Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
THE JOURNAL OF COMPARATIVE NEUROLOGY 424:356 –376 (2000) Organization and Significance of Neurons That Detect Change of Visual Depth in the Hawk Moth Manduca sexta MARTINA WICKLEIN AND NICHOLAS J. STRAUSFELD* Arizona Research Laboratories Division of Neurobiology, University of Arizona, Tucson, Arizona 85721 ABSTRACT Visual stimuli representing looming or receding objects can be decomposed into four parameters: change in luminance; increase or decrease of area; increase or decrease of object perimeter length; and motion of the object’s perimeter or edge. This paper describes intracellular recordings from visual neurons in the optic lobes of Manduca sexta that are selectively activated by certain of these parameters. Two classes of wide-field neurons have been identified that respond selectively to looming and receding stimuli. Class 1 cells respond to parameters of the image other than motion stimuli. They discriminate an approaching or receding disc from an outwardly or inwardly rotating spiral, being activated only by the disc and not by the spiral. Class 2 neurons respond to moving edges. They respond both to movement of the spiral and to an approaching or receding disc. These two classes are further subdivided into neurons that are excited by image expansion (looming) and are inhibited by image contraction (antilooming). Class 2 neurons also respond to horizontal and vertical movement of gratings over the retina. Stimulating class 1 and 2 neurons with white discs against a dark background results in the same activation as stimulation with dark discs against a white background, demonstrating that changes in luminance play no role in the detection of looming or antilooming. The present results show that the two types of looming-sensitive neurons in M. sexta use different mechanisms to detect the approach or retreat of an object. It is proposed that cardinal parameters for this are change of perimeter length detected by class 1 neurons and expansion or contraction visual flow fields detected by class 2 neurons. These two classes also differ with respect to their polarity, the former comprising centripetal cells from the optic lobes to the midbrain, the latter comprising centrifugal neurons from the midbrain to the optic lobes. The significance of these arrangements with respect to hovering flight is discussed. J. Comp. Neurol. 424:356 –376, 2000. © 2000 Wiley-Liss, Inc. Indexing terms: looming neurons; insect vision; neuroanatomy; electrophysiology; hovering flight Perception of depth is a key feature of vision. It is used to detect and avoid objects; to pursue visual targets; to maintain distance from targets; and to discriminate among objects in the third dimension. Animals employ many strategies to compute depth perception. These include stereopsis, in which images from each retina are combined centrally to provide representation of depth by computing binocular retinal disparities (Julesz, 1972; Livingston and Hubel, 1988). Vergence is another binocular mechanism that uses the degree of rotation of each eye, measured by eye muscle spindles, to compute absolute depth of a perceived target (Judge, 1991). Vergence could, in principle, be used by arthropods with movable eyes on stalks, although it appears that distance perception by triangulation is the method employed (Zeil et al., 1986). © 2000 WILEY-LISS, INC. Insect eyes are integrated in the head cuticle and are virtually immobile. In almost all insect species, the distance between the two eyes is not sufficient for stereopsis Grant sponsor: NIH National Center for Research Resources; Grant number: RR08688; Grant sponsor: A. v. Humboldt Foundation (Feodor Lynen stipend); Grant sponsor: University of Arizona’s NSF Integrative Graduate Education and Research Traineeship Program; Grant sponsor: NSF Plant-Insect Interactions Group. Martina Wicklein’s current address is: the Salk Institute, 10010 Torrey Pines Rd., La Jolla, CA 92037-1099. *Correspondence to: Nicholas J. Strausfeld, Arizona Research Laboratories Division of Neurobiology, 611 Gould-Simpson, University of Arizona, Tucson, AZ 85721. E-mail: [email protected] Received 2 December 1999; Revised 25 April 2000; Accepted 27 April 2000 LOOMING-SENSITIVE NEURONS IN MANDUCA Fig. 1. Differences between stimuli used for this study. A: The retinal image of an approaching (or receding) disc expands (or contracts) over the retina: the area of the image increases; the perimeter length of the image increases; and the edge of the expanding image provides motion stimuli across ommatidia. B: A spiral that is fixed at one plane subtends an unchanging area of the retina. However, when it turns, the edges of the spiral move across ommatidia, outward or inward depending on the direction of rotation. This stimulus provides motion stimuli only, without changing perimeter length or image area. or vergence other than in odonate larvae and in the preying mantis, which use true stereopsis (Baldus, 1926; Rossel, 1986). A special case of binocular vision is used by male calliphorid flies, which possess an area of high retinal acuity that provides information about the centered image of a pursued target. This is usually a female fly, the immediate proximity of which is detected when its image falls on both retinas (Collett and Land, 1975). Both vertebrates and arthropods also employ monocular mechanisms for depth perception. Not yet demonstrated in insects, but known from human psychophysics, is depth perception by occlusion where a depth map is created by objects that are close to the moving eye appearing to pass in front of more distant objects, blocking them and relatively larger areas of the background from view. Distance perception by motion parallax has been identified in insects, particularly in locusts (Collett, 1978; Erikson, 1980) and in honey bees (Srinivasan et al., 1989). This type of depth perception relies on a comparison of velocities and motion directions at different planes in the visual scene. It should be noted that depth perception by occlusion or by motion parallax requires information about self-motion and about background. These strategies contrast with yet another mechanism for detecting change in depth: that of looming. This is the perception of an approaching object, the image of which expands (or contracts) across the retinal surface. Looming is also an interesting case of depth perception because it involves neither background features nor self-motion of 357 Fig. 2. Histogram shows four examples of depth-sensitive neurons (mean responses of five repeated stimulations). Both class 1 and 2 looming-sensitive neurons respond to an expanding disc with an increase of spike frequency (preferred direction) and a contracting (receding) disc with a decrease of spontaneous spike activity (null direction). The class 1 and 2 antilooming-sensitive neurons respond in the opposite manner: to a contracting disc with an increase of spike frequency (preferred direction) and to an expanding (approaching) disc with a decrease of spontaneous spike activity (null direction). The symbols used here and in other figures are as follows: a wedge pointing to the left indicates expansion; a wedge pointing to the right indicates contraction. the observer. The only requirement is that the object