Distinct memory traces for two visualfeatures in the Drosophila brain

Q1

Gang Liu1,2*, Holger Seiler3*, Ai Wen1, Troy Zars3,4, Kei Ito5, Reinhard Wolf3, Martin Heisenberg3 & Li Liu1

The fly Drosophila melanogaster can discriminate and remember visual landmarks. It analyses selected parts of its visualenvironment according to a small number of pattern parameters such as size, colour or contour orientation, and storesparticular parameter values. Like humans, flies recognize patterns independently of the retinal position duringacquisition of the pattern (translation invariance). Here we show that the central-most part of the fly brain, thefan-shaped body, contains parts of a network mediating visual pattern recognition. We have identified short-termmemory traces of two pattern parameters—elevation in the panorama and contour orientation. These can be localized totwo groups of neurons extending branches as parallel, horizontal strata in the fan-shaped body. The central location ofthis memory store is well suited to mediate translational invariance.

Drosophila tethered to a torque meter, with its head (and hence itseyes) fixed in space, can control its orientation with respect to theartificial scenery in a flight simulator1. In this set up, the fly isconditioned to avoid certain flight directions relative to virtuallandmarks (Fig. 1) and recognizes these visual patterns for up to atleast 48 h (ref. 2). Visual pattern recognition in Drosophila has beenstudied in some detail3–7. Flies store values of at least five patternparameters: size, colour, elevation in the panorama, vertical com-pactness, and contour orientation. Moreover, they memorize spatialrelations between parameter values. The neuronal substrate under-lying visual pattern recognition is little understood in any organism.In Drosophila, memory traces can be localized to groups of

neurons in the brain8. Using the enhancer GAL4/UAS expressionsystem9,10, short-term memory traces of aversive and appetitiveolfactory conditioning have been assigned to output synapses ofsubsets of intrinsic neurons of the mushroom bodies (MBs). TheRutabaga protein—a type 1 adenylyl cyclase that is regulated byCa2þ/Calmodulin and G protein, and is considered a putativeconvergence site of the unconditioned and conditioned stimulus inolfactory associative learning11–14—selectively restores olfactorylearning if expressed in these cells in an otherwise rutabaga (rut)-mutant animal15,16. Moreover, expressing a mutated constitutivelyactivating Gas protein (Gas*) in the MBs interferes with olfactorylearning17. Blocking the output from these neurons during memoryretrieval has the same effect, while blocking it during acquisition hasno effect16,18,19. Interestingly, memory traces for other learning tasksseem to reside in other parts of the brain: for remembering itslocation in a dark space, the fly seems to rely on a rut-dependentmemory trace in neurons of the median bundle and/or the ventralganglion20.In the present study, we localize short-term memory traces for

visual pattern recognition to the fan-shaped body (FB), the largestcomponent of the central complex (CX; also called the central bodyin other species). The CX is a hallmark of the arthropod brain. It hasbeen characterized functionally as a pre-motor centre with promi-nent, but not exclusive, visual input (for a review see refs 21, 22). In

the locust, large-field neurons sensitive to the e-vector orientation ofpolarized light have been described in the CX23. Because of itsrepetitive structure and the precisely ordered overlay of fibre projec-tions from the two hemispheres in the FB, neighbourhood relationsof visual space might still be partially preserved at this level (retino-topy)24. Using the genetic approach15,16,20, we now show that a smallgroup of characteristic stratified neurons in the FB house a memorytrace for the pattern parameter ‘elevation’, and a different set ofneurons forming a parallel stratum contain a memory trace for‘contour orientation’.

Central complex defects impair visual pattern memory

Of ten mutants with structural abnormalities in the CX, all wereimpaired in visual pattern recognition25,26. They were able to flystraight and to avoid heat, yet they failed to remember the patterns(Fig. 2a). Did they really lack thememory or had they lost their ability Q2

to discriminate between patterns? Fortunately, individual flies oftendisplay spontaneous preferences for one of the patterns (see Fig. 2c andthe Supplementary Information for the evaluation procedure(Table S1)). In three lines, these preferences were consistent enoughto reveal intact pattern discrimination, suggesting that aberrantcircuitry of the central complex can affect visual learning independentof visual pattern discrimination.As the developmental and structural defects in these mutants are

not well characterized, we used the GAL4/UAS system to acutelyinterfere with CX function. We chose a GAL4 driver line (c205–GAL4) with expression in parts of the CX (expression pattern shownin Fig. 3c) and, as the effector, the gene for tetanus toxin light chain(CntE)27. CntE blocks neurons by cleaving neuronal Synaptobrevin, aprotein controlling transmitter release28. For temporal control, weadded the temperature-sensitive GAL4-specific silencer GAL80under the control of a tubulin promoter (tub–GAL80ts)10. Flies(UAS–CntE/þ; tub–GAL80ts/c205–GAL4) were raised at 19 8C, andwere transferred for 14 h to the restrictive temperature (30 8C)just before the behavioural experiment to induce GAL4-driventoxin expression. Flies kept at the low temperature showed normal

ARTICLES

1State Key Laboratory of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences, 15 Datun Road, Chaoyang District, Beijing 100101, China. 2GraduateSchool of the Chinese Academy of Sciences, Beijing 100039, China. 3Theodor-Boveri-Institut fur Biowissenschaften, Lehrstuhl fur Genetik und Neurobiologie, Am Hubland,97074 Wurzburg, Germany. 4Division of Biological Sciences, 219 Lefevre Hall, University of Missouri, Columbia, Columbia, Missouri 65211, USA. 5Institute of Molecular andCellular Biosciences, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-0032, Japan.*These authors contributed equally to this work.

0 2005|doi:10.1038/nature04381

1

memory scores, while after inactivation of GAL80ts no patternmemory was observed (Fig. 2b). Again, flight control and heatavoidance were normal, and Fourier analysis confirmed that flies atthe high temperature had retained their ability to tell the patternsapart (Supplementary Table S1). As with the structural mutants,interrupting the circuitry of the CX by tetanus toxin expressionseemed to specifically interfere with visual pattern memory. The useof tub–GAL80ts, in addition, excluded the possibility that toxinexpression in unknown tissues during development might causethe memory impairment in the adult. These results do not, as yet,address the question of memory localization.

