Standardized atlas of the brain of the desert locust, Schistocerca gregaria

Page 1

REGULAR ARTICLE

Standardized atlas of the brain of the desertlocust, Schistocerca gregaria

Angela E. Kurylas & Torsten Rohlfing &

Sabine Krofczik & Arnim Jenett & Uwe Homberg

Received: 14 January 2008 /Accepted: 31 March 2008 /Published online: 27 May 2008# Springer-Verlag 2008

Abstract In order to understand the connectivity ofneuronal networks, their constituent neurons should ideallybe studied in a common framework. Since morphologicaldata from physiologically characterized and stained neuronsusually arise from different individual brains, this can onlybe performed in a virtual standardized brain that compen-sates for interindividual variability. The desert locust,Schistocerca gregaria, is an insect species used widely forthe analysis of olfactory and visual signal processing,endocrine functions, and neural networks controlling motoroutput. To provide a common multi-user platform for neuralcircuit analysis in the brain of this species, we havegenerated a standardized three-dimensional brain of thislocust. Serial confocal images from whole-mount locustbrains were used to reconstruct 34 neuropil areas in tenbrains. For standardization, we compared two differentmethods: an iterative shape-averaging (ISA) procedure by

using affine transformations followed by iterative nonrigidregistrations, and the Virtual Insect Brain (VIB) protocol byusing global and local rigid transformations followed bylocal nonrigid transformations. Both methods generated astandard brain, but for different applications. Whereas theVIB technique was designed to visualize anatomical vari-ability between the input brains, the purpose of the ISAmethod was the opposite, i.e., to remove this variability. Anovel individually labeled neuron, connecting the lobula tothe midbrain and deutocerebrum, has been registered into theISA atlas and demonstrates its usefulness and accuracy forfuture analysis of neural networks. The locust standard brainis accessible at http://www.3d-insectbrain.com.

Keywords Virtual Insect Brain protocol .

Iterative shape-averaging . Three-dimensional reconstruction .

Standard brain . Virtual neuroanatomy.

Schistocerca gregaria (Insecta)

Introduction

Neural networks are composed of synaptically connectedand functionally associated neurons. Analysis of neuralcircuits, therefore, requires the staining of the participatingneurons selectively, to reconstruct their three-dimensional(3D) morphology and to visualize them within a commonframework. Although the anatomical and functional prop-erties of the brains of a particular species follow a commonstructural design, interindividual differences in shape andsize do indeed exist. In order to combine neuronal datafrom different individual brains, it is highly desirable toperform this in a representative reference brain model thatcompensates these differences in shape and size. With theuse of special averaging procedures, we have established a

Cell Tissue Res (2008) 333:125–145DOI 10.1007/s00441-008-0620-x

This work was supported by the Deutsche Forschungsgemeinschaft(U. Homberg; HO 950/14-3) and the National Institute on AlcoholAbuse and Alcoholism (T. Rohlfing; AA05965 and AA13521).

A. E. Kurylas :U. Homberg (*)Fachbereich Biologie, Tierphysiologie,Philipps Universität Marburg,35032 Marburg, Germanye-mail: [email protected]

T. RohlfingNeuroscience Program, SRI International,Menlo Park CA 94025, USA

S. KrofczikInstitut für Neurobiologie, Freie Universität Berlin,14195 Berlin, Germany

A. JenettInstitut für Genetik und Neurobiologie, Universität Würzburg,97074 Würzburg, Germany

Page 2

locust brain atlas that can be used to integrate neuronalstructures from different individual brains.

The 3D visualization of neuronal structures and theembedding of functional data into standardized brain mapsof humans and of other vertebrate species have greatlyadvanced neuroinformatic efforts to analyze brain complexity(Toga 2002; Van Essen 2002; Martone et al. 2004). The smallsize of insect brains together with their uniquely identifiableneurons renders them ideally suited for an analysis offunctional circuits within the framework of the whole brain.Toward this goal, 3D standard insect brains have beengenerated for the fruitfly, Drosophila melanogaster (Rein etal. 2002), and for the honeybee, Apis mellifera (Brandt et al.2005). In addition, individual 3D maps of whole brains orreconstructions of single brain structures such as the antennallobe have been established for various species, including thecockroaches Diploptera punctata (Chiang et al. 2001) andLeucophaea maderae (Reischig and Stengl 2002), the mothsLobesia botrana (Masante-Roca et al. 2005), Manduca sexta(Rospars and Hildebrand 2000; Huetteroth and Schachtner2005), Helicoverpa assulta, and Heliothis virescens (Berg etal. 2002; Rø et al. 2007), Agrotis ipsilon (Greiner et al.2004), and Spodoptera littoralis (Sadek et al. 2002), thehoneybee (Galizia et al. 1999), the fruit fly (Laissue et al.1999; Iyengar et al. 2006; Jefferis et al. 2007), and theparasitoid wasps Cotesia glomerata and C. rubecula (Smidet al. 2003).

In addition to fruit flies, honeybees, and moths, locustsare well-established model systems in neuroscience.Locusts are major pest insects. Since their large brain iseasily accessible, they have been favorable subjects forstudies of the visual system (Rind 1987, 2002; Simmons2002), the olfactory system (Laurent 1996, 2002), braindevelopment (Ludwig et al. 2001; Boyan et al. 2003),endocrine functions (Veelaert et al. 1998), the control offlight and walking (Burrows 1996), and mechanisms ofspatial orientation (Homberg 2004; Pfeiffer and Homberg2007; Heinze and Homberg 2007). The locust brain, likethat of other insects, consists of three divisions termedthe protocerebrum, deutocerebrum, and tritocerebrum. Thelargest part of the brain, the protocerebrum, includes theoptic lobes, the mushroom bodies, the central complex, andvarious areas in the inferior, ventral, and superior proto-cerebrum (Homberg 1994). The two optic lobes are theprimary visual centers of the brain. Each optic lobe isdivided into the lamina, medulla, and lobula complex. Thelobula complex is subdivided into an anterior, a dorsal, aninner, and an outer lobe (Elphick et al. 1996). Visualprojection areas in the median protocerebrum include theanterior optic tubercles, consisting of upper and lower units.The lower unit is part of the polarization vision pathway(Homberg et al. 2003; Pfeiffer et al. 2005). The pairedmushroom bodies are conspicuous neuropil structures in the

median protocerebrum; they consist of a stalk, two lobes,and a primary and an accessory calyx. The mushroombodies have been implicated in olfactory coding (Jortner etal. 2007; Cassenaer and Laurent 2007) and, as demonstrat-ed in other species, are essential for olfactory learning andmemory (Davis 1993; Heisenberg 1998; Rybak und Menzel1998; Müller 1999; Farris et al. 2001; Lozano et al. 2001;Malun et al. 2002). The central complex is a prominentstructure in the center of the brain. It consists of theprotocerebral bridge and the central body. The central bodyis divided into an upper and lower division and a pair ofnoduli. The central complex serves a role in sky compassorientation (Heinze and Homberg 2007) and, as shown inflies, in right-left motor coordination (Strauss and Heisenberg1993; Strauss 2002) and visual memory (Liu et al. 2006).The paired lateral accessory lobes are closely associated withthe central complex. The antennal lobes, the antennalmechanosensory and motor centers, and the glomerularlobes are subdivisions of the deutocerebrum. Antennal-lobeprojection neurons provide direct olfactory input to thecalyces of the mushroom body and to the lateral horn of theprotocerebrum (Laurent 1996; Anton and Hansson 1996).The tritocerebrum, posterior and ventral to the deutocere-brum, connects the brain to the stomatogastric nervoussystem (Homberg 1994).

In the present study, we have established two 3Dstandard brain atlases of the locust, Schistocerca gregaria,as a basis for future analysis of the neuroarchitecture of thelocust brain. A standard brain must faithfully representshapes and sizes of the brain and any of its substructuresthat are characteristic for the particular species, in order tofacilitate the mapping of the morphology of additionalindividual elements into the spatial representation of theatlas. To meet these requirements, various approaches ofimaging and mapping methods have been developed (forreviews, see Toga and Thompson 2001; Toga 2005). Toestablish a locust standard brain, we have evaluated twodifferent published techniques: (1) the Virtual Insect Brain(VIB) protocol as used for the Drosophila standard brain(Rein et al. 2002; Jenett et al. 2006), and (2) the iterativeshape-averaging (ISA) technique, based on the method usedfor the honeybee standard brain (Rohlfing et al. 2001;Brandt et al. 2005). The results obtained by these twotechniques are compared, and the usefulness of both stan-dards for different purposes is discussed.

To demonstrate the value of the locust standard brain,we have registered a single neuron into the locust ISAstandard. The selected neuron, an interneuron connectingthe lobula complex to the midbrain and deutocerebrum, isrelated to a group of lobula-protocerebrum neuronsreported in the migratory locust Locusta migratoria(Gewecke and Hou 1992; Gewecke and Hou 1993; Sternand Gewecke 1993).

126 Cell Tissue Res (2008) 333:125–145

Page 3

Materials and methods

Animals

Experiments were performed on adult males (n=29) andfemales (n=12) of the desert locust, Schistocerca gregaria, inthe gregarious phase. Animals were reared under crowdedconditions at the University of Marburg at 28°C and on a12:12 light/dark cycle. For comparison of body size of theanimals, we measured the length of the pronotum and thedistance between the eyes by using digital vernier calipers(accuracy 10 μm; Guogen, Matatakitoyo, Taiwan). Aftertissue preparation, photographs of the brains were taken witha Zeiss Axioskop compound microscope equipped with aPolaroid DMC Ie digital camera (Polaroid, Cambridge,Mass., USA). Brain sizes were quantified by measuring thedistances between both optic stalks and the distances betweenthe dorsal and ventral midline furrows. We tested for acorrelation between body size and brain size (Pearson’scorrelation) and compared body and brain sizes of male andfemale locusts (two-sided t-test, unpaired; Fig. 1). We ensuredthat the brains used for the standardization were an adequatesample of the 41 locusts with respect to their size. Theultimate choice was made in favor of the brains that exhibitedthe most intense staining and thus were an appropriate basisfor segmentation. Statistical evaluation was performed by

using Excel XP (Microsoft, Redmond, Wash.) and Origin 6.0(Microcal Software, Northampton, Mass.).

