pH 7.4) plus 0.1% Triton X-100 (PBT) and challenged with anti-Tau antiserum (1:1000) in PBT overnight at 4°C. A series of washes and challenge with biotinylated horse anti-mouse serum (1:100) in PBT for 1 hr at room temperature followed. Signal was detected following Vectastain ABC elite kit instruction with a modified substrate preparation (0.67 mg/ml 3,3′diaminobenzidine dihydrochloride and 0.03% H₂O₂ in PBT).

Results and Discussion

The spatial learning paradigm, in which the position of a fly can be continuously monitored, is shown in Figure 1A. Individual flies' responses to the training protocol were examined and reveal several qualitative changes in behavior. The spontaneous behavior of flies in this chamber is to walk from one end to the other (Fig. 1B,C). This behavior is evident in the pretest period (the first 30 sec of the position traces in Fig. 1B,C) and in control experiments in which flies are allowed to run without punishment in the chamber for several minutes (not shown). There are no obvious differences betweenrutabaga mutant (Fig. 1B), wild-type (not shown), and rescued (see below, Fig. 1C) flies in this behavior. Walking speed first increases during training, perhaps as a reaction to the heat, and later slows to below spontaneous levels (period marked with “tr” and a black bar). Also evident during training is a marked preference for the chamber side not associated with punishment. There are differences between rutabaga mutant and wild-type flies during this phase. The rutabaga mutant flies have a longer high speed walking phase compared with wild type and are slower to avoid the side associated with punishment. The last phase in this experiment is a test without heat punishment (the period labeled with “test” and a hollow bar). Memory of training is evident in the continued avoidance of the chamber side associated with punishment, even after a fly re-explores that side. This avoidance eventually disappears and flies return to the pretraining behavior of walking from chamber end to end. The difference between rutabaga mutant and wild-type flies in this phase is the length of time before the pretraining behavior returns.

The performance deficits observed in rutabaga mutant flies are rescued with neuronal expression of a wild-type rutabaga cDNA. Both rutabaga mutant alleles (rut¹ andrut²⁰⁸⁰) tested in the spatial learning paradigm are defective (Wustmann et al. 1996; Fig. 2). Indeed, both the training and memory test avoidance performances are lower in rutabagamutant flies. This, together with the thermosensitivity assay results (see below) imply that these flies are defective in the learning process itself. Two independent insertions of anUAS_GAL4–rut ⁺ cDNA were tested for their ability to rescue this defect when combined with the elavneuronal GAL4 driver (Luo et al. 1994; Yao and White 1994; Ito et al. 1998). The UAS_GAL4–rut ⁺ cDNAs are labeledrut ⁺ cDNA #1 (B) and rut ⁺ cDNA #2 (C) throughout the text. The training period in Figure 2, A–D, is the first 4 min and is followed by 3 min of testing. Figure 2A shows that the white-eyed wild-type strain w CS10, with eitherrut ⁺ cDNA insert #1 or #2 (dark gray bars), have significantly higher avoidance scores during training and test phases than the rutabaga mutants with either rut ⁺ cDNA but no GAL4 driver (light gray bars, see figure legend for detailed genotypes). There were no significant differences between the w CS10 wild-type flies with rut ⁺ cDNA #1 or withrut ⁺ cDNA #2. Therefore, the wild-type data were pooled in Figure 2A for easier comparison to the rutabaga mutant. Figure2B also shows significantly higher avoidance scores during training and testing for flies carrying the rut ⁺ cDNA #1 and the GAL4–elav driver (dark gray bars) than for therutabaga mutant with the rut ⁺ cDNA #1 but no GAL4 driver (light gray bars). The second rut ⁺ cDNA in therutabaga mutant background with the GAL4–elav driver (Fig. 2C, dark gray bars) also shows significantly higher training/test phase avoidance scores compared to therutabaga mutant control (light gray bars). Finally, Figure 2D shows that in the rutabaga mutant with either rut ⁺cDNA #1 or #2, but no GAL4 driver (dark gray bars, pooled data), learning/memory scores are not significantly different from the parental rutabaga mutant strain (light gray bars). In addition, no significant difference was observed between the w CS10 wild-type UAS_GAL4–rut ⁺ cDNAs and either rescue line (Fig. 2A–C, dark gray bars). Also, no differences were detected comparing the rutabaga mutant with and without the GAL4–elav driver (data not shown). These results demonstrate the full rescue of the rutabaga mutant learning/memory phenotype in the spatial learning paradigm. In addition, because rutabaga mutant flies are rescued with expression in all or most neurons, ectopic expression is not deleterious to the fly for this behavior. Moreover, the rescue provides an essential tool to determine rutabaga's minimally sufficient spatial expression in the brain for rescue of performance in the spatial learning paradigm. Finally, together with the identification of the molecular defects in several rutabagaalleles (Levin et al. 1992) the rescue clearly identifies therutabaga gene as a Drosophila type I AC.