The rutabaga gene is required for pattern memory

Visual pattern memory in the flight simulator requires an intact rutgene (Fig. 2c, left)29. Mutant rut flies (rut2080) showed normal visualflight control, heat avoidance and pattern discrimination (Fig. 2d).To confirm that the defect was indeed due to the mutation in the rutgene rather than an unidentified second-site mutation, we rescued itby the expression of the wild-type rut cDNA (UAS–rutþ) using the

Figure 1 | Flight simulator for measuring visual pattern recognition.a, Experimental setup. The fly is attached to a torque meter and its yawtorque controls the angular velocity of the panorama surrounding it. It canfly straight and chooses its flight direction relative to the visual patterns inthe panorama. With a beam of infrared light, the fly can be conditioned toavoid certain flight directions (see the Methods for more details). b, Courseof experiment. Bars show performance indices (PI) of successive 2-minintervals of pretest (PI1, PI2), training (dotted bars; PI3, PI4, PI6, PI7) andmemory test (PI5, PI8, PI9) (see the Methods for experimental details anddefinition of PI). The following figures all show PI8 (hatched bar)exclusively. Error bars are s.e.m. throughout. WT, wild type.

Figure 2 | Visual pattern memory is impaired in central complex mutants.Memory (PI8) for the pattern parameter ‘elevation’ was tested using eitherhorizontal bars at different elevations in the panorama or upright andinverted T’s. (The fly can conditionally distinguish the latter only by thedifferent elevations of their centres of gravity6.) a, Visual pattern memory inwild-type (WT) Berlin flies (hatched bar) and ten CX mutants in sixindependent genes. Mutants are in the genetic background of WT Berlin.Experimental details can be found in the Methods, SupplementaryInformation and Fig. 1. None of the mutant memory scores are significantlydifferent from zero. b, Expression of tetanus toxin light chain (CntE) in thedriver line c205–GAL4 labelling parts of the CX interferes with visualpattern memory. Owing to the tub–GAL80ts element, a conditional silencerof GAL4, CntE is expressed only in the adult during a 14 h incubation at30 8C (the restrictive temperature for tub–GAL80ts) before the experiment.c, Visual pattern memory is impaired in the mutant rut2080. Memory scoresafter operant (left) and classical (pavlovian) conditioning (right). Note thatduring classical conditioning, flies are not in command of panoramamotionand are exposed to heat for precisely 50% of the time (PI ¼ 0). Panoramamotion alternates between smooth clockwise and anticlockwise rotation,each for 12 s at an angular velocity of 308 sec21. Memory test is the same as inthe operant experiment. d, Pattern discrimination in the flight simulator. Ifflies show spontaneous individual pattern preferences, orientationhistograms (not shown) have a 1808 periodicity owing to identical patternsin opposing quadrants. Left and middle panels show principal Fouriercoefficients during the 4-min pre-test period for single WT (left) and rut(middle) flies. The 1808 Fourier components (black bars) have the largestamplitude. The same evaluation of a 4-min pre-test with four identicalpatterns (upright Ts) is also shown (right panel). The 1808 component is atbackground level, and the high 908 component indicates equal choice for all4 quadrants. Discrimination values (D) are calculated as D ¼ 2A180/(A120 þ A72) (subscripts refer to Fourier components; A ¼ amplitude).Numbers for D above Fourier spectra are averages ^s.e.m. for n flies (forfurther details, see the Supplementary Information). Numbers below bars ina–c indicate numbers of flies.

ARTICLES NATURE|0 2005

2

pan-neuronally expressing driver line elav–GAL4. Indeed, flies of thegenotype rut 2080/Y;elav–GAL4/UAS–rutþ had normal memory(Supplementary Table S2A).Visual pattern memory in the flight simulator has been shown to

depend upon at least two kinds of behavioural plasticity. For one, anassociative classical (pavlovian) memory trace is formed linking aparticular set of values of pattern parameters to heat. Second, the fly’scontrol of the panorama operantly facilitates the formation of thismemory trace30. Either of the two processes might depend upon theRut cyclase.To address this issue, we tested rutmutant flies in a purely classical

variant of the learning paradigm. During training, panoramamotionwas uncoupled from the fly’s yaw torque and the panorama wasslowly rotated around the fly. Heat was made contingent with theappearance of the ‘punished’ pattern in the frontal quadrant ofthe fly’s visual field. All other parameters were kept as described(see the Methods). For testing memory, panorama motion wascoupled again to yaw torque and the fly’s pattern preferencewas recorded as usual. Even in the absence of operant facilitation,visual pattern memory required the intact rut gene (Fig. 2c, right).Therefore, the rut-dependent memory trace investigated here repre-sents the association of a property of a visual pattern with thereinforcer.

Local rescue of visual pattern memory in the rut mutant

As a first step in localizing the memory trace, we asked in whichneurons of the rutmutant expression of the wild-type rut gene wouldbe sufficient to restore learning. To this end, a total of 27 driver linesexpressing GAL4 in different neuropil regions of the brain were usedto drive the UAS–rutþ effector gene in the rut mutant background.The parameter ‘elevation’ was measured. With seven of the driverlines, pattern memory was restored (104y, 121y, 154y, 210y, c5, c205and c271; Fig. 3 and Supplementary Fig. S2; SupplementaryTable S2B, C).