Reconstruction of locust brains

Histology For construction of the standard brain, ten adultmale locusts were immobilized with ice and decapitated. Theheads were prefixed for about 60 min in 4% formaldehyde/0.01 M phosphate buffer/0.3 M NaCl (4% FA/0.01 M PBS;pH 7.4) at room temperature to minimize tissue distortionsand dislocation of the optic lobes. After dissection inhyptotonic saline (Clements and May 1974), brains werefixed overnight in 4% FA/0.01 M PBS at 4°C. After rinses in0.1 M PBS containing 0.3% Triton X-100 (PBT; pH 7.4), theganglionic sheath was made permeable by treatment with1 mg/ml collagenase-dispase (in 0.05 TRIS-HCl, pH 7.6) for1 h. Following another washing step, the brains werepreincubated overnight with 5% normal goat serum (NGS)and 0.02% sodium azide in PBT at 4°C. To visualize theirsubdivisions, the brains were incubated with a monoclonalantibody against the synaptic protein synapsin (SYNORF1,kindly provided by Dr. E. Buchner, Würzburg, Germany)diluted at 1:50 in PBT, 1% NGS, and 0.02% sodium azidefor 5–6 days at 4°C. The brains were then washedthoroughly with PBT. Secondary antibody (Cy5-conjugatedgoat anti mouse; Jackson Immunoresearch, Westgrove, Pa.)

Fig. 1 Comparison of body and brain size between female (n=12) andmale (n=29) locusts, and the selection of male locusts used to createthe standard brain (sample, n=10). For quantification of body size, thelength of the pronotum (P) and the distance between the outer rims ofthe compound eyes (interocular distance, IO) were measured (see bodyscheme, top view). Means and standard deviation: P males 8.41±0.39 mm, females 9.06±0.47 mm, sample 8.39±0.39 mm; IO males5.91±0.22 mm, females 6.18±0.18 mm, sample 5.93±0.21 mm. Forquantification of brain size, the distance between the optical stalks

(OS) and the distance between the dorsal and ventral midline furrows(F) were measured (see black brain scheme, frontal view). Means andstandard deviation: OS males 1.47±0.08 mm, females 1.50±0.10 mm,sample 1.49±0.08 mm; F males 0.98±0.07 mm, females 0.95±0.09 mm, sample 0.95±0.09 mm. Differences in body size betweenmale and female locusts are significant, but no significant differenceswere observed in brain size. The ten brains used for the standardizationwere a representative sample of all locust brains measured

Cell Tissue Res (2008) 333:125–145 127

Page 4

was used at a dilution of 1:300 in PBT, 1% NGS, and 0.02%sodium azide for up to 3 days. After being washed, brainswere dehydrated in an ethanol series (25%, 50%, 70%, 90%,95%, 100%, 15 min each), prepared for clearing in a solutionof 50% ethanol/50% methyl salicylate, and cleared withmethyl salicylate (Merck, Darmstadt, Germany) until trans-parent (at least 40 min). Finally, the brains were mounted inPermount (Fisher Scientific, Pittsburgh, Pa.) between twoglass cover-slides, which were separated by spacing rings toavoid squeezing.

Confocal microscopy Optical sections were obtained byusing a confocal laser scanning microscope (CLSM, LeicaTCS SP2) equipped with a 10× oil objective (HC PL APO10×/0.4 Imm Corr CS; Leica, Bensheim, Germany). Fluores-cence was excited with the 633-nm line of a HeNe laser,detected with an emission spectrum of 650–750 nm, andquantized with a resolution of 8 bits. We estimated the scalingfactor in the z-direction to be 1.6 (Bucher et al. 2000; Brandt etal. 2005). Because of their large size (about 4.1×2.7×0.8 mm), brains were imaged sequentially in 2×3 partiallyoverlapping single scans (three stacks from anterior and fromposterior, respectively). Each stack was scanned at 512×512pixel resolution in the xy-plane and 68–214 optical sections(3 μm) axially. Stacks were combined semi-automaticallywith the help of the AlignSlices editor and the Transformeditor in Amira 3.1 software (Mercury Computer Systems,San Diego, Calif.). The merged image data had a final size ofabout 300 MBytes per image. For further processing, it wasnecessary to downsample the image data laterally to half ofthe original dimensions. This resulted in final image volumesof 241–278 slices in the z direction, with each slice consistingof 616–731 pixels in the x direction and 396–502 pixels inthe y direction and a voxel size of 5.9×5.9×3 μm.

Image segmentation and reconstruction Neuropil areas ofinterest were labeled with the segmentation editor in Amira3.1 on a PC (AMD Athlon64 3000+ processor, 1.81 GHz,1 GB RAM running Windows XP; graphics card: ATIRadeon X800XL; ATI Technologies, Ontario, Canada). Inthis procedure, each voxel was assigned to a particularneuropil, thereby providing anatomical meaning to eachvoxel, in addition to the merely physical property of itsstaining intensity. The corresponding volumetric data ofthis labeling process were called the “label field”. The labelfields created were regular cubic grids with the samedimensions as the underlying microscopy images. Thereconstruction of polygonal surface models, morphometricanalysis, and shape-averaging were mainly based on theselabel fields. In some brains, we reconstructed 48 neuropils,but only 34 of these could be reconstructed reproduciblyand thus were used to compose the standard brain. Thevolumes of segmented neuropils in our sample of ten brains

were determined by using the “TissueStatistics” tool inAmira. It enabled us to count the number of voxelscontained in a region, the volume of a particular neuropil,and the coordinates of the neuropil center. Conventionalstatistical analysis was performed on these data by usingExcel XP. We averaged the volumes and the coordinates ofthe centers of the reconstructed neuropils. The brain thatmost closely resembled the averages in neuropil volumesand locations was defined as the template brain. In addition,we ensured that its neuropil counterstaining was mostsuitable for the iterative registration process.

The terminology for general brain structures is based onthe nomenclature of Strausfeld (1976), for central-complexsubdivisions on Müller et al. (1997), and for lobulacomplex subdivisions on Gouranton (1964). The orientationof brain structures is given with respect to the body axis ofthe animal.

Creating the standard brain/registration

The goal for constructing the locust standard brain was tocreate an approximation to the typical brain shape for thisspecies. We achieved this by computing an average-shapebrain from ten individual male brains. Two different methodsfor establishing the standard brain were used. For comparisonof the resulting averaged label images of either technique, weperformed a comparative analysis based on relative volumesand relative distances. For volumetric analysis, we comparedthe mean relative volumes of each segmented structure of theindividual brains with the relative volumes of the VIB andISA results. Relative volumes were obtained by normalizingthe volume of a particular structure to the sum of all structures.Relative distances were calculated by normalizing thedistances of the centers of each segmented structure relativeto the center of the entire label field. For analyzing the relativelocations, we compared the averaged relative distances for aparticular brain structure of the individual brains with therelative distances of the VIB and ISA structures. The requiredcenters of gravity for the respective segmented structures wereobtained with the “TissueStatistics” tool in Amira. Theanalysis was performed by using Excel XP and Origin 6.0.

VIB protocol The registration method and application ofthe VIB protocol have been described in detail elsewhere(Zöckler et al. 2001; Rein et al. 2002; Schindelin 2005;Jenett et al. 2006). Therefore, we will summarize onlybriefly the components of the VIB protocol that we used tocreate the locust standard brain. The VIB protocol wasimplemented by using Amira 3.1. First, because of thecomputational limitations of PC resources, the image datawere downsampled once again. The final image volume forthis procedure contained 121–139 slices in the z direction,

128 Cell Tissue Res (2008) 333:125–145

Page 5

whereas single slices consisted of 308–365 pixels in thex direction and 198–251 pixels in the y direction with avoxel size of 11.7×11.7×6 μm. Second, for each segment-ed neuropil structure, the mean and standard deviation of itsvolume and its center of gravity were computed automat-ically in the VIB protocol. The alignment was a three-stepapproach. For initial alignment of the brains, a global rigidtransformation (rotation, translation, and isotropic scaling)was calculated according to the centers of gravity of thelabel fields. Next, the labeled structures were aligned moreaccurately with a local rigid transformation. Hereby, thestructures were matched individually by maximizing theoverlap volume with the corresponding template structurewithout taking into account the other structures. Thus, asmany contradictory rigid transformations were computed asstructures were defined during the segmentation. In the laststep, the contradictory transformations were merged andextrapolated to the areas in between by using a diffusionalgorithm (Jenett et al. 2006). The resulting vector field ofthis nonrigid transformation was applied to the 3D images,thereby bringing them into a common coordinate system.Finally, two 3D images were computed: an average gray-level image and a probabilistic atlas of the labeled neuropilstructures. The latter encodes the probability of eachstructure being present at a given location. These probabil-ity maps were also used as a measure of transformationquality, as they represent the degree of overlap between theindividual images through the registration process.

ISA method The shape-averaging method is an improvedversion of a technique introduced by Rohlfing et al. (2001)and applied to the brain of the honeybee, Apis mellifera, byBrandt et al. (2005). In an iterative procedure, successivelyrefined average images are generated, starting with a fuzzyaverage after aligning all input CLSM images by transla-tion, rotation, and anisotropic scaling, i.e., by using affinetransformations. In subsequent iterations, increasingly finerinter-individual shape differences are averaged out by usingnonrigid alignment of the individual images to the currentaverage image.