In a first attempt to localize the minimal brain region sufficient forrutabaga memory formation, arbitrarily chosen GAL4 enhancer trap lines were used to express the rut ⁺ cDNAs in subsets of neurons. The mean 3-min memory was tested for rescue in 17 lines (Table 1). GAL4 enhancer trap lines 30y, 72y, 103y, 201y, 238y, 1407, c115, c232, c271, c492b, c739, and c772 with either of the tworut ⁺ cDNAs were tested. Lines c115, c271, and c772 showed full rescue of the rutabaga mutant memory defect with bothrut ⁺ cDNAs compared with the GAL4–elav rescue flies and wild type w CS10. With line c271 rescue was slightly (but not significantly) lower than with lines c115 and c772. With therut ⁺ cDNA #1 rescue in this line was significantly different from only the rutabaga mutant but not the “no driver” control. The other GAL4 enhancer trap lines did not show significantly higher memory scores with either rut ⁺ cDNA. Five other GAL4 lines (121y, 188y, 189y, H24, and c522) were initially tested with only rut ⁺ cDNA #1. Line c522 had the second highest rescue score of all GAL4 enhancer trap lines withrut ⁺ cDNA #1, but surprisingly, the performance scores withrut ⁺ cDNA #2 failed to reach wild-type levels. It is possible that a threshold for rut ⁺ expression is not reached with this combination of GAL4 driver andUAS_GAL4–rut ⁺. Thus, four of 17 GAL4 enhancer trap lines rescued the rutabaga 3-min memory. Three of the four showed rescue with both rut ⁺ cDNA inserts, whereas line c522 only rescued with rut ⁺ cDNA #1.

Several conclusions can be drawn from these rescue experiments. First, the rescue of the learning/memory phenotype with four GAL4 enhancer trap lines corroborates the GAL4–elav driven rescue, arguing against, for example, unknown genetic interactions in the GAL4–elav rescue flies. Second, only one of 13 GAL4 enhancer trap lines rescues with only one of the two rut ⁺cDNAs. This consistency gives weight to the specificity of the ability of GAL4 enhancer trap lines to rescue and argues against genetic background effects. Finally, if the rescue in the four lines is indeed due to the specific expression of the rut ⁺ cDNA in the neurons labeled in these strains, this suggests that the cAMP-dependent synaptic plasticity is required in only a small subset of neurons for this behavioral plasticity to work properly (see below).

We tested the thermal sensitivity of wild-type and rutabagamutant flies to address the possibility that the rutabagalearning defect was simply due to a defect in heat perception. In the experiment of Figure 3, flies are exposed to a reference temperature (24°C) and a higher test temperature in a chamber with split peltier elements so that only half of the chamber is heated to a preset temperature independent of the fly's behavior. The side of the reference temperature changes, and the test temperature increases in a step-like manner (30°C, 37°C, and 45°C) every minute, alternating sides in each half-experiment. A fly's avoidance of the higher temperatures is an indication of its thermosensitivity. Wild-type and rutabaga mutant flies are similar in sensitivity to temperature differences over the range tested. The thermosensitivity profile of wild-type and rutabaga flies is drastically different from flies lacking antennae, which have no thermopreference at 30°C, yet show normal performance indices in a learning experiment in the spatial learning paradigm (Sayeed and Benzer 1996; T. Zars, unpubl.). It is clear from these results that therutabaga mutation does not affect temperature perception per se, but likely affects the association of spatial position with temperature punishment.