has to contrast against a background. On its own, looming cannot be used to define a three-dimensional (3D) map of a scene. However, the perception of a looming stimulus initiates escape behavior by locusts and flies and is used by nectar feeding moths and hummingbirds for maintaining station in front of a target. It is thus not surprising that looming-sensitive neurons can be found in insects and vertebrates. Examples are in sphingid moths (Wicklein, unpublished data), grasshoppers (Rind, 1999; Gabianni et al., 1999), flies (Holmqvist and Srinivasan, 1991), pigeons (Sun and Frost, 1999), and primates (Tanaka and Saito, 1989; Graziano et al., 1994)). This account addresses the question of what features of the looming stimulus are detected by uniquely identifiable neurons and how such neurons might collaborate to provide information that would regulate the flight motor to maintain distance from a target. The paper describes experiments performed on the tobacco hawk moth Manduca sexta—a crepuscular and nocturnal sphingid. Sphingids are rapid and highly maneuverable flyers with large compound eyes, which hover in front of flowers where they perform stationary flight while ingesting nectar. Typically, a moth will approach a flower at high velocity and then abruptly decelerate to arrive at a distance from the flower that just permits insertion of the proboscis. To feed successfully, the moth maintains its position even if the flower moves, irrespective of the direction of movement. Flowers selected as targets are those that offer stark contrast in front of uniform foliage, which under these light conditions comprises a virtually unstructured background. 358 M. WICKLEIN AND N.J. STRAUSFELD MATERIALS AND METHODS Animals Manduca sexta (Lepidoptera: Sphingidae) were reared on an artificial diet (modified after Bell and Joachim 1976) and maintained on a long-day photoperiod (17-hour light, 7-hour dark) at 25°C at 50 – 60% relative humidity. Stage 18 pupae were taken out of the colony and put into an environmental chamber to hatch. Their light cycle was reversed in the chamber, and animals were kept under reversed conditions for at least a full cycle to ensure that they switched to a diurnal rhythm. Electrophysiology: preparation Adult 1- to 3-day-old M. sexta were cold anesthetized and waxed to a holder to stabilize the thorax ventrally. The head was tilted forwards at a 30° angle from the horizontal axis and waxed in place, leaving the eyes uncovered. The proboscis was cut to prevent muscle action, and the cut was sealed with low melting point wax. A narrow window was cut into either the left or right posterior surface of the head capsule, directly above the optic peduncle. Tracheal air sacs and the neural sheath were carefully removed to expose the optic peduncle and the most medial part of the optic lobes. The preparation was perfused with a Ringer’s solution to prevent desiccation (Christensen and Hildebrand, 1987). The head of the moth was aligned in the rig in a standard fashion using the antennal bases and the base of the proboscis as fiducial points that were oriented to markers on the table. The eyes’ pseudopupils were used to orient the head so that it assumed a standard position with respect to the stimuli, or the monitor, in front of it. To achieve this alignment the pseudopupils were revealed by illumination through an operation microscope, and the head was moved until the pseudopupils were in the most medial and dorsal corner of each eye. These were then aligned with vertical and horizontal markers on the recording table. Intracellular recording and staining Intracellular recordings were made from neurons in the optic lobes. Intracellular recording and dye injection were performed using silicon glass microelectrodes filled with a 1% neurobiotin-KCl solution backed up with 1 M KCl. The electrodes had tip resistances of between 60 and 90 M⍀ (when in contact with the tissue). A silver wire inserted into the thorax served as the indifferent electrode. Recordings were amplified (Axoprobe 1A, Axon Instruments, Burlingame, CA) and displayed using standard equipment and stored (Video Recorder, Vetter, Rebersburg, PA). After recording, the cells were filled iontophoretically for 5–10 minutes using a 1-nA negative current. Signals were analyzed off-line on a desktop computer using a data analysis system (Spike2, CED, Cambridge, UK). The response frequency was calculated by subtracting the spontaneous activity of the cell from the spike frequency during stimulation. An estimate of spontaneous activity was obtained by calculating the means from the spike rate measured in a 2-second period before and after each set of motion stimuli. Stimuli Both physical objects (light or dark discs, gratings, rotating spirals) and their simulations were used as stimuli. All stimuli subtended a minimum of 30° on the retina. Stimuli used to test looming responses were a white disc (5 cm in diameter) in front of a black background or a black disc in front of a white background. The disc was moved with a constant velocity of 1 m/sec toward the animal, stopped 10 cm in front of the moth where it subtended an angle of 30° on the retina, and then moved back to its starting position (30 cm away from the moth). Spirals (10 cm in diameter) of different degrees of curvature, and presented at a standard distance from the eye, were rotated using a small motor that could be reversed (to provide apparent expansion or contraction). The spiral could be rotated at a range of velocities. In each experiment, the spiral was rotated with constant velocity, stopped, and then rotated in the opposite direction with the same velocity. Spiral stimuli were presented to the front part of the visual field. To compare the effects of different stimuli, all the descriptions below refer to stimulation of this frontal field of view. The stimulus used to test horizontal and vertical motion sensitivity consisted of a frame holding a movable grating. Gratings were moved vertically and horizontally in different parts of the visual field of the animal to test the response of a cell to translatory motion. Stimuli were also simulated: displayed on a fast computer monitor (NEC M500, 160-Hz refresh rate) and presented to the animal via a surface-glazed mirror positioned to stimulate the frontal visual field (extending 70° in the horizontal plane and 50° in the vertical plane). These stimuli consisted of moving horizontal and vertical gratings; expanding and contracting light and dark discs; inwardly and outwardly rotating spirals. The starting and end size of the discs and the velocity with which they expanded and retreated could be preprogrammed, as could the size and curvature of the spiral and its velocity of rotation. Stationary stimuli were presented first, then moved in one direction, stopped, and moved in the opposite direction. This cycle was repeated five times for each stimulus. The order in which stimulus types were presented was randomized. Histology The procedure for revealing neurobiotin was based on the original method of Horikawa and Armstrong (1988), using a protocol for marine Crustacea (Schmidt and Ache 1990) or modifications for Diptera (Gronenberg and Strausfeld, 1992). After iontophoresis (5–10 minutes 1nA) and 1 hour for tracer diffusion (at 4°C), the brain was opened