To align the individual CLSM images with each otherand with the evolving standard brain is an image registra-tion problem, i.e., it requires finding the set of coordinatetransformations that maps anatomically correspondingpoints in all images onto each other. We solved thisproblem by applying a fully automatic, image intensity-based, nonrigid registration algorithm similar to the oneintroduced by Rueckert et al. (1999). The particularimplementation used herein has been described in detailby Rohlfing and Maurer (2003). Developments in imageregistration methodology have led to improved robustnessof the procedure. Therefore, we used CLSM imagesdirectly for the averaging procedure, unlike prior work

(Rohlfing et al. 2001), which applied the averaging tosegmented label images. As an immediate benefit, coordi-nate transformations inside anatomical structures (whichhad no internal texture in label images) were no longerdetermined exclusively by the properties of the coordinatetransformation model, but instead took advantage of therich local structure visible in the CLSM images. Wequantified the convergence through a recently developedmeasure of groupwise image overlap (Crum et al. 2005)that was computed from the ensemble of individual labelmaps, transformed into the standard image space. Tosubstantiate the claim that the result of the averagingprocedure indeed approximated the “average” shape of theinput sample of locust brains, we computed the magnitudesof the deformation between the standard brain and each ofthe input brains and compared these numbers with thedeformation magnitudes required to map each of theindividual brains onto each of the others (analogous toFig. 8 in Brandt et al. 2005). The “nonrigidity” of eachdeformation was measured by its second-order bendingenergy (Wahba 1990). It was computed as the integral

Z@2T

@x2

� �2

þ @2T

@y2

� �2

þ @2T

@z2

� �2

þ2@2T

@x@y

� �2

þ @2T

@y@z

� �2

þ @2T

@z@x

� �2" #

d x; y; zð Þ

over the entire reference-image coordinate space of therespective deformation field. This energy is related to thephysical bending energy of a thin metal plate. It is zero foraffine transformations, which makes it appropriate toquantify the degree of “warping” of a deformation.

Lobula projection neuron

Histology The neuron was dye-injected during intracellularrecording. The tip of the recording electrode (resistance:60–140 MΩ) was filled with 4% Neurobiotin in 1 M KCl.A positive current (3 nA for 3 min) was passed through thetip of the electrode to inject Neurobiotin into the neuron. Adiffusion time of about 1 h was allowed before the brainwas dissected out of the animal, fixed overnight in 4% FA/0.01 M PBS at 4°C, and rinsed in phosphate-buffer (0.1 M).

Since the registration of the neuron depends on thetransformation of corresponding labels of neuropils, neuro-pil counterstaining was performed with the anti-SYNORF1monoclonal antibody. The immunostaining procedure wasthe same as that used for the standard brain individuals,with the exception that, during incubation with thesecondary antibody Cy5-GAM (1:300 in PBT, 1% NGS,and 0.02% sodium azide), visualization of the neuron was

Cell Tissue Res (2008) 333:125–145 129

Page 6

achieved by adding Cy3-conjugated streptavidin (1:1000;Jackson Immunoresearch Westgrove, Pa.). After thoroughrinses with 0.1 M PBS, the brain was dehydrated in anincreasing ethanol series (15 min each) and cleared inmethylsalicylate for 35 min. Finally, the brain was mountedin Permount between coverslips. Spacers were used toprevent squeezing.

Confocal imaging The whole-mount preparation was evalu-ated by using confocal microscopy (Leica TCS-SP2). In orderto register the neuron into the standard brain by applying thetransformation of labeled neuropil structures (see below), wescanned neuron- and neuropil-staining with two channelssimultaneously. Neuropil fluorescence was excited with the633-nm line of a HeNe laser and was detected with anemission spectrum of 650–750 nm. Fluorescence of theneuron was excited with the 543-nm line of the HeNe laserand detected with an emission spectrum of 550–620 nm. Wescanned the brain sequentially at 10× magnification (HC PLAPO 10x/0.4 oil lens) in two partially overlapping singlescans. The resolution in the xy-plane was 1024×1024 pixelwith a step size of 3 μm in the z direction.

Reconstruction The tiled scans obtained from the CLSMwere combined to a complete dataset for both channels. Forneuron reconstruction, a semi-automated tool was used(Schmitt et al. 2004; Evers et al. 2005), which wasimplemented as a custom module in Amira 3.1 (Schmitt etal. 2004). The neuron was reconstructed by selectingbranching points and end points of the cell. The centerlineand diameter of each neuronal branch was fitted by usinglocal intensity gradients (described in detail in Schmitt et al.2004; Evers et al. 2005). Finally, a surface was assigned tothe fitted skeleton by using the local intensity gradients ofthe original dataset.

Fitting of neuron into the standard brain The procedure formapping the neuron into the standard brain followed theprocedure described by Brandt et al. (2005). The neuronwas registered into the standard brain by applying thetransformation parameter calculated for the registration ofthe corresponding label field of the neuropils into the locuststandard brain. Therefore, we segmented the same neuropilstructures as were used for the standard brain. As describedin Brandt et al. (2005), the registration of the labeledneuropils into the standard brain was then determined byusing a two-step registration. First, the label fields werealigned by translation, rotation, and anisotropic scaling,with a 9-degree of freedom affine registration. Second, anonrigid registration was performed to compensate localshape differences. The underlying metric takes the spatialcorrespondence of two label fields into account. Theresulting transformation matrix and the deformation field

(vector field) containing the nonrigid component of thetransformation were applied to the geometric representationof the neuron image. To assess the influence of registrationon the morphology of the labeled neuron, (1) we analyzedthe distances between all corresponding points of theneuron (vertices) before and after registration, and (2) wecompared the lengths of all segments of the reconstructions,defined as the distances between two branching points,before and after registration.

Results

We have constructed a locust standard brain from tenindividual male brains. For establishing the locust standardbrain, neuropils with high synaptic density were recon-structed. We used the antibody SYNORF1 for the stainingof synaptic areas, whereas somata and tracts largelyremained unstained. We applied and compared two differ-ent standardization methods. The body size between maleand female locusts differed significantly, but no differencesin brain size were observed (two-sided t-test, unpaired;Fig. 1). The choice of the ten male brains was arepresentative sample of our population of 41 locusts withrespect to their brain size (Fig. 1).

Reconstructed neuropils of the locust brain

Because the intensity of synapsin immunostaining was similarin different neuropils, the specification of particular neuropilsby distinct gray values failed. Therefore, we manuallyassigned each voxel to a defined label field, representing oneof the preselected brain structures (Fig. 2). We used data from34 neuropils to compose the standard brain. A conventionalvolumetric analysis was performed with the labeled imagesof our sample of ten brains (Table 1); additionally, we

Fig. 2 Confocal images from a locust brain immunostained forsynapsin. Neuropil regions were labeled manually and were stored inseparate image volumes, which were related to defined color codes.a Image from a frontal slice through the brain at 190 μm below thebrain surface. b Frontal section at the level of the central body atapproximately 340 μm depth. c Horizontal slice at 90 μm from the topof the midbrain and about 700 μm depth from the top of the opticlobes. d Horizontal slice at approximately 430 μm from the top of themidbrain (ACa accessory calyx of the mushroom body, AL antennallobe, ALo anterior lobe of the lobula complex, AMe accessorymedulla, AOTu anterior optic tubercle with LU being the lower unitand UU being the upper unit, CBL lower division of the central body,CBU upper division of the central body, d dorsal, DLo dorsal lobe ofthe lobula complex, DRMe dorsal rim area of the medulla, ILo inner lobeof the lobula complex, l lateral, LH lateral horn, Me medulla, mL mediallobe of the mushroom body,MN midbrain neuropil, No noduli, OLo outerlobe of the lobula complex, p posterior, P peduncle, PB protocerebralbridge, PCa primary calyx of the mushroom body, vL vertical lobe of themushroom body). Bar in d 200 μm (also applies to a-c)

�

130 Cell Tissue Res (2008) 333:125–145

Page 7

Cell Tissue Res (2008) 333:125–145 131

Page 8

Tab

le1

Meanvo

lume(M

V),standard

deviation(SD),standard

error(SE),relativ

evo

lume(RV),relativ

estandard

deviation(RSD

),andrelativ

estandard

error(RSE

)of

34segm

entedlocustbrain

compartments(n=10

).The

volume(V)andrelativ

evo

lume(RV)of

theiterativ

eshape-averaging(ISA

)andVirtual

Insect

Brain

(VIB)results

aregivenrigh

t

Structure

Abb

reviations

MV

(μm

3)

SD

(μm

3)

SE(μm

3)

RV%

RSD%

RSE%

ISA

VIB

V(μm

3)

RV%

V(μm

3)

RV%

Midbrainneurop

ilMN

2.90

×10

81.04

×10

81.81

×10

741

.65

35.74

6.22

2.78

×10

841

.77

3.23

×10

843

.11

Medulla

(right)

Me-r

1.22

×10

84.40

×10

77.65

×10

617

.48

36.06

6.28

1.19

×10

817

.95

1.26

×10

816

.81

Medulla

(left)

Me-l

1.23

×10

84.91

×10

78.56

×10

617

.67

39.89

6.94

1.2×10

817

.98

1.27

×10

816

.99

Accessory

medulla

(right)

AMe-r

6.29

×10

52.27

×10

53.95

×10

40.09

36.11

6.29

4.41

×10

50.07

7.05

×10

50.09

Accessory

medulla

(left)

AMe-l

5.30

×10

51.98

×10

53.45

×10

40.08

37.43

6.52

3.83

×10

50.06

5.90

×10

50.08

Anteriorlobe

ofthelobu

lacomplex

(right)

ALo-r

4.25

×10

61.43

×10

62.49

×10

50.61

33.63

5.85

3.7×10

60.56

4.73

×10

60.63

Anteriorlobe

ofthelobu

lacomplex

(left)

ALo-l

4.55

×10

61.18

×10

62.06

×10

50.65

25.94

4.52

4.82

×10

60.72

5.07

×10

60.67

Dorsallobe

ofthelobu

lacomplex

(right)