The expression patterns of the GAL4 enhancer trap lines were examined to identify the CNS structures sufficient for memory formation in this paradigm. Frontal brain sections of the four GAL4 enhancer trap lines that rescued the rutabaga phenotype and three that did not are shown in Figure 4. Sagittal brain sections of the four rescue lines are shown in Figure 5. The ventral ganglion was also examined with serial sections (Fig. 5N, and data not shown). At a gross level, the rescuing GAL4 lines show expression in six common structures. These are the antennal lobes, mushroom bodies, median bundle, ellipsoid body, fan-shaped body, and ventral ganglion. Closer examination shows that different substructures are often labeled. We use the nomenclature of Hanesch et al. (1989) andCrittendon et al. (1998) to describe the central complex and mushroom body substructures. The horizontal layers in the fan-shaped body (I–VI) and ellipsoid body (I–III) are diagrammed in Figure 5H. We propose that the ellipsoid body can be divided into three vertical layers (A–C; hence nine domains are distinguished in the ring by the intersections of horizontal and vertical layers) based on differences in GAL4 enhancer trap expression patterns. This is an elaboration of anterior and posterior rings previously proposed (Hanesch et al. 1989). The layers in the fan-shaped body and ellipsoid body, as well as the three lobe systems in the mushroom body (α/β, α′/β′,γ), allow a rough mapping of expression patterns within these structures.

Line c522 strongly labels a subset of antennal lobe glomeruli, as well as fibers in the antennomechanosensory center and the subesophageal ganglion. Weak labeling is observed in the α′ and β′ lobes of the mushroom bodies and the corresponding peduncular and calycal regions (Figs. 4 and 5I). Further stained are subsets of ellipsoid body ring neurons (Figs. 4 and 5D, possibly R1 and R4 neurons). These strongly innervate domains BI and CIII and spare domains AI and AII. A central portion of the median bundle (Fig. 4), two layers of the fan-shaped body (possibly II and III, Figs. 4 and 5D) from two different medial fan-shaped body large field neurons, and other neuron tracts are highlighted.

Line c772 weakly labels a subset of the antennal lobe glomeruli. More strongly labeled are fibers in the antennomechanosensory tract and subsets of Kenyon cells in all three mushroom body substructures (Figs.4 and 5J). The ellipsoid body is labeled in most domains, also possibly from R1 and R4 type ring neurons. They spare domain AII and are restricted to the periphery in BIII and possibly AII (Figs. 4 and 5E). A middle layer (possibly III) of the fan-shaped body is faintly labeled from a lateral type large field tangential neuron (Figs. 4 and 5E). Finally, the central part of the median bundle is strongly stained (Fig. 4).

In line c115, most of the antennal lobe glomeruli as well as most or all parts of the mushroom bodies are detailed. The central part of the median bundle, many of the ellipsoid body ring neurons (sparing AII), and a distal layer (possibly V) of the fan-shaped body are also stained (Figs. 4 and 5 F,K). Finally, line c271 displays a subset of antennal lobe glomeruli and the α,β, and γ (but not α′β′) mushroom body lobes. In the ellipsoid body, exclusively small parts of domains AI and AII are innervated from a novel “ring” neuron that does not have arborizations in the lateral triangles. Labeled fibers project through the ellipsoid body canal. Also, most or all of the median bundle (Figs. 5A,B), and two layers of the fan-shaped body (possibly II and V) are innervated from a median large field fan-shaped body fiber (Figs. 4 and 5G,L).

The ventral ganglions of the four rescuing GAL4 lines were also examined. In all rescuing GAL4 lines at least some expression was found in the leg neurons, the dorsal and ventral paths of the cervical fasciculi, the ventral ellipse, several commissures, as well as other structures (Fig. 5; data not shown). Some of the immunopositive neuropil in the ventral ganglion arose from cell bodies in the ventral ganglion rind. Interestingly, the dorsal and ventral projection pathways from the cervical fasciculi include neurons of the median bundle (Power 1948). The ventral ellipse is an intriguingly organized ventral ganglion structure. Its function is, however, unknown.