and immersed in 4% formaldehyde in 0.2 M Sorensen’s phosphate buffer. Half an hour later the brain Fig. 3. Reponses of class 1 looming cells to expanding discs and spirals. Histograms (A,B) show relative spike frequencies elicited by a disc moving toward and away from the animal (wedges), a spiral turning outward and inward (indicated by arrowed spirals), horizontal progressive and regressive movements of gratings (horizontal arrows), and up-to-down or down-to-up motion of gratings (vertical arrows). The histograms show that both recorded neurons had the same properties: strict tuning to disc expansion, inhibition to disc contraction, and no or negligible responses to directional motion. C–J: Spike trains supporting these conclusions. C,D: Excitation to disc expansion, inhibition to disc contraction. E,F: No responses to apparent looming (spiral). G–I: Movement of gratings shows no effect on background discharge. Figure 3 360 M. WICKLEIN AND N.J. STRAUSFELD was dissected and fixed overnight. After washing in buffer for at least 8 hours, the tissue was dehydrated through an ethanol series, rinsed in propylene oxide (10 minutes), and then rehydrated. The purpose of this step is to permeabilize the tissue for a whole-mount avidin-biocytin procedure (Gronenberg and Strausfeld, 1992). The brain was then incubated overnight in buffer with avidin-horseradishperoxidase (Vectastain ABC-kit, Vector, Burlingame, CA) and 1% Triton X-100. After washing in buffer for 8 hours, the brain was presoaked for 4 hours in the dark at 4°C in diaminobenzidine (DAB) solution (Vectastain DAB-kit, Vector). Another buffer wash followed, and the tissue was then transferred into the same DAB solution containing 0.001% hydrogen peroxide (Vectastain, DAB-kit, Vector) and processed in the dark at 4°C for 2 hours. After a 1-hour rinse in buffer in the dark at room temperature, the brain was processed for an additional hour in the dark at room temperature. The tissue was rinsed in buffer overnight and then incubated in osmium tetroxide for 2–12 hours in the dark at 4°C. The tissue was then dehydrated and embedded in Durcupan (Fluka, Heidelberg, Germany). Serial horizontal or frontal sections were cut at 35 m. Preparations were observed and photographed with a Leitz Orthoplan microscope. Drawings and reconstructions were made using a camera lucida attachment. Selected areas were serially digitized at 2-m steps using a Sony DC 5000 CCD and overlaid using Adobe Photoshop. Single images were collapsed into one image using the Photoshop darkening function. Reconstructions have been normalized to show one-half of the brain, as seen from the back or top, with the optic lobes to the left of the illustration. Schematics showing the organization of the insect visual system are provided in Strausfeld and Lee (1991). RESULTS More than 200 neurons were recorded from 148 moths. Most neurons were motion sensitive but not looming sensitive. Sixty-seven neurons that responded to looming stimuli were recorded from the optic lobes. Of these, 14 were filled with neurobiotin, with 14 others showing partial fills or multiple fills. The latter were discarded. Looming neurons can be generally grouped into two major classes according to their tuning properties and structure. Class 1 neurons respond to approaching or receding discs whereas class 2 neurons respond both to these stimuli and to rotational motion of a spiral pattern (Fig. 1). Class 1 and 2 neurons can be subdivided into those that are excited by an object moving toward the eye (type 1a, 2a neurons) and those that are excited by an object receding away from the eye (type 1b, 2b neurons; Fig. 2A,B). Of those neurons that were filled with neurobiotin, class 1 neurons were revealed to be centripetal cells leading from the optic lobes to the brain whereas class 2 neurons are centrifugal cells, leading from centers in the posterior midbrain (called the posterior slope) out to the medulla. The recorded neurons responded with action potentials having amplitudes that ranged from 10 to 60 mV. Responses were mostly sustained tonic excitations or were inhibition of background activity (Figs. 3A,B, 4A,B). Most class 2 neurons showed phasic-tonic responses and higher response frequencies at stimulus onset when stimulated with a rotating spiral (Figs. 4A,B, 9 –11, 13). Significantly, none of the neurons showed habituation to repetitive stimulation. Most neurons maintain a background (resting) activity. Thus, suppression of background in response to visual stimulation indicates inhibition in the null direction. Conversely, excitation is an increase of the spike frequency compared with the background. All the neurons tested had large (up to 120° horizontal, 90° vertical) receptive fields in the frontal or frontoventral part of the monocular visual field. Responses to a moving disc or spiral were the same as those to movement simulated on a computer monitor. Encoding of approaching and retreating objects Neurons responded best to objects that move along the longitudinal body axis of the moth toward or away from the eye (i.e., the z-axis). We classified neurons as looming or antilooming sensitive and direction selective. Four examples are shown in Figure 2. Two cells respond with an increase in spike frequency when stimulated with a disc moving toward the eye and a decrease in spike frequency when a disc recedes from the eye (cells 1,2a). These neurons are classified as looming-sensitive neurons. Two other neurons show an increase of spike frequency when stimulated by a disc moving away from the eye and a decrease in spike frequency when the disc moves toward the eye (cells 1, 2b). These cells exemplify antilooming neurons. Different types of stimuli that are associated with change in distance reveal two classes of neurons that respond selectively to specific visual cues (Fig. 1): 1) expanding or contracting light and dark discs in front of a contrasting background provide change in object area, change of object perimeter (edge) length, and edge motion; 2) inward and outward rotation of a spiral isolates edge motion since this stimulus neither changes its area nor its perimeter length; and 3) movement of horizontal and vertical gratings test for directionally selective motion sensitivity along the x and y axes. Responses to these stimuli demonstrate that class 1 and 2 neurons are sensitive to certain, but not all features of an advancing or receding stimulus. Class 1 cells respond only to an approaching or retreating disc but not to rotating spirals or moving gratings. Class 2 cells respond to all these different stimuli. Figures 3 and 4 exemplify responses of two class 1 (Fig. 3) and two class 2 neurons (Fig. 4) elicited by these stimulus elements. The bar graphs in Figure 3A and B summarize the response properties for the two cell classes, each bar representing the mean response of five identical stimulations. The two class 1 neurons are clearly tuned to an expanding disc. Their spike trains (Fig. 3C–H) show actual responses to different stimuli: an approaching (expanding) disc causes excitation (Fig. 3D); a spiral simulating expansion elicits no response Fig. 4. Reponses of class 2 looming cells to expanding discs and spirals. As in Figure 3, histograms A and boxes C, E, G, and I are from one neuron; histogram B and boxes D, F, H, and J are