DLo-r

1.27

×10

63.14

×10

55.46

×10

40.18

24.67

4.30

9.42

×10

50.14

1.49

×10

60.20

Dorsallobe

ofthelobu

lacomplex

(left)

DLo-l

1.46

×10

64.37

×10

57.61

×10

40.21

29.98

5.22

1.16

×10

60.18

1.63

×10

60.22

Innerlobe

ofthelobu

lacomplex

(right)

ILo-r

3.92

×10

61.14

×10

61.98

×10

50.56

29.05

5.06

3.36

×10

60.51

4.31

×10

60.57

Innerlobe

ofthelobu

lacomplex

(left)

ILo-l

4.19

×10

61.44

×10

62.50

×10

50.60

34.32

5.97

4.51

×10

60.68

4.48

×10

60.60

Outer

lobe

ofthelobu

lacomplex

(right)

OLo-r

1.28

×10

74.74

×10

68.24

×10

51.84

37.00

6.44

1.09

×10

71.63

1.34

×10

71.79

Outer

lobe

ofthelobu

lacomplex

(left)

OLo-l

1.32

×10

75.84

×10

61.02

×10

61.89

44.24

7.70

1.21

×10

71.82

1.38

×10

71.84

Dorsalrim

area

ofthemedulla

(right)

DRMe-r

8.83

×10

52.66

×10

54.63

×10

40.13

30.11

5.24

4.32

×10

50.06

9.76

×10

50.13

Dorsalrim

area

ofthemedulla

(left)

DRMe-l

9.27

×10

53.28

×10

55.70

×10

40.13

35.37

6.16

6.55

×10

50.10

1.01

×10

60.13

Low

erun

itof

thecentralbo

dyCBL

8.17

×10

53.55

×10

56.17

×10

40.12

43.40

7.56

7.54

×10

50.11

8.85

×10

50.12

Upp

erun

itof

thecentralbo

dyCBU

5.83

×10

62.69

×10

64.68

×10

50.84

46.13

8.03

4.83

×10

60.73

6.07

×10

60.81

Nod

ulus

(right)

No-r

9.42

×10

45.81

×10

41.01

×10

40.01

61.63

10.73

8.13

×10

40.01

8.63

×10

40.01

Nod

ulus

(left)

No-l

9.64

×10

45.71

×10

49.93

×10

30.01

59.19

10.30

7.65

×10

40.01

8.98

×10

40.01

Protocerebral

bridge

PB

1.66

×10

65.86

×10

51.02

×10

50.24

35.21

6.13

1.37

×10

60.21

1.78

×10

60.24

Pedun

culus(right)

P-r

1.33

×10

74.94

×10

68.60

×10

51.91

37.09

6.46

1.31

×10

71.97

1.42

×10

71.90

Pedun

culus(left)

P-l

1.20

×10

73.80

×10

66.62

×10

51.72

31.80

5.54

1.19

×10

71.79

1.30

×10

71.74

Primarycalyx(right)

PCa-r

1.25

×10

74.25

×10

67.40

×10

51.80

33.90

5.90

1.13

×10

71.70

1.38

×10

71.84

Primarycalyx(left)

PCa-l

1.17

×10

73.42

×10

65.95

×10

51.68

29.10

5.06

1.03

×10

71.55

1.32

×10

71.75

Accessory

calyx(right)

ACa-r

5.70

×10

62.57

×10

64.47

×10

50.82

45.07

7.84

4.13

×10

60.62

6.01

×10

60.80

Accessory

calyx(left)

ACa-l

5.66

×10

62.47

×10

64.29

×10

50.81

43.63

7.59

4.58

×10

60.69

6.17

×10

60.82

Lateral

horn

(right)

LH-r

1.66

×10

67.47

×10

51.30

×10

50.24

44.95

7.82

9.09

×10

50.14

1.67

×10

60.23

Lateral

horn

(left)

LH-l

1.78

×10

69.32

×10

51.62

×10

50.26

52.33

9.11

1.11

×10

60.17

1.81

×10

60.24

Low

erun

itof

theanterior

optic

tubercle

(right)

AOTuL

U-r

2.31

×10

57.27

×10

41.26

×10

40.03

31.41

5.47

4.54

×10

50.07

2.02

×10

50.03

Low

erun

itof

theanterior

optic

tubercle

(left)

AOTuL

U-l

3.03

×10

51.09

×10

51.90

×10

40.04

35.96

6.26

3.97

×10

50.06

2.74

×10

50.04

Upp

erun

itof

theanterior

optic

tubercle

(right)

AOTuU

U-r

7.26

×10

53.05

×10

55.30

×10

40.10

41.94

7.30

1.14

×10

60.17

7.23

×10

50.10

Upp

erun

itof

theanterior

optic

tubercle

(left)

AOTuU

U-l

7.27

×10

52.61

×10

54.54

×10

40.10

35.88

6.25

1.12

×10

60.17

7.50

×10

50.10

Antennallobe

(right)

AL-r

1.89

×10

77.34

×10

61.28

×10

62.72

38.76

6.75

1.85

×10

72.78

1.97

×10

72.63

Antennallobe

(left)

AL-l

1.94

×10

76.51

×10

61.13

×10

62.78

33.63

5.85

1.89

×10

72.84

2.05

×10

72.73

132 Cell Tissue Res (2008) 333:125–145

Page 9

calculated means and standard deviations of the absolute andrelative volumes of the brain compartments. For someindividual brains, we could reconstruct 48 neuropils(Fig. 3), but because of technical difficulties, some neuropilscould not be reconstructed reproducibly (see below).

Distinct immunostaining allowed us to distinguish sevenneuropils in the optic lobe reliably: the medulla, the dorsalrim area of the medulla, the accessory medulla, and theanterior, dorsal, outer, and inner lobes of the lobulacomplex (Fig. 4a,e,f). Small remains of retina led toconfocal scan artefacts in the lamina so that black “spots”in the lamina made it impossible to reconstruct itcompletely. The reconstruction of the lamina shown inFig. 3 was performed by interpolating these parts withoutstaining information. However, this procedure was judgedinadequate to create a representative standard brain.

Therefore, the lamina and its dorsal rim area (Fig. 3) werenot included in the standard brain.

In themedian protocerebrum,we reconstructed and includedthe lateral horns, the anterior optic tubercles (upper and lowersubunits), the mushroom bodies (primary calyces, secondarycalyces, and combined peduncle-lobes; Fig. 4b,g,h), and thecentral complex (protocerebral bridge, upper and lowerdivision of the central body, noduli; Fig. 4c,i,j) in the standardatlas. In the deutocerebrum, the antennal lobes were includedin the standard brain. The remaining neuropil areas of thebrain were difficult to distinguish reproducibly and weretherefore segmented and assigned collectively as “midbrainneuropil”. These included the lateral accessory lobes of theprotocerebrum (Fig. 4d,i), the glomerular lobes, the mediancrescents and the antennal mechanosensory and motor centersof the deutocerebrum, and the tritocerebrum (Fig. 4d,k,l).

Fig. 3 Surface reconstruction ofan individual locust brain:anterior view (a), posteriorview (b). Neuropils not includedin the standard brain are thelamina (La), dorsal rim area ofthe lamina (DRLa), antennalmechanosensory and motorcenter (AMMC), glomerular lobe(GL), median crescent (MC),lateral accessory lobe (LAL), andtritocerebrum (TC). Otherneuropils are also shown (ALantennal lobe, AMe accessorymedulla, AOTu anterior optictubercle, Ca calyx, CC centralcomplex, DRMe dorsal rim areaof the medulla, Lox lobulacomplex, LH lateral horn, MBmushroom body, Me medulla,PC protocerebrum). Bar 400 μm

Cell Tissue Res (2008) 333:125–145 133

Page 10

Fig. 4 Surface reconstruction of neuropils of an individual locust brain,anterior views. a-d Three-dimensional (3D) polygonal surface models of(a) the lobula complex, (b) the mushroom body, (c) the central complex,and (d) the ventro-median areas of the brain. a Four subdivisions weredistinguished in the lobula complex: the anterior (ALo), dorsal (DLo),inner (ILo), and outer (OLo) lobe of the lobula complex. b The primary(PCa) and the accessory calyces (ACa) of the mushroom body weresegmented separately, whereas the peduncle (P) and the medial andvertical lobes (mL, vL) were assigned to a common label. c In the centralcomplex, the protocerebral bridge (PB), the upper (CBU) and lower(CBL) divisions of the central body, and the two noduli (No) could bedistinguished. The upper division of the central body is transparent, tovisualize the posteriorly located paired noduli. d Neuropil areas in the

ventro-median brain. For better visualization the antennal lobe wasomitted. e, f Confocal images of the anti-synapsin-stained lobula complexat approximately 180 μm in e and 350 μm in f beneath the frontal surfaceof the optic lobe. g, h Confocal images of the mushroom body atapproximately 130 μm (g) and 350 μm (h) depth from the frontal brainsurface. i, j Confocal images of the central complex at approximately320 μm (i) and 390 μm (j) depth below the frontal brain surface. kConfocal image of the deutocerebrum at approximately 360 μm depthfrom the frontal brain surface. l Confocal image at approximately 480 μmdepth from the frontal brain surface (AMMC antennal mechanosensoryand motor center, GL glomerular lobe, LAL lateral accessory lobe, MCmedian crescent, TC tritocerebrum). Bars 100 μm (a [also applies to e, f],c, i-l), 200 μm (b [also applies to g, h], d)

134 Cell Tissue Res (2008) 333:125–145

Page 11

The locust standard brain

We compared two different methods to establish a locuststandard brain: the VIB protocol (Rein et al. 2002; Jenett etal. 2006) and the ISA procedure modified from thetechnique used by Rohlfing et al. (2001). After reconstruc-tion of all brains, a representative template brain wasselected for both registration methods. We selected thebrain that most closely resembled the averaged label fieldsin neuropil volumes and location of centers of gravity. Anadditional criterion was the quality of neuropil stainingregarding intensity and distinctness. The latter criterion wasimportant for the iterative registration, because the CLSMimages were used directly in our averaging procedure.