Comparison of the expression patterns of the 27 lines allowed us tonarrow down the putative site of the memory trace to a small groupof neurons in the brain. The seven rescuing lines all showed transgeneexpression in a stratum in the upper part of the FB. In three of themstaining is rather selective. It comprises, in addition to the FB, only alayer in the medulla, several cell clusters in the suboesophagealganglion and a few other scattered neurons.Evidently, rutþ expression in theMBs is neither necessary (104y, c5,

c205, 154y) nor sufficient for rescue (Supplementary Table S2B). Thisresult is in line with the earlier observation that elimination of morethan 90% of the MBs by hydroxyurea treatment of first-instar larvae31

has no deleterious effect on visual pattern memory32. We ablated theMBs in one group of rescue flies (rut2080/Y;UAS–rutþ/þ;c271/þ).They showed full visual pattern memory (data not shown).Although GAL4 expression in the optic lobes is prominent in all

seven rescuing lines, it occurs in distinctly different layers that do notoverlap (Supplementary Fig. S2). For instance, in 104y expression isrestricted to layer 2, whereas in 210y it is found only in the serpentinelayer (layer 7)33. A similar situation is found for the suboesophagealganglion, although there the staining patterns are more difficult toevaluate. Finally, expression in the ellipsoid body is again notnecessary (104y, c5, c205, 154y) or sufficient (c232, 78y, 7y, and soon) for rescue. Thus, the expression patterns favour the conclusionthat the neurons of the upper stratum of the FB (SupplementaryFig. S2) might be the site of the memory trace for the parameter‘elevation’ in visual pattern memory.Neurons in this stratum, labelled in all seven rescuing lines, have a

very characteristic shape (Supplementary Fig. S3). Their cell bodiesare located just lateral to the calyces. Their neurites run slightlyupward in an antero–medial direction, forming an upward-directedtufted arborization just behind the a/a 0-lobe of the MB. From there,the fibre turns sharply down and backward towards the midline justin front of the FB. Finally, it turns horizontally backward, spreadingas a sharp stratum through all of the FB across the midline. These

Figure 3 | Expression patterns of three driver lines rescuing patternmemory. GAL4-driven green fluorescent protein (GFP) expression in threeof the seven ‘rescuing’ driver lines. Flies mutant for rut show normal visualpattern memory (PI8, green bars on the right), if 104–GAL4, 121y–GAL4 orc205–GAL4 drive rutþ cDNA. Frontal virtual sections of whole-mount

preparations stained with anti-nc82 antibody (staining the synapticneuropil, magenta) and anti-GFPantibody (GAL4 expression, green; overlaywhite). Scale bars 50mm. All ‘rescuing’ driver lines (see SupplementaryFig. S2) show labelling in a dorsal horizontal layer (layer 5) of the FB.

NATURE|0 2005 ARTICLES

3

neurons have been described before in Golgi preparations34.They belong to a larger group of tangential FB neurons calledF neurons34. Besides the stratum in FB, most of them have anarborization in a particular part of the unstructured neuropil.We tentatively classify the layer stained in 104y, and the other sixrescuing lines, as layer 5 (from bottom upward), and hence provi-sionally call the neurons F5, although, without further markers, it isdifficult to reliably number the layers. In summary, expression of Rutcyclase in F5 neurons rescued the rut-dependent memory defect forpattern elevation, whereas no rescue effect was observed in any of 20strains without expression of Rut cyclase in F5 neurons (thoughRut cyclase was expressed in other regions of the brain). Hence, arut-dependent memory trace for pattern elevation may reside inF5 neurons.This finding does not exclude the possibility that memory is

redundant, and that other rut-dependent memory traces for patternelevation might be found elsewhere. Therefore, we asked whetherplasticity in the F5 neurons is necessary for visual pattern memory.As mentioned above, the Rut cyclase is regulated by G protein11–14,and olfactory learning/memory can be blocked by a constitutivelyactive form of the Gas protein subunit (Gas*)

17. We expressed theGas* mutant protein in the FB using the driver line c205, and testedthe flies for their memory of ‘elevation’. Memory was fully suppressed(Fig. 4a). As in olfactory learning17, overexpression of the wild-typeprotein did not interfere with learning (Fig. 4a). These resultssupport the hypothesis that continuous upregulation of Rut cyclasein the F5-neurons interferes with visual short-term memory, imply-ing that F5 neurons are the only site of a rut-dependent memory tracefor pattern elevation.

Pattern specificity of the memory trace

The patterns used in the experiments so far exclusively addressed theparameter ‘elevation’ (upright and inverted T’s or horizontal bars atdifferent elevations). We wondered whether the mutant defect in rutand the Rut rescue in the F5 neurons affected only this parameter, orwhether it applied to other pattern parameters as well. We thereforeextended the study to two further parameters: ‘size’ and ‘contourorientation’. Three driver lines—c205, NP6510 and NP2320—werechosen showing different expression patterns in the FB (Fig. 5a–c). Inthe line NP6510, as in c205, a group of F neurons is marked. They areputatively classified as F1, as their horizontal stratum lies near thelower margin of the FB. Their cell bodies form a cluster in the dorso-frontal cellular cortex above the antennal lobes. Like the F5 neurons,they have large arborizations in the dorsal unstructured neuropil.The line NP2320 expresses the driver in columnar neurons runningperpendicular to the strata of F neurons, with their cell bodiesscattered singly or in small groups between the calyces. As theyseem to have no arborizations outside the FB, they are tentativelyclassified as pontine neurons34.First, we showed that pattern memory required the rut gene for

each of the three parameters (data not shown). Next, we studied theRut rescue flies (for example, rut2080/Y;c205/UAS–rutþ). In the linec205, memory was restored only for ‘elevation’, not for ‘size’ or‘contour orientation’ (Fig. 5a). Correspondingly, the memoryimpairment by expression of dominant-negative Gas* in this driverline should be specific for ‘elevation’, as is indeed the case (Fig. 4b).With the driver line NP6510, memory was not restored for either‘elevation’ or for ‘size,’ but memory was restored for ‘contourorientation’. The third driver line, NP2320, labelling columnarneurons of the FB, did not restore the memory for any of the three

Figure 4 | Rescue and suppression of pattern memory occur in the adult.a, Expression of a constitutively active Gas* protein subunit in the FB(driver line c205–GAL4) abolishes pattern memory. Overexpression ofwild-type Gas protein has no effect. b, Gas* driven by c205–GAL4 disruptsmemory for ‘elevation’ but leaves memory for ‘contour orientation’ intact.c, Temporal specificity of pattern memory. Expression of rutþ cDNA in amutant rut background is blocked during development using tub–GAL80ts

(light hatched bars). Only after inactivation of GAL80ts at 30 8C in the adult,Rut expression is induced and pattern learning/memory restored (blackbars). d, Gas* protein interferes with pattern memory if expression isdelayed until adulthood. Temperature control as in panel c. Numbers belowthe bars indicate numbers of flies.