from a second neuron. Both are class 2 elements that are strongly excited by disc expansion, outward spiral rotation, regressive motion, and downto-up motion. The cells are strongly inhibited by the respective opposite directions of motion or apparent motion. Figure 4 362 M. WICKLEIN AND N.J. STRAUSFELD (Fig. 3E,F); and likewise, horizontal (Fig. 3G,H) and vertical movement of gratings (Fig. 3I,J) are ineffective. Class 1 neurons are further divided into two types: the 1a neurons just described are looming neurons, excited by an approaching object and inhibited by a receding one; and the type 1b neurons are antilooming neurons, inhibited by an expanding object and excited by a receding one. Both types of neurons respond feebly, if at all, when presented with moving gratings. Class 1 neurons distinguish expansion of the retinal image from contraction. Contraction and expansion involve three variables: change of the perimeter length (edge length) over the retina; change in the area of the image on the retina; and change of luminance. Change in luminance probably plays no role in looming or antilooming detection, however. A type 1a neuron that is activated by an expansion of a white disc against a dark background is equally activated by an expansion of a dark disc against a pale background (Fig. 5), where the overall luminance decreases. Two cells, the activities of which are shown in Figure 4, exemplify responses of class 2 neurons. The bar graph again summarizes the response properties for each of the two cells; each bar represents the mean response of five repeated stimuli. The spike trains show responses of the neurons to the same stimuli used to test class 1 cells (Fig. 3). Figure 4A and B demonstrates that class 2 cells are sensitive to expanding and receding discs. Here, both cells are type 2a looming neurons, being excited by expansion and inhibited by contraction (Fig. 4B,C). The neurons increase their firing rates to outwardly rotating spirals (expansion) and decrease their firing rates to inwardly rotating spirals (Fig. 4A,B,E,F). Thus, as in humans, the outwardly rotating spiral provides the moth with the illusion of expansion. Again, class 2 neurons are divisible into two types: 2a responding to looming stimuli, and type 2b responding to receding (antilooming) stimuli. Class 2 neurons also respond to horizontal and vertical movement of gratings over the retina (Fig. 4G–J). The preferred directions of the neuron summarized in Figure 4A are front-to-back (regressive horizontal movement; Fig. 4G) and down-to-up vertical movement. The second class 2 neuron (Fig. 4B) preferred downward vertical motion and regressive horizontal motion (Fig. 4I). In each case null directions were indicated by a reduction of the resting frequency. Class 2 neurons were also tested with dark discs expanding or contracting against a pale background. As is the case for class 1 neurons, the cells respond to changes of image size regardless of whether the stimulus is darker or lighter than the background—in other words, independently of luminance (Fig. 5). Anatomical differences of class 1 and 2 neurons The two classes of looming and antilooming neurons differ in their morphology (Figs. 6 –14). Class 1 neurons derive from cell bodies in the optic lobes. Their terminals invade the ipsilateral optic foci of the posterior slope of the midbrain. Their dendrites in the optic lobes are grouped into three distinct arbors: one in the medulla resides exclusively in its innermost stratum; a second in the lobula consists of processes that ascend through the lobula to provide minute twigs within its outer stratum; and a third in the lobula plate penetrates all its strata. Fig. 5. Class 2 type 2 neuron, excited by disc contraction independent of luminance. Contraction of a white disc against a dark background (white columns) excited the cell, as does expansion of a dark disc against a light background (filled columns). Class 2 neurons derive from large cell bodies situated posteriorly in the cell body rind between the calyces of the mushroom bodies. The dendritic trees of class 2 neurons reside in posterior slope neuropil on one or both sides of the brain. Their axons extend out to the medulla on the same or the opposite side of the brain as their cell bodies. Their terminals in the optic lobes are restricted to the outermost medulla stratum, extending across a large area of its retinotopic mosaic. Descriptions of class 1 neurons Figure 6 shows a class 1 neuron filled with neurobiotin and reconstructed from serial sections in the vertical plane (normal to the z-axis). The cell responds with a higher spike frequency to a receding stimulus but no activation when stimulated with an inward rotating spiral, thereby indicating that it is an antilooming neuron. Its dendritic branchlets (Fig. 7A) are arranged distoproximally in the innermost stratum of the medulla and in the outermost stratum of the lobula where they arise from stouter branches that ascend through this neuropil (Fig. 7B). A third level of dendrites branch throughout the depth of the lobula plate (Fig. 7C). Figure 8A and B shows a second class 1 neuron, the dendrites of which are concentrated mainly within the lower half of the medulla and lobula plate, suggesting that it is activated by stimuli subtending the lower half of the visual hemisphere. However, its dendritic field in the lobula extends across the dorsoventral extent of this neuropil. The neuron terminates in the posterior slope of the midbrain. Original spike trains are shown in response to stimulation with an approaching or retreating disc (Fig. 8B) and an outward or inward rotating spiral (Fig. 8C). These are the typical response characteristics of type 1b antilooming neurons, showing selective activation to a receding stimulus but no response to an inwardly rotating spiral. Figure 9A illustrates a species of neuron that can be characterized as a type 1a looming cell, with a sustained response to an expanding image on the retina (Fig. 9B). It is strongly inhibited by a contracting image (Fig. 9B). An expanding spiral may slightly diminish background frequency, although this is ambiguous (Fig. 9C). Dendrites in the medulla LOOMING-SENSITIVE NEURONS IN MANDUCA 363 Fig. 6. Reconstruction and responses of a class 1 type 2 antilooming neuron. A: To show the full organization of this cell, the lobula is shown separated to the right from the medulla and lobula plate. Reconstruction of dendrites showing their distoproximal arrangements at three consecutive levels: in the medulla, lobula, and lobula plate. Main branches ascend through the second chiasma to provide retinotopic branchlets in the inner medulla stratum and through the lobula plate. Branches ascend outward through the depth of the lobula to its outer stratum, where they terminate. The inset shows an enlargement of medulla arborizations. Abbreviations for this and subsequent figures: me, medulla; lo, lobula; lop, lobula plate. B: Recording showing the response of this cell to an expanding and contracting disc and (C) to an outward rotating spiral. Scale bar ⫽ 100 m. serve the lower frontal part of the eye’s receptive field as do dendrites in the lobula plate. Dendrites in the lobula are concentrated in its ventral half and have diffuse arrangements dorsally. An axon projects to neuropil flanking the esophagus. A thin collateral leads off the axon to provide a recurrent field of terminal processes situated at the base of the optic peduncle (Fig. 9A, inset). Its terminal in the posterior slope was incompletely filled. 