VIB protocol The resulting data files of the VIB protocolwere an average gray image and probability maps thatcontained all transformed neuropils of all brains in acommon file (Fig. 5). To generate a 3D polygonal surfacemodel from the probability maps (Fig. 5a-c), the averagelabel field was calculated by applying an adaptive thresh-olding procedure. To assign certain voxels to a corre-sponding label, a threshold for the respective probabilityvalue was determined (in the best case: 100%). With someexceptions (the threshold for the noduli and the anterioroptic tubercles were 30%), this threshold was about 50%. Adirect volume rendering of the resulting average labelimages displayed the standardized label images in acommon view without manually interfering by determininga threshold (Fig. 5d,e). Hereby, nonconformity of labelssuch as the lower subunits of the anterior optic tubercles orthe noduli is visible. The probability map for the averagedlabeled structures, used as a measure of transformationquality (Fig. 5f), indicates the variation of position andshape.

ISA method The resulting standard brain of the ISA methodwas an average image, which, in the process of the iterativeprocedure, converged to become increasingly sharp andwell defined and at last was successively refined (Fig. 6).As illustrated in Fig. 7, the overlap between the transformedindividual label fields, which has a maximum value of 1.0for perfect overlap of all label fields, increased monoton-ically through the iteration and appeared to reach an upperlimit at the end of the fourth (and final) nonrigidregistration pass. The resulting images of the fourthnonrigid registration iteration were selected as the standardbrain. The resulting files of the ISA method were anaverage gray image and an average label image (Fig. 8).The surface reconstruction and a direct volume rendering ofthe corresponding average gray image of the locuststandard brain is shown in Fig. 8 in different views. Toconfirm quantitatively that the deformation of an individual

image to match the average is smaller than the mean inter-subject deformation, we calculated nonrigid distancesbetween the ISA standard brain and each of the individualbrains, and for a comparison between each pair of twoindividual brains (Fig. 9).

Comparison of VIB and ISA results For a comparison ofthe results of both standardization methods, we analyzedtheir relative volumes and relative distances in relation tothe respective means of the individual brains (Fig. 10).Relative volumes are the normalized volume of a particularstructure to the sum of all structures. Relative distances arethe normalized distances of the center of each segmentedstructure relative to the center of the entire label field. Therelative volumes of structures of the VIB standard showedless deviation from the mean relative volumes of individualbrain structures than those of the ISA standard (Fig. 10a,c).This was no surprise because preservation of structurevolumes is built into the design of the VIB method.However, the ISA brain showed larger invariance in relativedistances (Fig. 10b,d).

Registration of a single neuron into the ISA standard

To elucidate central nervous mechanisms underlying sen-sory processing and motor control, electrophysiologicalrecordings combined with intracellular staining of neuronsin the locust brain are performed routinely (Anton andHansson 1996; Heinze and Homberg 2007; Jortner et al.2007; Kinoshita et al. 2007). Typically, only data from asingle neuron are obtained from an individual brain. Foranalysis of neural networks, it is therefore desirable to fitneuronal morphologies obtained from different preparationsinto a standardized reference brain. Because of interindi-vidual shape differences between brains, the brains con-taining the respective neurons have to be merged to thesame shape. Additionally, the local relationship betweenneuropils has to be preserved.

To demonstrate the usefulness of the locust standardbrain, we registered a visual interneuron into the locust ISAstandard (Fig. 11). The neuron derived from an electro-physiological recording; it showed phasic-tonic excitationsto frontal and dorsal illumination but was not sensitive tothe E-vector orientation of polarized light (data not shown).The neuron was labeled by intracellular injection ofNeurobiotin, and the brain was additionally treated withthe neuropil marker anti-synapsin. The transformationparameters were calculated by the registration of thecorresponding segmented neuropils and were applied tothe reconstructed neuron.

The neuron chosen is a projection neuron from the lobulacomplex to the proto- and deutocerebrum (Fig. 11). Its soma

Cell Tissue Res (2008) 333:125–145 135

Page 12

lies anterior-ventrally in the optic stalk. The neuron haswide-field dendritic arborizations in the ventral hemisphereof the inner layer of the outer lobe of the lobula and has asecond dendritic arbor in dorsal stratum 2 of the anterior lobe

of the lobula (Fig. 11a,e,f; for terminology, see Homberg etal. 2003). Its axon runs toward the brain midline and givesrise to two major collaterals. One fiber projects with beadedterminals to the ventro-median protocerebrum, the second

Fig. 5 Locust standard brain obtained by using the VIB protocol. a-cSurface reconstruction of the resulting average label images (left panel)and a direct volume rendering of the corresponding average intensity map(right panel). For better visualization, the midbrain neuropil (MN) isshown as being transparent (ACa accessory calyx, AL antennal lobe,AOTu anterior optic tubercle, CX central complex, DRMe dorsal rim areaof the medulla, LH lateral horn, Lox lobula complex, MB mushroombody, Me medulla, No noduli, PCa primary calyx). a Anterior view. b

Horizontal view. c Posterior view. d, e Direct volume rendering of theresulting average label images. d Anterior view. e Posterior view. Forvisualization of brain neuropils, the midbrain neuropil has been omitted. fThe probability map for the labeled structures indicates variations inposition and shape. The color code indicates the probability to find asingle structure at a particular position (red 100%, dark blue 0%). Scalebars 600 μm (a-c); 400 μm in d (also applies to e), f

136 Cell Tissue Res (2008) 333:125–145

Page 13

fiber projects anteriorly and joins the inner antenno-cerebraltract toward the deutocerebrum. Beaded terminals areconcentrated in the antennal mechanosensory and motorcenter (arrowheads in Fig. 11a,d,i).

To quantify the influence of registration on the morphol-ogy of the neuron, reconstructions before and after registra-tion were superimposed (Fig. 11b). The reconstructionsshowed 60 branching points and 121 segments with a totallength of 6680.96 μm for the original neuron and6279.16 μm for the registered neuron. Distances weremeasured between all corresponding points of the recon-structions (13.454 vertices; Fig. 11c). These distances variedbetween 10.21 μm and 105.46 μm with a mean distance of

50.24 μm and a median of 42.97 μm. Furthermore, wecompared the differences in length of the segments (n=121)before and after registration. Differences in segment lengthvaried from 0.015 μm to 66.31 μm with a mean of 6.91 μm.The relative proportion of differences in length is 0.1% of thetotal length of the neuron reconstruction. The relativefrequency of differences smaller than 10 μm is 76.03%,and of these, 20.66% are smaller than 1 μm. Only twosegments differ by more than 30 μm (-66.31 μm, 42.45 μm).

Discussion

To provide a digital basis and common framework for futureanatomical analysis, we have generated, using differentprocedures, two standardized 3D brain atlases of the locustSchistocerca gregaria. Both atlases are based on the same setof ten individual male brains immunostained as whole-mounts. For quantification, 34 neuropil structures weresegmented and incorporated into the atlas with the aid ofconfocal microscopic images. For standardization, we usedtwo different methods: the ISA procedure and the VIBprotocol. A comparison of these methods revealed that theVIB and ISA standards are suitable for different purposes.As an example for future application of the ISA atlas, weregistered a lobula projection neuron into the locust ISAstandard brain.

Immunostaining

Confocal microscopy is an established imaging method forinvestigating 3D structures. For the imaging of the largewhole-mount locust brains, however, penetration of anti-bodies and fluorescent compounds was a major problem.We developed a whole-mount protocol that resulted indistinct and reproducible neuropil staining throughout thelocust brain. This was achieved by treatment with enzy-matic digestion by collagenase-dispase and incubation inprimary antibody for up to 6 days. Moreover, treatment

Fig. 6 Frontal slices at the centerof the brain through the evolvingstandard brain as the ISA proce-dure progresses. a After affinealignment to the template brain.b-e Successively refined nonrigidregistrations to the respectivepreceding average image.f Average label fieldcorresponding to the CLSMimage in e. Colors indicate thevarious neuropils. Bar 400 μm(a; also applies to b-f)

Fig. 7 Evolution of groupwise label field overlap through the iterativestandard brain (ISA) generation procedure. Upper curve Volume-weighted overlap, i.e., larger neuropil structures contribute more to theoverlap measure, which is a result of all pixels receiving equal weightindependent of their label. Lower curve Equal weighting of neuropilstructures, i.e., the contribution of all structures is normalized relativeto the others. The equally weighted overlap numbers are generally at alower level, because smaller structures are harder to align, andmisalignment at their surface results in relatively larger overlapreduction (relative to the structure’s volume)

Cell Tissue Res (2008) 333:125–145 137

Page 14

with various temperatures and the addition of Triton X-100and sodium azide led to improved staining results. The useof the 10× oil objective made imaging possible up to adepth of about 600 μm, when scanning the brain from bothsides. To visualize the immunostaining, the brains had to bedehydrated and cleared in methyl salicylate. Tissue dehy-dration, clearing, and the fixation procedure lead toconsiderable tissue shrinking, as shown by Bucher et al.(2000), but are essential for confocal imaging. Because theimmunostaining procedure for single neurons requiressimilar treatments leading to similar spatial distortions, thismight not be a severe limitation when integrating neuronsfrom different preparations into the ISA standard brain.