Figure 5 | Memory traces for pattern parameters are spatially separated.a–c, Three GAL4 lines—c205 (a), NP6510 (b), NP2320 (c)—driving rutþ

cDNA in a mutant rut background, were tested for 3-min memory usingpatterns specifically testing for the single parameters ‘elevation’, ‘size’ or‘contour orientation’. Expression patterns in the FB, using GFP as effector,are shown on the right (staining as in Fig. 3). Note that in NP2320 (c), onlycolumnar elements (presumably ‘pontine neurons’)34 are stained. d, Driverline NP6561 has a similar expression pattern as NP6510. In flies of thegenotype rut2080/Y;NP6561/UAS–rutþ, again only the memory for patternparameter ‘contour orientation’ is rescued. Number of flies: n ¼ 20 in allexperiments.

ARTICLES NATURE|0 2005

4

pattern parameters (Fig. 5c). Among the 27 GAL4 lines above(Supplementary Table S1) we found a second with a very similarexpression pattern as NP6510 (NP6561). The P-element insertions inthe two lines are only 124 nucleotides apart from each other. LikeNP6510, NP6561 restores the memory for ‘contour orientation’ butnot for ‘size’ or ‘elevation’ (Fig. 5d). These results strongly suggestthat memory traces for distinct visual pattern parameters are locatedin different parts of the FB, and that, in addition to the memorytrace in F5 neurons, a memory trace for the parameter ‘contourorientation’ is located in F1 neurons.

Developmental versus adult restoration

A pertinent question in rescue experiments is whether the rescue isdue to the provision of an acute function in the adult or to theavoidance of a developmental defect. Therefore, the tub–GAL80ts

transposon was added to the system. We chose the driverlines c205 and NP6510. Groups of adult males (for example,rut2080/Y;þ/tub–GAL80ts;NP6510/UAS–rutþ), raised at 19 8C, werekept as adults for 14 h at 19 8C or 30 8C. Afterwards, pattern memoryfor the corresponding pattern parameter was tested. In both cases,flies that had been kept at 30 8C showed normal memory, indicatingthat Rut cyclase induced just a few hours before the experiment hadrestored an immediate neuronal function rather than preventinga developmental defect (Fig. 4c). This conclusion was furthersupported by the finding that Gas* expression in the adult (usingtub–GAL80ts) was sufficient to disrupt memory (Fig. 4d).

Conclusions

Several conclusions can be drawn from the above results. Memorytraces in Drosophila are associated with specific neuronal structures:odour memories with the MBs (reviewed in ref. 8), visual memorieswith the CX, and place memory (tentatively) with the medianbundle20. Memory traces are not stored in a common all-purposememory centre. Even within the visual domain, memories fordistinct pattern parameters are localized within distinct structures:a rut-dependent short-term memory trace for the pattern parameter‘elevation’ to F5 neurons, and a corresponding memory trace for‘contour orientation’ to F1 neurons. Moreover, if the constitutivelyactivating Gas* protein indeed interferes with the regulation ofRut cyclase, it follows that the brain contains no other redundantrut-dependent memory traces for these pattern parameters. TheRut-mediated plasticity is necessary and sufficient, at least inF5 neurons. As in the earlier examples, the memory traces areconfined to relatively small numbers of neurons. At least in flies,and probably in insects in general, memory traces appear to be part ofthe circuitry serving the respective behaviour.Our study provides a first glimpse of the circuitry within a neural

system for visual pattern recognition. Though the picture is far fromcomplete, it invites (and may guide) speculation. The FB is a fibrematrix of layers, sectors and shells34. The F1- and F5-neurons formtwo sharp parallel horizontal strata in this matrix. If the width of theFB represents the azimuth of visual space as recently proposed24, thehorizontal strata of the F neurons would be well suited to mediatetranslation invariance (see ref. 7 and the Supplementary Infor-mation). In any case, it is satisfying to find a translation invariantmemory trace in the CX where visual information from both brainhemispheres converges35,22. These first components of the circuitrymay encourage modelling efforts for pattern recognition in smallvisual systems7 (see the Supplementary Information for a discussionof the pattern recognition mechanism).

METHODSThe apparatus for studying visual learning is shown in Fig. 1a (see also refs 36, 37).The tethered fly, suspended at a torque meter, is flying stationarily in the centreof an arena that is illuminated from behind and carries two upright and twoinverted T-shaped patterns in alternating sequence on its wall. The arena isrotated such that its angular velocity is proportional to, but directed against, the

fly’s yaw torque; that is, the fly’s yaw torque determines the angular velocity ofthe arena instead of the fly’s own body. This arrangement allows the tethered flyto stabilize and choose its flight orientationwith respect to the arena by adjustingits yaw torque. A computer continuously registers yaw torque and angularposition of the arena. For visual pattern learning, a beam of infrared light isdirected at the fly as an instantaneous source of heat. The arena is virtuallydivided into four quadrants with the patterns at their respective centres (Fig. 1a;Supplementary Fig. S1). During training, the computer switches the heat beamonwhenever the fly is heading towards a quadrant with, for example, an uprightT, and switches it off when the fly is oriented towards one of the other twoquadrants. Hence, half of all possible orientations in the arena are paired withheat, the other with ambient temperature. During tests, heat is permanentlyswitched off. (In some experiments, a circular LED display is used instead of themechanically rotating arena.)

Angular position is recorded every 50ms and orientation preferences arecalculated for nine consecutive 2-min periods (performance index (PI) 1–9).Pattern A is paired with ambient temperature during training and pattern B withheat. The two patterns alternate as A and B from fly to fly. If tA is the time thefly spends heading towards the quadrants of pattern A, and tB the timeheading towards pattern B quadrants, the performance index is calculated asPI ¼ (tA 2 tB)/(tA þ tB). Supplementary Fig. S1 gives a further explanation.Technical details regarding the flight simulator and data processing routines canbe found in refs 36, 37. Error bars in the figures are s.e.m.; asterisks indicate levelsof significance against zero (***P # 0.001; **P # 0.01; *P # 0.05; n.s., P . 0.05(one-sample t-test; two-sided P-value)).