364 M. WICKLEIN AND N.J. STRAUSFELD Descriptions of class 2 neurons The next four figures illustrate class 2 neurons, all of which have their cell bodies and dendrites in the midbrain, as exemplified by the cell shown in Figure 10. This cell, which has been reconstructed from serial horizontal sections, has its terminal arborization confined to the outermost layer of the medulla. Its dendrites reside in the lateral deutocerebrum but originate from a large cell body situated contralaterally in the dorsal cell body rind. The fine structure of the arborizations in the brain are suggestive of dendrites, indicating that this and other class 2 neurons receive their inputs in the midbrain (see Fig. 12 for details). This neuron showed spontaneous bursts of activity, which were possibly autonomous to this cell type but are more plausibly a consequence of injury and current leakage. The neuron responded to an expanding retinal image and was inhibited by a contracting one. It showed an increase of spike frequency when stimulated with an outward rotating spiral (illusory looming). The initial response to this stimulus was phasic The neurons shown in Figures 11, 13, and 14 were reconstructed from frontal sections. Figure 11 shows a neuron whose dendritic tree and cell body are contralateral to its terminal in the medulla. The latter provides a dense field of beaded processes (Fig. 12A) that extend across the whole outermost stratum of the medulla. The dendrites branch profusely in the posterior slope of the midbrain and some extend distally into the base of the optic peduncle. Dendritic branches are visited by axons originating from the lobula plate (Figs. 11A, inset, 12B,C). The neuron is a looming cell showing an increase in activity in response to an approaching disc or an outwardly rotating spiral. Figure 13 shows a class 2 cell that has two fields of dendrites, one on the same side as its terminal in the medulla, but contralateral to the cell body. Its axon prolongates over the esophageal foramen to enter the posterior slope. Neurobiotin failed to resolve details of the dendritic tree on that side, however. As in the previous examples, a dense system of terminal processes extends across the outer stratum of the medulla. The cell is a looming-sensitive neuron, typical of other type 2a cells, responding to both an expanding disc and outward spiral motion (Fig. 13A,B). The cell is completely silent during stimulation with the nonpreferred direction of movement, namely, the retreating disc or the inwardly rotating spiral. The last example (Fig. 14) shows a type 2a neuron with dendrites in the posterior slope of the midbrain. Its cell body lies just at the other side of the midline, however. The terminal processes in the medulla are restricted to the lower half of the neuropil, in an area corresponding to the lower half of the retina. Figure 14B and C shows that the neuron is excited by an expanding retinal image and is inhibited by a contracting image. An outwardly rotating spiral vigorously excites the cell. DISCUSSION This paper describes neurons in the visual neuropil of Manduca sexta that are tuned to looming stimuli. Two cell classes can be distinguished, based on their anatomy and their visually induced response properties. Class 1 neurons receive inputs in the inner stratum of the medulla, the outer stratum of the lobula and through all strata of the lobula plate. Their cell bodies are in the optic lobes, and their axons end ipsilaterally in the posterior slope of the midbrain. Class 2 cells are centrifugal neurons. They receive their inputs in the posterior slope and terminate in the medulla. Their cell bodies reside posteriorly near the midline of the brain. The two classes of neurons also differ in their physiological responses to visual stimuli. Class 1 cells respond only to an approaching or retreating disc but not to a rotating spiral or moving grating. Class 2 cells respond to both types of stimuli. The different response properties of the neurons suggest that they detect different stimulus properties associated with looming. In nature, tobacco hawk moths feed while they hover in front of a flower that is selected by the moth for its high contrast (Raguso et al., unpublished data). While feeding, moths maintain a constant distance from the flower using visual cues (Wicklein and Willis, unpublished data). Three features of the target could provide information that is used by the visual system to detect change of distance: 1) change in the area of the retinal image; 2) change in length of the image perimeter; and 3) detection of flow field properties, such as expanding or contracting edges. The present results demonstrate that at least two of these mechanisms are realized within one and the same system. Two components of the stimulus can provide information about change in area: a change in luminance detected by pooling the intensity signal across the whole retina; or detection of change of image size on the retina by recruitment or divestment of individually stimulated retinotopic channels. Cells that increase their firing rate to an expanding bright target against a darker background also respond in the same way to an expanding dark target against a bright background. This excludes the possibility that the effective signal is increased luminance because the second stimulus condition is associated with an overall luminance decrease. Do changes in perimeter length of the image provide information about looming? To test this, neurons were stimulated with expanding and contracting discs. Their responses were compared with responses elicited by inwardly and outwardly rotating spirals. A looming object expanding over the retina’s surface not only provides outward moving edges but also increasing area and an increase of perimeter (edge) length. A spiral rotated in one direction provides an outward motion of edges and in the opposite direction an inward motion of edges, in both cases without any change in area or perimeter length. Thus, if perimeter length is a key stimulus for looming detection, an outwardly rotating spiral should not excite a cell that is excited by an expanding disc. Fig. 7. Features of the neuron shown in Figure 6. A: Tufts are restricted to the innermost stratum (is) of the medulla’s inner layer (il). The outer layer (ol) and serpentine layer (sl) are not invaded by this neuron. B: Detail showing branches ascending through the inner medulla strata (is). C: Branches also arborize through the outer (ol) and inner (il) layers of the lobula. The inset (from the boxed area) shows delicate brushlets of processes that invade the outer lobula layer, at the level of terminals of type 1 transmedullary cells from the medulla (Fig. 15). D: A third system of branches invades the lobula plate, extending from its outer surface through all three directionand orientation-specific layers (1–3). p op t; groove in the outer surface of the lobula carrying axons of the posterior optic tract. Scale bars ⫽ 50 m in A; 25 m in B; 10 m in C,D. Figure 7 366 M. WICKLEIN AND N.J. STRAUSFELD Fig. 8. A class 1 