Sexual differences in brain anatomy

Despite females having a significantly larger body size, wefound no significant differences in overall brain size betweenmale and female locusts. Although the locust standard brainis based on male brains only, future registration of singleneurons into the standard brain might be obtained from bothfemale and male locusts. However, sex-specific differencesmight still exist for particular brain compartments. Toanalyze this, neuropils of interest can now be segmented in

Fig. 8 Locust standard brainobtained by using the ISAmethod. Left panel Surfacereconstruction. Right panelDirect volume rendering of thecorresponding average grayimage. For better visualization,the midbrain neuropil (MN) isshown as being transparent.a Anterior view. b Horizontalview. c Posterior view (ACaaccessory calyx, AL antennallobe, AOTu anterior optic tuber-cle, CX central complex, DRMedorsal rim area of the medulla,LH lateral horn, Lox lobulacomplex, MB mushroom body,Me medulla, No noduli, PCaprimary calyx). Bar 600 μm(c; also applies to a, b)

Fig. 9 Plot of nonrigid distances between individual brains and betweenindividual brains and the ISA standard brain. Each box-and-whisker plotshows the distribution of distances from one of the individual brains to theremaining nine brains (boxes 25th and 75th percentiles, whiskersminimum and maximum, horizontal black bars medians, Xs distancesto the ISA standard). The rightmost plot (Population) represents thestatistics and averages of the entire set of ten brains. The ISA standardbrain has relatively smaller distances from each of the individual brainsthan the individual brains have from each other

138 Cell Tissue Res (2008) 333:125–145

Page 15

female locusts and can be compared statistically with theirmale counterparts for volumetric differences.

Comparison of the ISA method and VIB protocol

Two different registration methods to establish an insectstandard brain have been used: the VIB protocol (Rein et al.2002; Jenett et al. 2006), and the ISA method based on theprocedure introduced by Rohlfing et al. (2001). We appliedand compared these methods. The VIB protocol distin-guishes itself by its relatively easy application, a benefit fromits integration within the graphical user interface of Amirasoftware. It guides the operator reliably through all necessarysteps, which encompass, in addition to data-processing andstatistical calculations, three registration steps. After initialglobal alignment of the label fields, the segmented neuropilswere aligned by using a local rigid transformation. The last

step is an extrapolation that uses a diffusion algorithm totransform unlabeled neuropil regions. This task allowsstructures in the gaps between labeled neuropils to becometransformed according to the previous transformation of theneighboring labeled structures (Jenett et al. 2006). The VIBprotocol (Jenett et al. 2006) is accessible at http://www.neurofly.de and is supplied with detailed instructions.

The ISA method results in more distinct and less fuzzybrain structures than the VIB protocol (Figs. 5, 8). The maincriterion for a representative standard brain that meets ourpurpose is the property that the deformation occurring whena single brain is fitted into the standard brain is smaller thanthe deformation occurring when mapping two individualbrains. This requirement is met with the ISA method, asshown in Fig. 9. With the ISA method, we obtained anaverage label image with higher resolution. In contrast to theVIB protocol, it used stand-alone tools written by one of the

Fig. 10 Comparison of the standardized label images resulting fromthe VIB protocol (white bars) and the ISA method (black bars) withmeans from the individual reconstructed brains. Deviation of relativevolumes (a) and relative distances of the centers of each segmentedstructure to the centers of the entire label field (b) of the structures ofthe ISA result and of the VIB result from the means of the individual

brains. For abbreviations of the neuropils (x-axis) see Table 1.Frequency distribution of the deviations of relative volumes (c) andrelative distances of the structures (d) of the ISA result and of the VIBresult from the means of the individual brains. The VIB standardperforms better with respect to relative volumes, whereas the ISAstandard shows smaller variance in relative distances

Cell Tissue Res (2008) 333:125–145 139

Page 16

authors (available free of charge for selected computingplatforms from http://flybrain.stanford.edu). Further applica-tion, primarily the registration of single neurons into the ISAstandard, can be performed by using commercially availablesoftware such as the registration tools provided by Amira.

At first glance, the resulting ISA label average imageappears to be more symmetrical and more detailed. Theright and left anterior optic tubercles more closely matcheach other in the ISA standard than in the VIB standard(Figs. 5, 8). For unknown reasons, the ellipsoidally shapedlateral horn (Fig. 3) is lengthened particularly in the ISAstandard brain. The two noduli (Fig. 4c) appear greatlydifferent in size in the VIB standard (Fig. 5c). Afterresampling, which was performed for the VIB protocol,each nodulus is represented by 100–150 voxels only. Ingeneral, small neuropils will be more poorly representedafter transformation, because smaller structures are moredifficult to align because of the low resolution of scansobtained with the 10× objective, and misalignment at theirsurface results in relatively larger overlap reduction(Fig. 7). The registration of the lobula projection neuronprovided in this study was not affected by these problems,since it was based on the use of larger neuropils (Fig. 11).Nevertheless, in future studies, small neuropils that displaypoor matching results should be reconstructed and stan-dardized independently in more detail by means of high-resolution scans. The resulting standardized structure canthen be integrated into the standard brain. The task ofsubdividing the central complex has just begun.

Statistical comparison has revealed the different advan-tages of the ISA and VIB methods. Therefore, each standardmay be useful for a different purpose. Although, the twomethods have been applied at different resolutions, thecomparison is valid since the ratio of the volume of aparticular neuropil to the summed volume of all neuropilsdoes not change significantly in resampled data. The sameapplies to the relative distances, which are based on the centersof gravity that remain largely unchanged in resampled data.

The volumetric consistency is preserved in a standardbrain generated with the VIB protocol. Therefore, it is idealfor inter- and intraspecific comparisons of variability at thelevel of neuropil structures or other organs (McGurk et al.2007). With the aid of the VIB protocol, sex-specificvolumetric differences of neuropils can be analyzed. Sincethe locust standard brain was created exclusively from malebrains, this can serve as counterpart for neuropils segment-ed in female locusts. Moreover, the VIB standard brain willbe a valuable tool to study anatomical differences betweensolitary and gregarious locusts. Anton et al. (2002), forexample, have assumed the antennal lobes to be slightlylarger in solitary than in gregarious locusts, as they haveobserved that more antennal receptor neurons project intothe antennal lobe in solitary locusts.

In comparison with the VIB standard brain, the ISAstandard provides a better representation of relative loca-tions of brain areas. This means that the ISA standard ispredestined to combine neuronal data from differentindividual brains. In addition, it may be used as a templateto standardize immunostaining patterns. The small defor-mations of the registered lobula projection neuron supportthe suitability of this brain atlas. The accuracy ofregistration of the neuron was assessed by determiningdistances between corresponding points and differences insegment length of the reconstructions before and afterregistration. In future studies, it will be interesting to seewhether the degree of deformation depends on thecomplexity of the neuron reconstruction, e.g., on thenumber of branches in comparison with the total length ofthe reconstruction or on the existence of neuropils as areference for the registration. In principle, deformations inregistration depend on two sources of variability: (1)interindividual variability in neuronal morphology and (2)deformations based on the registration algorithm. In order

Fig. 11 Fitting of a lobula projection neuron into the locust standard brain(ISA). a-f The skeleton graph of the neuron derived from a reconstructionof aligned scans at 10× magnification. a Frontal view of the neuronreconstruction registered into the ISA standard brain. The soma of theneuron is located in the optic stalk. The primary neurite runs dorsally andgives rise to wide-field arborizations in the anterior (ALo) and outer(OLo) lobe of the lobula complex. The axon runs toward the brainmidline and gives rise to two major collaterals. One fiber projects withbeaded terminals to the posterior and ventro-median protocerebrum,whereas the second branch joins the inner antenno-cerebral tract andprojects with beaded endings to the antennal mechanosensory and motorcenter (arrowhead). b Frontal view of the neuron reconstruction before(red) and after (blue) registration. To quantify the deformation of theneuron, the reconstructions were rigidly superimposed to determine thedistances between corresponding points (vertices). The boxed area isenlarged in c. c Distances between corresponding points of thereconstruction before and after registration were measured as illustrated.d Sagittal view showing the two axon collaterals in the posterior brain.One collateral runs straight downward to the ventro-median protocere-brum (arrow), whereas the second branch joins the inner antenno-cerebral tract to targets in the antennal mechanosensory and motor center(arrowhead). e Enlarged frontal view of the lobula complex. For a betterview of the lobula arborizations, the anterior lobe of the lobula complexhas been omitted. f Sagittal view. The neuron has two main fields ofarborizations: one in the ventral hemisphere of the inner layer of the OLoand a second one posteriorly in the upper stratum 2 of the ALo (doublearrowhead). g-i Confocal images of the neuron. g Maximum projectionof the aligned and merged confocal image stack. The boxed areas areenlarged in h and i. h, i Stacks of confocal images at about 405–555 μmdepth from the frontal surface of the optic lobe. h Note the arborizations(arrow) in the outer lobe (OLo) of the lobula complex. i The axon givesrise to two major collaterals: one (arrow) projecting to the ventro-medianprotocerebrum and the second (arrowhead) running toward the antennalmechanosensory and motor center (AMMC) of the deutocerebrum (aanterior, AL antennal lobe, d dorsal, DLo dorsal lobe of the lobulacomplex, ILo inner lobe of the lobula complex, m medial). Bars 250 μm(a, d), 200 μm (b, e-i), 10 μm (c)

�

140 Cell Tissue Res (2008) 333:125–145

Page 17

Cell Tissue Res (2008) 333:125–145 141

Page 18

to distinguish between these sources of variability, identicalneurons from different preparations need to be recon-structed and registered into the standard brain, as has beenpossible for olfactory projection neurons in Drosophila(Jefferis et al. 2007).

The standard brain is available to the scientific commu-nity, and other laboratories are encouraged to integrate theirneuronal data into this common framework. The lobulaprojection neuron registered into the ISA standard is relatedto a group of motion-sensitive lobula-protocerebrum neu-rons (termed LP1-LP16) characterized in the migratorylocust Locusta migratoria (Gewecke and Hou 1992, 1993;Stern and Gewecke 1993). Unlike that shown for the LPneurons in L. migratoria, the 3D analysis of the projectionneuron in S. gregaria demonstrates dendritic ramificationsin two subunits of the lobula complex and axonalprojections to the deep median protocerebrum and mecha-nosensory and motor center of the deutocerebrum. Neuronsof this type might, therefore, be involved in visually guidedantennal movements, which have been studied in severalinsect species (Erber and Schildberger 1980; Honegger1981; Erber et al. 1997; Kwon et al. 2004).