Three GAL4 enhancer trap lines represent the 13 expression patterns that do not rescue the rutabaga learning phenotype (Fig. 4). Some patterns are more restricted than those in the rescue lines but others have very similar structures labeled. In line c232, for instance, only a subset of the ellipsoid body ring neurons are marked. This is the most selective pattern in our collection. Line c492b shows subsets of Kenyon cells in all three mushroom body lobes, a subset of the antennal lobe glomeruli, a portion of the median bundle (lateral), and some ring neurons of the ellipsoid body. No expression is detected in the fan-shaped body. Finally, line 30y displays expression in a subset of the antennal lobe glomeruli, all mushroom body lobe systems, some ring neurons of the ellipsoid body, large field neurons of the fan-shaped body, and lateral fibers of the median bundle (not seen in these sections). Expression patterns for lines c772, c271, c492b, 30y, c232, and GAL4–elav have been described in some detail before (Luo et al. 1994; Yao and White 1994; O'Dell et al. 1995; Yang et al. 1995; Tettamanti et al. 1997; Armstrong et al. 1998; Ito et al. 1998). Lines c522 and c115 (as well as others) have been displayed on the Flytrap website (http://www.molgen.gla.ac.uk/flytrap/). The difference in overall gray levels in rescue versus nonrescue GAL4 lines is due to inadvertent differences in signal maturation and the images used here were picked because of their well-preserved cytoarchitecture. Control experiments in which serial sections of all GAL4 enhancer trap lines were treated identically on the same slide showed that c115 had the highest qualitative level of overall expression in the protocerebrum and c232 had the least. The other rescue and nonrescue lines had similar levels of expression in the protocerebrum (data not shown). A correlation between the number of brain cells expressing GAL4 in the different lines and their ability to rescue is not evident. Lines c522, c772, c115, c271, c492b, and 30y have detectable signal in the range of 60 cell bodies per hemisphere (not counting Kenyon cell bodies, which were in the range of hundreds per hemisphere). This is in contrast to c232 that has ∼20 stained cells per hemisphere and no detectable expression in Kenyon cells. This can be compared to the estimated total of 200,000 neurons in theDrosophila central brain (M. Heisenberg, unpubl.) of which ∼70,000 are in a single optic lobe (see Heisenberg and Wolf 1984). The fraction of immunopositive ventral ganglion cells were estimated to be <20% (data not shown). We have, therefore, successfully rescued the rutabaga learning/memory phenotype with expression in ∼1% of the central brain neurons (600/60,000) and a small fraction of ventral ganglion cells.

Can we determine the structure minimally sufficient forrutabaga-dependent learning/memory formation in this paradigm? The first conservative conclusion from these experiments is that in the rescue GAL4 enhancer trap lines, the sufficient set of cells (in the maximal range of a few percent) is expressing the sufficient level of transgenic rutabaga protein to allowrutabaga mutant flies to learn in the spatial paradigm. The second conclusion is that the sufficient cells for learning in this paradigm are neurons because the GAL4–elav line drives expression specifically in neurons (Luo et al. 1994; Yao and White 1994; Ito et al. 1998). Then, each of the rescue lines defines a set of neurons that are sufficient for learning and memory in the spatial learning paradigm. Assuming the UAS_GAL4–Tau expression represents the expression of theUAS_GAL4–rut ⁺ cDNA, we can now identify the common labeled brain structures in the rescue lines as the sufficient set. The most conspicuous regions of the central brain, the mushroom bodies, ellipsoid body, and fan-shaped body are all labeled to some extent. However, different lobe systems of the mushroom bodies are labeled in different lines [e.g., c522 (α′β′) vs. c271 (α,β,γ)]. Also, it has been found that the mushroom bodies are not necessary for learning in this paradigm (Wolf et al. 1998). The GAL4 enhancer trap lines c522, c772, and c115 have strong expression in several types of ring neurons of the ellipsoid body. c271, however, has weak expression in the ellipsoid body and this is in a novel ring neuron that does not have projections in the lateral triangles and seems not to be labeled in the rescue lines c522 and c115. Hence, the four rescue lines appear to have no ellipsoid body neurons stained in common. The fan-shaped body has the most diverse innervation pattern between lines, with some expressing in medially or laterally projecting large field neurons or small field neurons. No common staining pattern in the fan-shaped body emerges from the four lines. The differences in ellipsoid body, fan-shaped body, and mushroom body expression between the four rescue lines at the neuronal level strongly argue against these structures functioning in rutabaga-dependent spatial learning. Only if these structures function in a redundant fashion (e.g., rutabaga expression in any part of the ellipsoid body or fan-shaped body being sufficient) could these neuropil structures still be involved in rutabaga-dependent spatial learning.