Details about the visual figures, fly maintenance, mutant lines, histologicalprocedures and image processing are available in the Supplementary Information.

Received 27 May; accepted 25 October 2005.

1. Heisenberg, M. & Wolf, R. On the fine structure of yaw torque in visual flightorientation of Drosophila melanogaster. J. Comp. Physiol. [A] 130, 113–-130(1979).

2. Xia, S., Liu, L., Feng, C. & Guo, A. Memory consolidation in Drosophila operantvisual learning. Learn. Mem. 4, 205–-218 (1997).

3. Dill, M. & Heisenberg, M. Visual pattern memory without shape recognition.Phil. Trans. R. Soc. Lond. B 349, 143–-152 (1995).

4. Dill, M., Wolf, R. & Heisenberg, M. Visual pattern recognition in Drosophilainvolves retinotopic matching. Nature 365, 751–-753 (1993).

5. Dill, M., Wolf, R. & Heisenberg, M. Behavioral analysis of Drosophila landmarklearning in the flight simulator. Learn. Mem. 2, 152–-160 (1995).

6. Ernst, R. & Heisenberg, M. The memory template in Drosophila pattern vision atthe flight simulator. Vision Res. 39, 3920–-3933 (1999).

7. Tang, S., Wolf, R., Xu, S. & Heisenberg, M. Visual pattern recognition inDrosophila is invariant for retinal position. Science 305, 1020–-1022 (2004).

8. Gerber, B., Tanimoto, H. & Heisenberg, M. An engram found? Evaluating theevidence from fruit flies. Curr. Opin. Neurobiol. 14, 737–-744 (2004).

9. Brand, A. H. & Perrimon, N. Targeted gene expression as a means of alteringcell fates and generating dominant phenotypes. Development 118, 401–-415(1993).

10. McGuire, S. E., Le, P. T., Osborn, A. J., Matsumoto, K. & Davis, R. L.Spatiotemporal rescue of memory dysfunction in Drosophila. Science 302,1765–-1768 (2003).

11. Dudai, Y., Corfas, G. & Hazvi, S. What is the possible contribution ofCa2þ-stimulated adenylate cyclase to acquisition, consolidation and retentionof an associative olfactory memory in Drosophila. J. Comp. Physiol. [A] 162,101–-109 (1988).

12. Levin, L. R. et al. The Drosophila learning and memory gene rutabaga encodes aCa2þ/Calmodulin-responsive adenylyl cyclase. Cell 68, 479–-489 (1992).

13. Abrams, T. W., Yovell, Y., Onyike, C. U., Cohen, J. E. & Jarrard, H. E. Analysis ofsequence-dependent interactions between transient calcium and transmitterstimuli in activating adenylyl cyclase in Aplysia: possible contribution toCS–-US sequence requirement during conditioning. Learn. Mem. 4, 496–-509(1998).

14. Renger, J. J., Ueda, A., Atwood, H. L., Govind, C. K. & Wu, C. F. Role of cAMPcascade in synaptic stability and plasticity: ultrastructural and physiologicalanalyses of individual synaptic boutons in Drosophila memory mutants.J. Neurosci. 20, 3980–-3992 (2000).

15. Zars, T., Fischer, M., Schulz, R. & Heisenberg, M. Localization of a short-termmemory in Drosophila. Science 288, 672–-675 (2000).

16. McGuire, S. E., Le, P. T. & Davis, R. L. The role of Drosophila mushroom bodysignalling in olfactory memory. Science 293, 1330–-1333 (2001).

17. Connolly, J. B. et al. Associative learning disrupted by impaired Gs signalling inDrosophila mushroom bodies. Science 274, 2104–-2107 (1996).

18. Dubnau, J., Grady, L., Kitamoto, T. & Tully, T. Disruption of neurotransmissionin Drosophila mushroom body blocks retrieval but not acquisition of memory.Nature 411, 476–-480 (2001).

19. Schwaerzel, M., Heisenberg, M. & Zars, T. Extinction antagonizes olfactorymemory at the subcellular level. Neuron 35, 951–-960 (2002).

NATURE|0 2005 ARTICLES

5

20. Zars, T., Wolf, R., Davis, R. & Heisenberg, M. Tissue-specific expression of atype I adenylyl cyclase rescues the rutabaga mutant memory defect: in searchof the engram. Learn. Mem. 7, 18–-31 (2000).

21. Heisenberg, M. Central Brain Function in Insects: Genetic Studies on theMushroom Bodies and Central Complex in Drosophila. (eds Schildberger, K. &Elsner, N.) (Gustav Fischer, Stuttgart/Jena/New York, 1994).

22. Strauss, R. The central complex and the genetic dissection of locomotorbehaviour. Curr. Opin. Neurobiol. 12, 633–-638 (2002).

23. Homberg, U. In search of the sky compass in the insect brain.Naturwissenschaften 91, 199–-208 (2004).

24. Strauss, R. Die ubergeordnete Steuerung des Laufverhaltens durch dasInsektengehirn, studiert mit Methoden der Drosophila-Neurogenetik. Habilitationthesis, Bayerische Julius-Maximilians-Universitat, Wuerzburg (2002).

25. Weidtmann, N. Visuelle Flugsteuerung und Verhaltensplastizitat beiZentralkomplexmutanten von Drosophila melanogaster. Diploma thesis,Bayerische Julius-Maximilians-Universitat, Wuerzburg (1993).

26. Strauss, R. & Heisenberg, M. A higher control center of locomotor behaviour inthe Drosophila brain. J. Neurosci. 13, 1852–-1861 (1993).

27. Keller, A., Sweeney, S. T., Zars, T., O’Kane, C. J. & Heisenberg, M. Targetedexpression of tetanus neurotoxin interferes with behavioural responses tosensory input in Drosophila. J. Neurobiol. 50, 221–-233 (2002).

28. Broadie, K. et al. Syntaxin and synaptobrevin function downstream of vesicledocking in Drosophila. Neuron 15, 663–-673 (1995).

29. Eyding, D. Lernen und Kurzzeitgedachtnis beim operanten Konditionieren aufvisuelle Muster bei strukturellen und biochemischen Mutanten von Drosophilamelanogaster. Diploma thesis, Bayerische Julius-Maximilians-Universitat,Wuerzburg (1993).