type 2 antilooming neuron. A: Localized dendritic fields in the ventral medulla and lobula plate and across the lobula. The axons terminate in the posterior slope (p sl). Other abbreviations: la, lamina; mb, calyx and initial part of pedunculus of the mushroom body; p sl, posterior slope. B: Responses to disc expansion (inhibition) and contraction (excitation) and (C) an unchanged background activity during stimulation with an expanding (or contracting: not shown) spiral. Scale bars ⫽ 100 m. The present results show that only class 1 neurons are activated by expanding or contracting discs but not by the rotating spiral. We conclude that these neurons are tuned to changes of retinal image area or changes of perimeter (edge) length and that luminance and moving edges play no role in their activation. Class 2 neurons respond to both expanding and contracting discs and to rotating spirals. This shows that outward and inward movement of edges are their effective stimuli and that luminance, perimeter length, and changes of retinal image area are not necessary to excite this class of cells. In addition to their responses to spirals, class 2 cells are activated by moving gratings whereas class 1 cells are not. This supports the idea that class 2 cells are tuned to directional motion whereas class 1 cells are not. Responses of class 2 cells to changes of image size are independent of luminance. This fits well with natural conditions. A white flower illuminated by moonlight or starlight will appear pale against a dark background. The same flower may offer a dark profile against a pale background (the sky) at and just after sunset when the moths are already foraging. Anatomical polarities of class 1 and 2 neurons Both class 1 and 2 neurons are conventional nerve cells in that they have two separate fields of arborizations connected by an axon. Arborizations that are interpreted as dendrites can be either smooth or tapering, or can comprise many tapering branchlets equipped with knobs, spines, or varicosities. Structures inter- LOOMING-SENSITIVE NEURONS IN MANDUCA 367 Fig. 9. A class 1 type 1 looming neuron. A: Reconstruction showing that its processes are arranged across the central part of the medulla’s retinotopic mosaic. Processes are distributed across the whole lobula and lobula plate. The terminals arborize extensively in the neuropil, connecting the brain with the subesophageal ganglion. A small collateral from the axon provides a field of blebbed processes (boxed; see inset) within the optic peduncle, suggesting their relationship with axons of other visual interneurons. B: Spike train showing excitation elicited by an expanding disc and inhibition to a contracting disc. C: There are no responses to an expanding spiral. Scale bar ⫽ 100 m in A; 5 m in inset. preted as terminals are composed of arborizations decorated with varicosities or bead-like specializations (Strausfeld and Meinertzhagen, 1998). Reconstructions of class 1 and 2 neurons suggest that the two classes are differently polarized: class 1 neurons are centripetal, carrying information centrally from dendrites in optic lobe neuropils to terminals in the midbrain; class 2 neurons are centrifugal, relaying information from their dendrites in the posterior slope out to their terminals in the medulla. The locations of their cell bodies support these polarities: class 1 neurons have cell bodies in the optic lobes, and class 2 neurons have their cell bodies in the posterior rind of the midbrain. One caveat to these anatomical inferences is that neurons in insects, as in 368 M. WICKLEIN AND N.J. STRAUSFELD Fig. 10. Class 2 type 1 neuron. A: Reconstruction from serial horizontal sections shows the peripheral location of its terminals at the medulla’s outer surface and a densely branched dendritic tree ipsilaterally in the midbrain and its contralateral cell body. ant lob, antennal lobe. Other abbreviations as for previous figures. Although background bursts of spikes may be injury responses, the cell nevertheless showed a clear response to disc expansion (B) and to spiral expansion (C). Scale bar ⫽ 100 m. deuterostomes, can have pre- and postsynaptic structures on the same arbor. We were not able to perform physiological experiments confirming the assumed direction of information flow. Centrifugal cells have been previously described from the butterfly Papilio aegeus (Ibbotson et al., 1991) and from Manduca sexta (Milde, 1993). Ibbotson et al. (1991) found motion-sensitive neurons that connect the midbrain to the medulla. These neurons (termed MV1 and MH1) have cell bodies close to the midline of the brain. Their smooth tapering arborizations are located in the posterior slope, and their wide-field terminals reside in the outer layers of the medulla. These cells correspond structurally to the class 2 neurons described here. In Papilio, MH1 cells are selectively activated by horizontal motion with a preferred and null direction. MV1 neurons respond to vertical motion, also with a preferred and null direction. Interestingly, the MH1 and MV1 cells of Papilio are tuned only to horizontal or to vertical gratings, respectively, whereas the present class 2 cells in Manduca are sensitive to movement in both orientations, with a null and preferred direction for each. Cells termed cMt by Milde (1993) have their cell bodies located dorsally, close to the brain’s midline. Like the class 2 neurons described here, cMt neurons have smooth tapering processes in the posterior slope and beaded arborizations in the outer third of the medulla. Whereas processes of class 2 cells extend through large areas of the medulla’s retinotopic mosaic, the terminals of cMt cells are restricted to small patches of the medulla. The cMt neurons are motion sensitive and some show directional selectivity (Milde, 1993). Neither cMt nor MH1 and MV1 neurons were tested with looming stimuli and thus cannot be further compared with class 2 neurons described here. We assume that all these neurons— class 2 cells, cMt, MH1, and MV1 neurons— belong to a larger category of motion-sensitive elements that are supplied by afferents in the posterior slope and feed information back to the medulla. The nature of their inputs is discussed next. Polarities of looming neurons Observations of dipteran visual systems reveal that the posterior slope of the midbrain is the main termination area for wide-field motion-sensitive collator neurons originating in the lobula plate (Strausfeld and Bassemir, 1985a,b; Strausfeld and Gronenberg, 1990). These direc- LOOMING-SENSITIVE NEURONS IN MANDUCA 369 Fig. 11. A class 2 type 1 neuron reconstructed from vertical sections. A: Its large dendritic tree in the right side of the brain provides an axon that terminates in the lower frontal two-thirds of the contralateral medulla, there providing a dense plexus of beaded processes. Its dendritic tree is subdivided into branchlets (inset, upper right) each providing a claw-like group of processes that are invested with knobs and spicules. These appear to clasp the passage of un- stained axons that project from the right lobula complex (not shown) to the posterior slope. B: This neuron showed a high level of autonomous activity. The firing rate increased in response to an expanding disc but returned to its initial rate in response to a contracting disc. C: Its firing rate increased in response to an outwardly rotating spiral. Scale bar ⫽ 100 m. tionally selective neurons, which are variously tuned to specific orientations of motion, provide inputs to the dendritic trees of premotor descending neurons and are thought to be also presynaptic to a variety of centrifugal and heterolateral cells that extend back out into the optic lobes (Hausen and Egelhaaf, 1989; Strausfeld et al., 1995). The present results suggest that class 2 neurons are comparable to centrifugal neurons in Diptera, like the CH neuron (Hausen and Egelhaaf, 1989). Moreover, observations of biocytin-filled class 2 neurons reveal that their dendrites in the posterior slope are visited by many largediameter profiles from the lobula plate. Given this anatomical relationship, it is not surprising that class 2 neurons are orientation and directionally motion sensitive. They respond to a preferred vertical as well as to a preferred horizontal direction, suggesting that their dendrites receive inputs from two converging popu- lations of lobula plate outputs: horizontal motion-sensitive neurons and vertical motion-sensitive neurons. Recent studies of the diurnal hummingbird hawk moth Macroglossum stellatarum (Wicklein and Varju, 1999) have confirmed that the sphingid lobula plate indeed provides separate populations of vertical and horizontal motionsensitive tangential neurons that are probably homologous to the horizontal and vertical motion-sensitive neurons of many brachyceran Diptera (Buschbeck and Strausfeld, 1997). Organization among the dendrites of class 1 neurons is more difficult to interpret. The entire dendritic tree provides three orderly and spatially segregated systems of branches that supply distal processes to three discrete levels of optic lobe: the inner stratum of the medulla; the outermost stratum of the lobula; and all strata through the lobula plate. Each branching system originates from a 370 Fig. 12. Details of the class 2 neuron shown in Figure 11. A: Terminal processes are equipped with bead-like specializations (arrowed circle) that invade retinotopic columns. B: This panel shows an enlargement of the area boxed to the right in Figure 11, identifying M. WICKLEIN AND N.J. STRAUSFELD dendritic processes in the posterior slope. C: Enlargements showing the clawed appearance of dendritic branchlets (arrows and boxes). D: Enlargement of dendritic tuft boxed to the right in B. Scale bars ⫽ 10 m. LOOMING-SENSITIVE NEURONS IN MANDUCA 371 Fig. 13. Reconstruction of a class 2 type 1 neuron showing its terminal in the outer stratum of the medulla. A: The bilateral dendritic tree is shown only in part with the right component indicated by the dotted line. B: The neuron showed almost no background activity and was excited by disc expansion and outward spiral motion (C). Scale bar ⫽ 100 m. large diameter axon. Branches are ordered: those supplying the medulla are distal to branches supplying the lobula; these are distal to branches supplying the lobula plate. In sphingids, as in brachyceran Diptera, levels receiving class 1 neuron dendrites are specifically associated with small-field retinotopic neurons that contribute to a discrete motion processing pathway (Douglass and Strausfeld, 1995, 1997) that is composed of elementary movement-detecting circuits (EMDs; Buchner, 1976). These circuits are repeated across the optic lobes, each circuit detecting the sequential change of luminance between adjacent visual sampling points (Franceschini et al., 1989). An integral part of each EMD is the type 1 transmedullary neuron (Tm1), which occurs in each retinotopic column. Tm1 neurons, which are common to both flies and sphingid moths (Fig. 15), are modulated by directional motion (Douglass and Strausfeld, 1995). Their collaterals invade the innermost medullary stratum, and their endings terminate in the outermost stratum of the lobula (Fig. 15). Tm1 neurons supply information about movement to the next processing stage, the bushy T-cells (T5), whose dendrites reside in the outer stratum of the lobula, as do dendritic twigs of class 1 neurons. Bushy T-cells terminate in orientation- and direction-specific strata in the lobula plate. Again, class 1 neuron dendrites invade these layers. The coincidence of three levels of class 1 dendrites and the respective levels of Tm1 collaterals, Tm1 terminals, and T5 endings suggests that class 1 neuron dendrites receive inputs from movement-sensitive neurons across the whole eye. Why, though, should this occur thrice, at successive levels in the system? And why should class 1 neurons be “blind” to translatory motion if they are supplied by motion-sensitive elements? 372 M. WICKLEIN AND N.J. STRAUSFELD Fig. 14. A class 2 type 1 neuron with a dendritic tree in the ipsilateral posterior slope, but derived from a contralateral cell body. A: The terminal in the medulla invades only the lower half of the retinotopic mosaic. B: This neuron responded to an expanding disc and (C) to an outwardly expanding spiral. Scale bar ⫽ 100 m. One suggestion is that the relationship between Tm1 neurons and class 1 neuron dendrites in the medulla and outer lobula allows successive temporal sampling of changes of contrast between adjacent ommatidia, as would occur when the image perimeter expands or contracts across the retina. In the medulla and lobula, each dendritic branchlet of a class 1 neuron thus serves as an edge detector. The neurons’ lack of response to direction may be explained by directional information being discarded due to interactions in the lobula plate where class 1 neuron dendrites invade all its directionally sensitive layers. The question of why dendrites of class 1 neurons are distributed successively within the three optic lobe neuropils is more challenging. One possibility is that duplicating inputs from Tm1 at two levels adds redundancy to the system and thus increases accuracy. Alternatively, dendrites staggered at three successive levels may enable the cell to detect temporal and spatial displacements of the image edge from one ommatidium to the next during im- age expansion or contraction. Estimations of change of length of the image edge as it contracts or expands across the retina will be a function of the number of edge detectors triggered by edge motion. This calculation can ill Fig. 15. Camera lucida drawings of small-field retinotopic neurons homologous to those that are known to be involved in elementary motion-detecting circuits in Diptera and that terminate in layers that are invaded by the dendrites of class 1 looming neurons. Small and wide-field lamina monopolar cells (L2s, w) terminate in the outer stratum of the medulla, at the level of terminals of class 2 looming neurons and the dendrites of type 1 small and wide-field transmedullary neurons (Tm1 s,w). These second-order neurons provide narrow or wide-field collaterals within the inner stratum of the medulla and terminals in the outer stratum of the lobula. The dendrites of bushy T-cells (T4; see inset for details, T5) originate at these levels, each sending an axon into motion- and orientation-specific levels in the lobula plate (lop). These layers, like the inner stratum of the medulla and outer stratum of the lobula are invaded by the class 1 neuron dendrites. Scale bar ⫽ 100 m. Figure 15 374 afford noise