The registration of the lobula projection neuron is astarting point for the registration of a growing collection ofneuronal cell types into the standard brain of the locust.Since the standard brain is publicly available, neuronalmorphologies obtained in different laboratories can now beintegrated and co-registered within the same framework. Asthe number of registered neurons increase, neuronal wiringdiagrams can be constructed with proposed synapticconnectivities at increasingly finer detail. Through theseefforts, topographic relationships that could not be uncov-ered otherwise will become obvious in particular brainareas, as demonstrated for olfactory projection neurons inthe lateral horn of the fly brain (Jefferis et al. 2007). Finally,the digital atlas of neuronal cell types will facilitate avariety of statistical and cross-correlation analyses andmight allow a better understanding of spiking patternsbased on the branching patterns and arborization domainsof neurons.

The main difference between the ISA and the VIBstandard brain is that the VIB technique is designed tovisualize anatomical variability between the input brains,whereas the purpose of the ISA method is the opposite, i.e.,to remove this variability. The ISA method, however, doesnot completely remove variability but, instead, moves itfrom the image domain to the domain of coordinatecorrespondences between the images. Population variabilityis, therefore, not lost and can, for example, be analyzed bymeans of a statistical deformation model (Rueckert et al.2003) in which the major modes of shape variation withinthe population are decoupled and extracted from thedeformation fields by using principal component analysis.

Comparison with honeybee and fruit fly standard brains

The standard brain of Drosophila melanogaster wasgenerated from 28 brains, whereas for the honeybee, 20individual brains were used. For the comparatively largelocust brain, the immunostaining procedure and thereconstruction took disproportionately more time than itdid for the two other species. The ten brains used toestablish the locust standard brain were a well-balancedchoice taking into account limited time resources and themaximum possible quality.

Unlike the case with the honeybee and the locust, thetransformations used for the Drosophila standard brainwere restricted to isotropic scaling. The resulting standardbrain was represented by an individual brain that wasdefined as the label image closest to the average labelimage. As an enhancement, the VIB protocol introducesnonrigid transformations. Like the ISA method, it iscapable of handling multiple channels and working ondifferent platforms. For the task of analyzing biologicalvariability of gene expression patterns, as intended forDrosophila, the VIB protocol meets all requirements(Schindelin 2005; Jenett et al. 2006).

The ISA method, as applied here, is a standard braingeneration technique that is exactly complementary to thatdescribed by Rohlfing et al. (2001). Whereas Rohlfing et al.(2001) performed registration on the label fields and thenapplied the resulting transformations to the CLSM images,we registered the CLSM images and applied the resultingtransformation to the label fields. Interestingly, Brandt et al.(2005) observed that, although the label field overlapincreased monotonically, the resulting average intensityimages gained sharpness only through the first twoiterations but then turned fuzzy again. In our case, boththe label overlaps and the sharpness of the intensity imagesincreased monotonically; this illustrates the advantages ofusing intensity images for registration rather than texture-free label fields.

Analogous to Brandt et al. (2005), we verified the “meanshape” property of the ISA standard brain by computing its“distance” from all individual brains and comparing thesedistances with the distances between the individual brains.This reflects the common definition of a mean as the objectthat minimizes the total distance from all the objects in theoriginal set. Unlike Brandt et al. (2005), however, who usedthe total nonrigid deformation magnitude as their distancemetric, we considered the second-order bending energy(Wahba 1990) of registrations between brains. This isnecessary because total deformation as computed by Brandtet al. (2005) is affected by scale differences, and the locustbrains in this study have relatively larger scale differencesthan the bee brains. The bending energy, on the other hand,is zero for affine transformations, i.e., it is particularly

142 Cell Tissue Res (2008) 333:125–145

Page 19

insensitive to scale differences between the brains. Similarto Brandt et al. (2005), the standard brain generated by theISA procedure has relatively smaller distances from each ofthe individual brains than the individual brains have fromeach other (Fig. 9). The results provide sufficient evidencethat the locust standard brain achieves its goal of greatlyreducing the deformation required to map an individualbrain onto it, compared with the deformation required tomap one individual onto another.

Concluding remarks

We have calculated a standardized brain of the desert locust byusing two different methods. A comparison of these methodshas revealed that the VIB and ISA standards are suitable fordifferent purposes. Because of the volumetric consistency, theVIB standard is ideal for inter- and intraspecific comparativeanalysis of interindividual variability on the level of theneuropils. The ISA standard, on the other hand, shows largerinvariance in relative distances. The ISA standard is, therefore,predestined to combine neuronal data from different individualbrains. The availability of a locust 3D standard brain atlas is animportant prerequisite to visualize the complexity of neuronalnetworks in this species and to accomplish comparativestudies. Like the existing standard brains of Drosophilamelanogaster and Apis mellifera, the locust atlases will serveas a useful tool to combine and consolidate existing data andto embed new findings. The locust standard brains andinstructions on how data can be contributed are availablefrom http://www.3d-insectbrain.com.

Acknowledgements We thank Dr. Erich Buchner (UniversitätWürzburg, Würzburg, Germany) for providing antibodies, Dr. RobertBrandt (Mercury Computer Systems, Berlin, Germany) for providingthe AmiraZIB license and for help with Amira (especially concerningthe use of the “Arithmetic” module), and Ulrike Träger for providingconfocal images of the stained lobula projection neuron.

References

Anton S, Hansson BS (1996) Antennal lobe interneurons in the desertlocust Schistocerca gregaria (Forskal): processing of aggregationpheromones in adult males and females. J Comp Neurol 370:85–96

Anton S, Ignell R, Hansson BS (2002) Developmental changes in thestructure and function of the central olfactory system in gregariousand solitary desert locusts. Microsc Res Tech 56:281–291

Berg BG, Galizia CG, Brandt R, Mustaparta H (2002) Digital atlases ofthe antennal lobe in two species of tobacco budworm moths, theoriental Helicoverpa assulta (male) and the american Heliothisvirescens (male and female). J Comp Neurol 446:123–134

Boyan G, Reichert H, Hirth F (2003) Commissure formation in theembryonic insect brain. Arthropod Struct Dev 32:61–77

Brandt R, Rohlfing T, Rybak J, Krofczik S, Maye A, Westerhoff M,Hege HC, Menzel R (2005) Three-dimensional average-shapeatlas of the honeybee brain and its applications. J Comp Neurol492:1–19

Bucher D, Scholz M, Stetter M, Obermayer K, Pflüger HJ (2000)Correction methods for three-dimensional reconstructions fromconfocal images. I. Tissue shrinking and axial scaling. J NeurosciMethods 100:135–143

Burrows M (1996) The neurobiology of an insect brain. OxfordUniversity Press, Oxford

Cassenaer S, Laurent G (2007) Hebbian STDP in mushroom bodiesfacilitates the synchronous flow of olfactory information inlocusts. Nature 448:709–714

Chiang AS, Liu YC, Chiu SL, Hu SH, Huang CY, Hsieh CH (2001)Three-dimensional mapping of brain neuropils in the cockroach,Diploptera punctata. J Comp Neurol 440:1–11

Clements AN, May TE (1974) Studies on locust neuromuscularphysiology in relation to glutamic acid. J Exp Biol 60:673–705

Crum WR, Camara O, Rueckert D, Bhatia KK, Jenkinson M, Hill DL(2005) Generalised overlap measures for assessment of pairwiseand groupwise image registration and segmentation. In: DuncanJS, Gerig G (eds) Medical image computing and computer-assisted intervention, MICCAI 8th International Conference,Palm Springs, Calif., USA. Proceedings, Part I, vol 3749 ofLecture Notes in Computer Science. Springer, Berlin HeidelbergNew York, pp 99–106

Davis RL (1993) Mushroom bodies and Drosophila learning. Neuron11:1–14

Elphick MR, Williams L, O’Shea M (1996) New features of the locustoptic lobe: evidence of a role for nitric oxide in insect vision. JExp Biol 199:2395–2407

Erber J, Schildberger K (1980) Conditioning of an antennal reflex tovisual stimuli in bees (Apis mellifera L.). J Comp Physiol135:217–225

Erber J, Pribbenow B, Grandy K, Kierzek S (1997) Tactile motorlearning in the antennal system of the honeybee (Apis melliferaL.). J Comp Physiol [A] 181:355–365

Evers JF, Schmitt S, Sibila M, Duch C (2005) Progress in functionalneuroanatomy: precise automatic geometric reconstruction ofneuronal morphology from confocal image stacks. J Neuro-physiol 93:2331–2342

Farris SM, Robinson GE, Fahrbach SE (2001) Experience- and age-related outgrowth of intrinsic neurons in the mushroom bodies ofthe adult worker honeybee. J Neurosci 21:6395–6404

Galizia CG, Mellwrath SL, Menzel R (1999) A digital three-dimensional atlas of the honeybee antennal lobe based on opticalsections acquired by confocal microscopy. Cell Tissue Res295:383–394

Gewecke M, Hou T (1992) Structure and function of visual interneuronsin the locust brain. In: Singh RN (ed) Nervous systems, principles ofdesign and function. Wiley, New Delhi, pp 255–270

Gewecke M, Hou T (1993) Visual brain neurons in Locustamigratoria. In: Wiese K, Kapitsky S, Renninger G (eds) Sensorysystems of arthropods. Birkhäuser, Basel, pp 119–144

Gouranton J (1964) Contribution a l`étude de la structure desganglions cérébröides de Locusta migratoria migratorioides.Bull Soc Zool France 89:785–797