30. Brembs, B. & Heisenberg, M. The operant and the classical in conditionedorientation of Drosophila melanogaster at the flight simulator. Learn. Mem. 7,104–-115 (2000).

31. Prokop, A. & Technau, G. M. Normal function of the mushroom body defectgene of Drosophila is required for the regulation of the number and

proliferation of neuroblasts. Dev. Biol. 161, 321–-337 (1994).32. Wolf, R. et al. Drosophila mushroom bodies are dispensable for visual, tactile,

and motor learning. Learn. Mem. 5, 166–-178 (1998).33. Fischbach, K. F. & Dittrich, A. P. M. The optic lobe of Drosophila melanogaster.

A Golgi analysis of wild-type structure. Cell Tissue Res. 258, 441–-475 (1989).34. Hanesch, U., Fischbach, K. F. & Heisenberg, M. Neuronal architecture of the

central complex in Drosophila melanogaster. Cell Tissue Res. 257, 343–-366(1989).

35. Vitzthum, H., Muller, M. & Homberg, U. Neurons of the central complex of thelocust Schistocerca gregaria are sensitive to polarized light. J. Neurosci. 22,1114–-1125 (2002).

36. Wolf, R. & Heisenberg, M. Basic organization of operant behaviour as revealedin Drosophila flight orientation. J. Comp. Physiol. [A] 169, 699–-705 (1991).

37. Heisenberg, M. & Wolf, R. Reafferent control of optomotor yaw torque inDrosophila melanogaster. J. Comp. Physiol. [A] 163, 373–-388 (1988).

Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.

Acknowledgements We thank B. Gerber and H. Tanimoto for valuablecomments on the manuscript, C. Grubel, Haiyun Gong and Huoqing Jiang forexcellent technical assistance, and A. Jenett for visualizing gene expressionpatterns. Supported by Deutsche Forschungsgemeinschaft, Fonds derChemischen Industrie (M.H.), National Natural Sciences Foundation of China(L.L.), ‘973-program’ (L.L.), and by the Knowledge Innovation Project of theChinese Academy of Sciences (L.L.).

Author Information Reprints and permissions information is available atnpg.nature.com/reprintsandpermissions. The authors declare no competingfinancial interests. Correspondence and requests for materials should beaddressed to L.L. (liuli@sun5.ibp.ac.cn) or M.H.(heisenberg@biozentrum.uni-wuerzburg.de).

ARTICLES NATURE|0 2005

6

Distinct memory traces for two visual features in the Drosophila brain

Supplementary Information

Gang Liu1,2, Holger Seiler3, Ai Wen1, Troy Zars3,4, Kei Ito5, Reinhard Wolf3, Martin Heisenberg3, Li Liu1

1 State Key Laboratory of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences, 15 Datun Road, Chaoyang District, Beijing 100101, P.R. China

2 Graduate School of the Chinese Academy of Sciences, Beijing 100039, China

3 Theodor-Boveri-Institut für Biowissenschaften, Lehrstuhl für Genetik und Neurobiologie, Am Hubland, 97074 Würzburg, Germany

4 Division of Biological Sciences, 219 Lefevre Hall, University of Missouri, Columbia, Columbia, MO 65211, USA

5 Institute of Molecular and Cellular Biosciences, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-0032, Japan

___________________________________________________________________________ This document provides supplementary information for the above article in Nature. We

• show the dynamics of flight orientation behaviour of a single fly during different stages of a learning experiment in the flight simulator (Fig. S1),

• present methods and procedures used in the study, • discuss in detail the flies' ability to spontaneously discriminate visual patterns, • present additional data on this subject (Table S1), • show a compilation of the results of all 28 rutabaga rescue experiments

accomplished in this study (Table S2), • show the expression patterns for all 7 driver lines which rescue the rut-dependent

memory defect for pattern elevation (Fig. S2), • present a 3-D reconstruction of the tangential layer-5 Neurons (F5) in the fan-

shaped body (Fig. S3), • discuss the implications of our findings for models of visual pattern recognition,

and • briefly review the history of translation invariance in Drosophila.

___________________________________________________________________________

Flight performance and extraction of performance indices (PI's)

Fig. S1: Lower panel: Selected time traces for yaw torque T(t) (green) and relative flight direction ψ(t) (blue/red) of a single wild type (WT) fly. Shown are the three stages (pretest, training, and memory test) of a standard learning experiment. Quadrants of the panorama are marked with two different colors [yellow: quadrants containing pattern type associated with heat during training (dangerous); turquoise: quadrants containing pattern type not associated with heat during training (safe)]. Upper panel shows the position histograms, i.e. frequency distributions of all flight directions occurring during the 120 seconds of the experiment shown as a time trace below. During the first two minutes of pretest (left, raw data for PI1), the fly frequently changes its flight direction. It seemingly disregards the positions of the visual landmarks in the panorama. [Other flies may display spontaneous pattern preferences during the pretest (see below: supplementary notes on pattern discrimination ability).] The performance index PI can either be calculated from the time trace by separately accumulating the fractions of time for flight directions towards the dangerous quadrants (red) and the safe quadrants (blue), and subsequently dividing their difference by the total flight duration (see also methods), or it can be calculated from the differently colored areas in the histogram. Middle and right columns (PI3 and PI8): Time traces T(t), ψ(t), and position histograms for the first training period and for the first memory test. Already after the first two hits of the IR laser beam, the fly develops a pronounced preference for safe quadrants (middle). This preference persists beyond the training phase and is recorded during the subsequent memory test (right). Note, that prior to the beginning of every 120s-period, the orientation of the panorama is randomized.