This leaves the antennal lobes and median bundle in the central brain and the ventral ganglion. The antennal lobe is labeled in all four rescue lines and, several stained regions of the antennal lobe may be similar between them. The ventral ganglion expression may also have some common structures such as the dorsal and ventral projection pathways from the cervical fasciculi and the ventral ellipse. Finally, expression in the median bundle is similar between the four lines, with expression in at least the medial portion of the bundle and projections to the superior medial protocerebrum.

The median bundle is a good candidate for an integration site from two sensory inputs. The first should be a thermosensory input, possibly from the antennae as has been detected in flies and other insects (Altner and Loftus 1985; Sayeed and Benzer 1996). This could explain the common antennal lobe expression in rescue lines (although the antennal lobe has not been shown in any insect species to be the projection site of antennal thermosensitive neurons). The second is a postulated path integrator (Wehner 1992, 1996). Although the training chamber is relatively simple compared to the desert and forest environments that ants, bees, and spiders must be able to navigate, there is still no light (or other external cue) in this chamber by which Drosophila can orient (Wustmann and Heisenberg 1997). No nervous center has been proposed as yet to house the path integrator. However, the ventral ganglion is a candidate site and, interestingly, it is connected to the superior medial protocerebrum through the median bundle in Drosophila and large flies (Power 1948; Strausfeld 1976). Also, in animals in which necessary structures for a given behavior have been mapped, synaptic modifications generally occur in the sensory input pathway (Milner et al. 1998). This is also consistent with the putative role of the antennal lobe, ventral ganglion, and median bundle. It will be quite exciting to determine whether expression in any single structure can rescue therutabaga-dependent learning in this paradigm.

Is the wild-type Rutabaga protein expressed in any of the structures labeled in the rescue lines? The wild-type Rutabaga protein has elevated expression in the mushroom bodies, as well as expression in the ellipsoid body, antennal lobes, optic lobes, ventral ganglion, and a general low level staining in the protocerebrum (Han et al. 1992;Crittenden et al. 1998). There is also expression in a subset of neurons in the median bundle (T. Zars, unpubl. data). This expression is strongly reduced (but not absent) in a rutabaga P-element allele with the same insertion site as the one used in this study (Han et al. 1992). Therefore, in wild-type flies the candidate structures for minimally sufficient rutabaga-dependent spatial learning have endogenous Rutabaga protein. Whether the exact same neurons labeled in the median bundle, ventral ganglion, and antennal lobes of the rescue lines contain the Rutabaga type I AC is unknown and may be addressed with double labeling experiments.

In conclusion we have, for the first time, rescued the Drosophila rutabaga learning mutation using a wild-type cDNA encoding a type I Ca²⁺/CaM-dependent AC, confirming the identity of the Drosophila rutabaga gene. We have done so using the GAL4 transcription activation system (Brand and Perrimon 1993) that can be used to turn on gene expression in a spatially restricted fashion. We have localized the cells sufficient forrutabaga learning to neurons and likely localized the CNS regions sufficient for learning in this spatial learning paradigm to the antennal lobe, ventral ganglion, and the median bundle.

It should be remembered that there are several limitations of this study. First, the spatially restricted rescue of rutabagamutants can only address rutabaga-dependent synaptic plasticity. There could be other plastic synapses in these flies that are unaffected by the mutation and rescue. Second, the role of therutabaga type I AC in CNS synaptic plasticity is inferred from the larval neuromuscular junction (Zhong and Wu 1991). Third, only theUAS_GAL4–Tau reporter was available to address spatial expression, as all rescue experiments were done in a strong hypomorphic, but not null, rutabaga mutant background (Han et al. 1992). Fourth, precise single neuron anatomical analysis is not yet available nor is temporal control of spatially restricted expression. Finally, ablation-type studies that would give information about necessary brain structures have not yet been exhaustively performed in this paradigm. Once all these technical advances are available and the uncertainties are removed, the anatomy of this memory task will be experimentally accessible in unprecedented detail. Knowledge of the necessary and sufficient brain structures for memory formation may shed new light on the circuit properties of brains in general.