Greiner B, Gadenne C, Anton S (2004) Three-dimensional antennallobe atlas of the male moth, Agrotis ipsilon: a tool to studystructure-function correlation. J Comp Neurol 475:205–210

Heinze S, Homberg U (2007) Maplike representation of celestial E-vector orientations in the brain of an insect. Science 315:995–997

Heisenberg M (1998) What do the mushroom bodies do for the insectbrain? Learn Mem 5:1–10

Homberg U (1994) Distribution of neurotransmitters in the insectbrain. Progress in Zoology, vol 40. Fischer, Stuttgart

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

Homberg U, Hofer S, Pfeiffer K, Gebhardt S (2003) Organizationand neural connections of the anterior optic tubercle in the brain

Cell Tissue Res (2008) 333:125–145 143

Page 20

of the locust, Schistocerca gregaria. J Comp Neurol 462:415–430

Honegger HW (1981) A preliminary note of a new optomotorresponse in crickets: antennal tracking of moving targets. JComp Physiol 142:419–421

Huetteroth W, Schachtner J (2005) Standard three-dimensionalglomeruli of the Manduca sexta antennal lobe: a tool to studyboth developmental and adult neuronal plasticity. Cell Tissue Res319:513–524

Iyengar BG, Chou CJ, Sharma A, Atwood HL (2006) Modularneuropile organization in the Drosophila larval brain facilitatesidentification and mapping of central neurons. J Comp Neurol499:583–602

Jefferis GSXE, Potter CJ, Chan AM, Marin EC, Rohlfing T, MaurerCR Jr, Luo L (2007) Comprehensive maps of Drosophila higherolfactory centers: spatially segregated fruit and pheromonerepresentation. Cell 128:1187–1203

Jenett A, Schindelin JE, Heisenberg M (2006) The Virtual InsectBrain protocol: creating and comparing standardized neuroanat-omy. BMC Bioinformatics 7:544

Jortner RA, Farivar SS, Laurent G (2007) A simple connectivityscheme for sparse coding in an olfactory system. J Neurosci27:1659–1669

Kinoshita M, Pfeiffer K, Homberg U (2007) Spectral properties ofidentified polarized-light sensitive interneurons in the brain of thedesert locust Schistocerca gregaria. J Exp Biol 210:1350–1361

Kwon H-W, Lent DD, Strausfeld NJ (2004) Spatial learning in therestrained American cockroach Periplaneta americana. J ExpBiol 207:377–383

Laissue PP, Reiter C, Hiesinger PR, Halter S, Fischbach KF, StockerRF (1999) Three-dimensional reconstruction of the antennal lobein Drosophila melanogaster. J Comp Neurol 405:543–552

Laurent G (1996) Dynamical representation of odors by oscillatingand evolving neural assemblies. Trends Neurosci 19:489–496

Laurent G (2002) Olfactory network dynamics and the coding ofmultidimensional signals. Nature Rev Neurosci 3:884–895

Liu G, Seiler H, Wen A, Zars T, Ito K, Wolf R, Heisenberg M, Liu L(2006) Distinct memory traces for two visual features in theDrosophila brain. Nature 439:551–556

Lozano VC, Armengaud C, Gauthier M (2001) Memory impairmentinduced by cholinergic antagonists injected into the mushroombodies of the honeybee. J Comp Physiol [A] 187:249–254

Ludwig P, Williams L, Nässel DR, Reichert H, Boyan G (2001)Primary commissure pioneer neurons in the brain of thegrasshopper Schistocerca gregaria: development, ultrastructure,and neuropeptide expression. J Comp Neurol 430:118–130

Malun D, Plath N, Giurfa M, Moseleit AD, Müller U (2002)Hydroxyurea-induced partial mushroom body ablation in thehoneybee Apis mellifera: volumetric analysis and quantitativeprotein determination. J Neurobiol 50:31–44

Martone ME, Gupta A, Ellisman MH (2004) e-Neuroscience:challenges and triumphs in integrating distributed data frommolecules to brains. Nat Neurosci 7:467–472

Masante-Roca I, Gadenne C, Anton S (2005) Three-dimensionalantennal lobe atlas of male and female moths, Lobesia botrana(Lepidoptera: Tortricidae) and glomerular representation of plantvolatiles in females. J Exp Biol 208:1147–1159

McGurk L, Morrison H, Keegan LP, Sharpe J, O’Connell MA (2007)Three-dimensional imaging of Drosophila melanogaster. PLoSONE 2:e834

Müller M, Homberg U, Kühn A (1997) Neuroarchitecture of the lowerdivision of the central body in the brain of the locust(Schistocerca gregaria). Cell Tissue Res 288:159–176

Müller U (1999) Second messenger pathways in the honeybee brain:immunohistochemistry of protein kinase A and protein kinase C.Microsc Res Tech 45:165–173

Pfeiffer K, Homberg U (2007) Coding of azimuthal directions viatime-compensated combination of celestial time cues. Curr Biol17:960–965

Pfeiffer K, Kinoshita M, Homberg U (2005) Polarization-sensitive andlight-sensitive neurons in two parallel pathways passing throughthe anterior optic tubercle in the locust brain. J Neurophysiol94:3903–3915

Rein K, Zöckler M, Mader MT, Grübel C, Heisenberg M (2002) TheDrosophila standard brain. Curr Biol 12:227–231

Reischig T, Stengl M (2002) Optic lobe commissures in a three-dimensional brain model of the cockroach Leucophaea maderae:a search for the circadian coupling pathways. J Comp Neurol443:388–400

Rind FC (1987) Non-directional, movement sensitive neurones of thelocust optic lobe. J Comp Physiol [A] 161:477–494

Rind FC (2002) Motion detectors in the locust visual system: frombiology to robot sensors. Microsc Res Tech 56:256–269

Rohlfing T, Maurer CR Jr (2003) Nonrigid image registration inshared-memory multiprocessor environments with application tobrains, breasts, and bees. IEEE Trans Inf Technol B 7:16–25

Rohlfing T, Brandt R, Maurer CR Jr, Menzel R (2001) Bee brains, B-splines and computational democracy: generating an averageshape atlas. Proceedings of the IEEE Workshop on MathematicalMethods in Biomedical Image Analysis, MMBIA, Kauai,Hawaii, pp 187–194 (http://doi.ieeecomputersociety.org/10.1109/MMBIA.2001.991733)

Rø H, Müller D, Mustaparta H (2007) Anatomical organization ofantennal lobe projection neurons in the moth Heliothis virescens.J Comp Neurol 500:658–675

Rospars JP, Hildebrand JG (2000) Sexually dimorphic and isomorphicglomeruli in the antennal lobe of the sphinx moth Manducasexta. Chem Senses 25:119–129

Rueckert D, Sonoda LI, Hayes C, Hill DLG, Leach MO, Hawkes DJ(1999) Nonrigid registration using free-form deformations: appli-cation to breast MR images. IEEE Trans Med Imaging 18:712–721

Rueckert D, Frangi AF, Schnabel JA (2003) Automatic construction of3-D statistical deformation models of the brain using nonrigidregistration. IEEE Trans Med Imaging 22:1014–1025

Rybak J, Menzel R (1998) Integrative properties of the Pe1 neuron, aunique mushroom body output neuron. Learn Mem 5:133–145

Sadek MM, Hansson BS, Rospars JP, Anton S (2002) Glomerularrepresentation of plant volatiles and sex pheromone componentsin the antennal lobe of the female Spodoptera littoralis. J ExpBiol 205:1363–1376

Schindelin J (2005) The standard brain of Drosophila melanogasterand its automatic segmentation. PhD thesis, University ofWürzburg, Germany

Schmitt S, Evers JF, Duch C, Scholz M, Obermayer K (2004) Newmethods for the computer-assisted 3-D reconstruction of neuronsfrom confocal image stacks. Neuroimage 23:1283–1298

Simmons PJ (2002) Signal processing in a simple visual system: the locustocellar system and its synapses. Microsc Res Tech 56:270–280

Smid HM, Bleeker MAK, Loon JJA van, Vet LEM (2003) Three-dimensional organization of the glomeruli in the antennal lobe ofthe parasitoid wasps Cotesia glomerata and C. rubecula. CellTissue Res 312:237–248

SternM, GeweckeM (1993) Spatial sensitivity profiles of motion sensitiveneurons in the locust brain. In: Wiese K, Kapitsky S, Renninger G(eds) Sensory systems of arthropods. Birkhäuser, Basel, pp 184–195

Strausfeld NJ (1976) Atlas of an insect brain. Springer, BerlinStrauss R (2002) The central complex and the genetic dissection of

locomotor behaviour. Curr Opin Neurobiol 12:633–638Strauss R, Heisenberg M (1993) A higher control center of locomotor

behavior in the Drosophila brain. J Neurosci 13:1852–1861Toga AW (2002) Neuroimage databases: the good, the bad and the

ugly. Nat Rev Neurosci 3:302–308

144 Cell Tissue Res (2008) 333:125–145

Page 21

Toga AW (2005) Computational biology for vizualization of brainstructure. Anat Embryol 210:422–438

Toga AW, Thompson PM (2001) Maps of the brain. Anat Rec265:37–53

Van Essen DC (2002) Windows on the brain: the emerging role ofatlases and databases in neuroscience. Curr Opin Neurobiol12:574–579

Veelaert D, Schoofs L, De Loof A (1998) Peptidergic control of thecorpus cardiacum–corpora allata complex of locusts. Int RevCytol 182:249–302

Wahba G (1990) Spline models for observational data. CBMS-NSFRegional Conference Series, vol 59. SIAM, Philadelphia

Zöckler M, Rein K, Brandt R, Stalling D, Hege HC (2001) Creatingvirtual insect brains with Amira. ZIB Report 01-32:1–11

Cell Tissue Res (2008) 333:125–145 145