Methods

Two to six days old flies were used in the experiments. Wild-type strains Berlin (WTB) and Canton-S (WT-CS) as well as the following mutants were used: rutabaga with (rut2080; UAS-rut+) and without rut+-cDNA (rut2080; ry), no-bridgeks49 (nobks49), three alleles of ellipsoid-body-open (eboks263, ebo1041, ebo678), three alleles of central-body-defect (cbd2254, cbdks171, cbd ks96), central complex broad (ccbks127), central complex deranged (ccdks135), and central brain deranged (ceb849, neuroglianceb849) were used1. P[Gal4] enhancer trap lines were kindly provided by D. Armstrong, K. Kaiser and T. Raabe (see Table S2). The UASGal4-CntE stock was provided by S.T. Sweeney and C.J. O’Kane, the tub-GAL80ts, UAS-TNT stock by H. Tanimoto, Würzburg. It should be noted that only male flies were used in the rescue experiments and were F1 progeny of a genetic cross (e.g., rut2080; UAS-rut+ females crossed to P[Gal4] males).

For tethered flight, flies were briefly immobilized by cooling to 4°C and were glued by their head and thorax to a small hook (copper wire; diameter 0.05 mm). After return to room temperature they were kept individually in small moist chambers overnight with a few grains of sucrose.

The apparatus for studying visual learning is shown in Fig. 1a2,3. The tethered fly is suspended at a torque meter in the centre of an arena which is illuminated from behind and carries two upright and two inverted T-shaped patterns in alternating sequence on its wall. T patterns measure 36° vertically and 39° horizontally; the bars of the Ts are 12° wide. Of these patterns flies memorize only the different heights of their centres of gravity (COGs)4,5. In some of the experiments the Ts are replaced by horizontal bars (40° x 12°) placed at 12° above and below the horizon. The light source is a 100W, 12V tungsten-iodine bulb.

The arena is rotated by an electric motor such that its angular velocity is proportional to but directed against, the fly's yaw torque, i.e. the fly's yaw torque determines the angular velocity of the arena instead of rotating the fly's own body. This arrangement allows the tethered fly to stabilize and choose its flight orientation with respect to the arena by adjusting its yaw torque. A computer continuously registers yaw torque and angular position of the arena. In some of the experiments the mechanically rotating arena is replaced by a cylindrical display of 180 x 49 light emitting diodes (LED's). In this setup, the computer continuously calculates from the fly's yaw torque the hypothetical angular velocity of the fly and the corresponding displacements of the visual patterns displayed on the LED screen. For visual pattern learning, a beam of infrared light is directed at the fly as an instantaneous source of heat. With the mechanical arena setup, the beam can be intercepted by a computer-driven electric shutter, whereas with the LED arena setup an infrared laser beam (P = 0...150 mW, λ = 826 nm) is switched on and off by the computer . The arena is virtually divided into four quadrants with the patterns at their respective centers (Fig. 1a; Fig. S1). During training the computer switches the heat beam on whenever the fly is heading into a quadrant with e.g. an upright T, and switches it off when the fly is oriented towards one of the other two quadrants. Hence, half of all possible orientations in the arena are paired with heat, the other with ambient temperature. During tests heat is permanently switched off.

Angular position is recorded every 50 ms and orientation preferences are calculated for 9 consecutive 2-min periods (performance index, PI1-9). Pattern A is paired with ambient temperature during training and pattern B with heat. Each of the two patterns (i.e. the upright and inverted T in all experiments) is pattern A in half of the experiments. This procedure eliminates the effects of spontaneous or non-associatively induced pattern preferences. If tA is the time the fly spends heading towards the quadrants of pattern A, and tB the time heading

towards pattern B quadrants, the performance index (PI) is calculated as PI = (tA-tB)/(tA+tB). Mean PIs are obtained for 9 consecutive 2-min periods (PI1-9). Before each 2-min period the arena is rotated to an arbitrary position. Fig. S1 gives further explanations. Technical details regarding the flight simulator and data processing routines can be found in2,3. Error bars in the figures are S.E.M.s, asterisks indicate levels of significance against zero. *** p≤0.001; ** p≤0.01; * p≤0.05; n.s.: p>0.05 (one-sample t-test; two-sided p-value).

Pattern discrimination ability is derived from the flies' spontaneous pattern preferences during the two successive 2-min intervals of the pretest (PI1, PI2), by evaluation of the 180° periodicity in the flight orientation histograms. The 180° periodicity is measured as the amplitude of the second component in the Fourier spectrum. For more details, see the supplementary information.

Immunohistochemistry: For wholemount preparation and staining of brains6, flies were anesthetized and brains were dissected in Drosophila ringer by stripping off the head capsule including the eyes and laminae. Brains were fixed overnight in 2% para-formaldehyde at 4°C and stained using polyclonal rabbit anti-GFP antibody (1:1000, Molecular Probes) and monoclonal mouse antibody nc82 (1:100) for neuropil staining. As second antibodies we used anti-rabbit conjugated to Alexa 488 (1:100, Molecular Probes, Eugene, Oregon) and anti-mouse conjugated to Cy3.18 (1:250, Jackson Immuno Research, West Grove, Pennsylvania). One µm optical sections were acquired using a Leica CLSM / Aristoplan confocal microscope equipped with a Zeis pbjective (Plan Neofluar 20x; numerical aperture: 0.8). For virtual sections of 3D reconstructions extended AMIRA®-software5 was used. Supplementary notes on pattern discrimination ability

Several genotypes do not exhibit visual pattern memory. In principle, this may be due to a loss of memory or to a defect in pattern discrimination. Given the alternating sequence of the equally spaced patterns in the panorama (Fig. 1a), any flight direction with reference to a pattern type (e.g. upright Ts) occurs at a periodicity of 180°. Therefore, a periodicity of 180° in the flight orientation histogram during the pre-test (PI1, PI2) indicates the fly's ability to discriminate the patterns. However, if no 180° periodicity is observed, the inverse conclusion can not be drawn: The fly might still be able to discriminate the patterns. The 180° periodicity in the flight orientation histograms is measured as the amplitude of the second component in the Fourier spectrum. As the absolute amplitudes in the Fourier spectra differ considerably between flies, we decided to relate the amplitude of the 180° component to the mean amplitude of the (near-by) 120° and 72° components. These are considered 'background' since they have no relations to the periodicity of the patterns in the panorama.

The chance value for D is 1. As expected, in an experiment with 4 equal patterns, the pattern discrimination is not significantly different from chance (D = 1.3 ± 0.2, Fig. 2d). Interestingly, the rutabaga mutant, which neither in operant, nor in classical conditioning experiments shows any pattern memory (Fig. 2c), is able to discriminate the patterns as well as WT (Fig. 2d).

Definition of Pattern Discrimination Ability :

D = 2xA180/(A120+A72)

As already mentioned, an insignificant value of D does not prove a defect in pattern discrimination. For instance, in the rescue experiment line 121y shows a normal memory score (PI8 = 0.24 ± 0.06; Fig. 3b and table S2 C), yet, during pre-test it does not reveal any pattern discrimination (D = 1.3 ± 0.2). Significantly positive D values for genotypes showing no pattern memory are listed in Table S1.

ability of pattern discrimination line

significance parameter

PI8 in Fig.

rut ∗∗∗∗∗∗∗∗∗∗∗∗ height 2c

ebo678 ∗∗∗∗ height

eboKS263 ∗∗∗∗∗∗∗∗ height

nobKS49 ∗∗∗∗ height

2a

UAS-CntE/+, Gal80/c205 (30°C) ∗∗∗∗∗∗∗∗∗∗∗∗ height 2b

UAS-Gααααs∗∗∗∗/+; c205/+ ∗∗∗∗∗∗∗∗ height 4a, b

UAS-Gααααs∗∗∗∗/+; c205/Gal80 (30°C) ∗∗∗∗ height 4d Table S1. Pattern discrimination ability of lines deficient in pattern memory. For definition of D, see text. Levels of significance against D = 1.3 ± 0.2 (WT with 4 equal patterns, Fig.2d) are shown in the 2nd column. *** p≤0.001; ** p≤0.01; * p≤0.05 (one-sample t-test; two-sided p-value). For 7 strains with impaired pattern recognition pattern discrimination can be demonstrated.

Fig. S2: Expression patterns of driver lines rescuing pattern memory. GAL4-driven Green fluorescent protein (GFP) expression in seven ‘rescuing’ driver lines (104y; 121y; 154y; 210y; c5; c205; c271). Flies mutant for rut show normal visual pattern memory (PI8: green bars on right) if GAL4 drives rut+-cDNA. Frontal virtual sections of whole-mount preparations stained with anti-nc82 antibody (synaptic neuropil; magenta) and anti-GFP antibody (GAL4 expression; green; overlay white). Scale bars 50 µm. All driver lines show label in the 2nd most dorsal horizontal layer (layer 5) of the fan-shaped body. Little if any overlapping expression is found in other parts of the brain.

Table S2. Males of the general genotype rut2080/Y; driver/+; +/UAS-rut+ were tested for visual pattern recognition in the flight simulator (pattern parameter "height") as shown in Fig. 1. Only PI8 and levels of significance against zero are shown in the 3rd column. *** p≤0.001; ** p≤0.01; * p≤0.05; (one-sample t-test; two-sided p-value). In the 4th column, the most saliently marked structures of the central brain are indicated. The column on the right gives references for the GAL4 lines. (1) Flytrap; (2) Wuerzburg collection; (3) Flybase; (4) Zars et al., 2000a; (5) Martin et al., J Comp Physiol [A] 185, 277-88 (1999); (6) http://flymap.lab.nig.ac.jp/getdb.html. All P[GAL4] transposons are inserted on the 2nd or 3rd chromosomes.

Fig. S3: Three views of the cluster of F5-neurons in the central brain of 104y/+;+/UAS-GFP flies. F5-neurons stained with anti-GFP (yellow), neuropil (mushroom bodies, fan-shaped body, ellipsoid body) with anti-dCAST antibodies (blue transparent). (A) View from behind and above (about 60° downward); (B) also from behind but at an angle of about 30°; (C) oblique view from the front and side. Whether staining in lower part of fan-shaped body also belongs to this cell cluster is not known. See Experimental procedures and description of individual F5-neurons in the text.

Implications for models of visual pattern recognition

Pattern recognition in Drosophila is little understood, so far. Mechanisms of pattern

recognition that are translation invariant must be designed to locally detect values for pattern

parameters (features), i.e. one of several different colours, sizes, contour orientations, etc.. In

the conceptual model proposed earlier5 each parameter is divided into a small number of

domains and each domain has its own set of small-field feature detectors in the optic lobes, -

each detector reporting the presence or absence of the feature and each set covering the whole

visual field. The domains of a parameter must each be represented by at least one neuron in

the memory matrix. The model predicts that each of the F5- or F1-neurons represents one

parameter domain and that all neurons of one group together represent the whole parameter. If

all neurons of one stratum would represent only one single domain (e.g. contour orientation

+45° in F1-neurons), we would expect only a 50% rescue (Rut) and 50% suppression (Gαs*)

of the memory in the discrimination task used here, because in 50% of the flies heat was

associated with contour orientation -45°. Critical tests for the model (and improved models)

can now be designed.

Translational invariance

In early studies of pattern recognition in Drosophila7 flies, surprisingly, did not show

translational invariance in the flight simulator. They failed to recognize patterns shifted up or

down by 9° or more after training. Pattern recognition seemed to require the same retinal

coordinates for acquisition and retrieval. Later it was discovered that for these experiments

patterns had been chosen that could only be discriminated by the pattern parameter 'elevation

of the centre of gravity'4. Reinvestigation of this problem by Tang et al.5 revealed that indeed

vertical displacement specifically interferes with the parameter 'elevation'. Other parameters

such as size, colour and contour orientation, as well as relational cues are recognized after

vertical displacement. Finally, even the pattern parameter 'elevation' can be recognized at a

new retinal position, as Tang et al. showed using stabilized retinal images and horizontal

displacements.

Supplementary references 1) Strauss R, Heisenberg M, J. Neurosci. 13, 1852 (1993). 2) Wolf R, Heisenberg M, J. Comp. Physiol. [A] 169, 699 (1991). 3) Heisenberg M, Wolf R, J. Comp. Physiol. [A] 163, 373 (1988). 4) Ernst R, Heisenberg M, Vision res. 39, 3920 (1999). 5) Tang et al., Science 305, 1020 (2004). 6) Rein KH, et al., Curr. Biol. 12, 227 (2002). 7) Dill M et al., Nature 365, 751 (1993).