incurred by synaptic delays, as velocities of image edge expansion/contraction over the retina while hovering are fast, calculated as 0.2– 0.4 m/sec. Possibly, activated edge detectors are pooled first in the medulla, then in the lobula and then in the lobula plate. The organization of class 1 dendrites at these levels could allow comparison of rapid change in perimeter distribution on the retina by integrating edge detection within the dendritic tree at time t1 for inputs to the medulla, t2 for inputs to the lobula, and t3 for inputs to the lobula plate. Why are there looming and antilooming systems? Comparisons with other taxa The present results identify two classes of neurons that detect movement along the z-axis: cells that are insensitive (class 1) or sensitive to motion cues (class 2). Both classes are further divided into two types (1, looming; 2, antilooming) that are excited by expansion or contraction, respectively, of a retinal image. If each of these cell types can alone signal change of depth, what is the advantage of having a system comprising four types of neurons that signal the same event? In locusts and flies a single system of neurons detects looming to provide information about the expansion of an image over the retina. This information triggers an escape jump or a landing response, depending on the context in which the stimulus is encountered. In locusts this behavior is mediated, in part, by the lobula giant motion detector (LGMD), which synapses onto the descending contralateral motion detector (DCMD: Rowell, 1971; Rind, 1984)—itself supplying the hind-leg jump circuit. The LGMD responds to a looming stimulus that may (Gabbiani et al., 1999) or may not (Rind and Simmons, 1999) signal time-to-collision. In flies there is an isomorphic array of several hundred columnar neurons called Col A cells that subserve the whole eye. Each cell is triggered by a darkening stimulus entering its receptive field (Gilbert and Strausfeld, 1991), and the axon of each Col A cell is coupled by gap junctions to dendrites of the giant descending neuron (GDN) that supplies the midleg jump circuit (Strausfeld and Bassemir, 1983). This system reacts to a stimulus threshold defined by perimeter length of the retinal image rather than to time-to-collision (Holmqvist and Srinivasan, 1992). Swimming, flying, or arboreal animals would be expected to possess neuronal circuits that inform motor pathways about impending collision with objects. Such circuits are not exclusive to insects. They have been proposed for crustaceans (Glantz, 1974), and these circuits are likely to have evolved in arthropods before their appearance in chordates. Neurons having similar morphologies to those seen in insect looming systems have been identified in the nucleus rotundus of the pigeon (Luksch et al., 1998) and in layer SGS1 of the squirrel optic tectum (Major, Luksch and Karten, unpublished data). In each case the dendrites of these cells provide broad fan-like arrays of dendritic branches, each providing a small bottlebrush-like tuft onto which centripetal afferents terminate. As in the class 1 looming neurons, the shapes of these elements suggest a precise retinotopic input onto them from quite large areas of the visual field, which is M. WICKLEIN AND N.J. STRAUSFELD also the case for the locust LGMD and the ensemble of Col A cells in flies. There are interesting parallels between the selective properties of wide-field neurons in the looming subdivisions of the nucleus rotundus of pigeons (Wang and Frost, 1992; Wang et al., 1993) and type 1 class 1 and type 1 class 2 neurons in Manduca. In the nucleus rotundus, wide-field neurons are segregated at two discrete levels, the more distal cells receptive to velocity on the x and y axes, the deeper neurons tuned to motion on the z axis (Wang et al., 1993). Like class 1 cells, these neurons are not sensitive to directional motion but do respond to image expansion. Sun and Frost (1998) show three types of looming-sensitive neurons in the nucleus rotundus: one computing the relative rate of expansion; the second the absolute rate of expansion; and a third multiplying size-dependent responses and instantaneous angular velocity of the object in a manner comparable to findings from the locust LGMD (compare Gabbiani et al., 1999). What is so remarkable about these various systems is the apparent evolutionary convergence— both of morphological attributes and of functional properties—suggesting that common selective pressures, such as the necessity of avoiding objects during fast movements in three dimensions (water or air), have forced similar adaptations across phyla. It would be most interesting to study the nucleus rotundus of hummingbirds because, unlike pigeons, hummingbirds use looming and receding stimuli as cues for stationary flight. In this they show an exquisite parallel to sphingid hovering, a behavior that is utterly distinct from the escape and landing responses of locusts that rely solely on looming stimuli. There are similarities between the organization of control systems for stationary and for forward flight. During stationary flight the flight motor continuously makes compensatory fine adjustments to adjust for visual drift—signifying changing position toward or away from the target (Pfaff and Varju, 1991; Farina et al., 1994, 1995; Kern and Varju, 1998). In a comparable manner, compensatory movements stabilize the platform along the pitch, yaw, and roll axes to maintain visual balance during forward flight (Hausen and Egelhaaf, 1989). Although it could be argued that for forward flight one system would suffice to detect progressive and regressive horizontal motion, it seems that this is insufficient for fine control: several orthogonal systems have evolved that provide highly accurate information about all extractable features of the visual flow field (Hausen, 1982; Krapp and Hengstenberg, 1996). The several systems supporting stationary flight are likewise organized as discrete assemblages of nerve cells that collaborate in the detection of movement along the z axis. Even though systems of looming neurons in moths are crudely distinguished as motionindependent and motion-dependent neurons, their four divisions into looming and antilooming neurons described here may not be the minimum required for precise control of stationary flight. In this account we have shown that for any type of looming or antilooming neuron there exist several distinct morphs or species of nerve cells that have different field dimensions across the neuropil. Possibly, many of these morphological types are required to provide the exquisite accuracy of motor control manifested by the moth’s behavior. Com- LOOMING-SENSITIVE NEURONS IN MANDUCA parative studies of insects that show different grades of visually guided behaviors may resolve the question of whether systems of looming and antilooming neurons derive from an evolutionary elaboration of a simpler and phylogenetically basal system, like the locust LGMD. ACKNOWLEDGMENTS We thank Albert Brower, BS, for instructions in imaging and digital montaging, Robert Gomez, BS, for technical help, and A.A. Osman, PhD, for rearing moths. We are grateful to Holly Campbell, PhD, for discussing various aspects of this work. This account has also profited from Harvey Karten’s (MD) valuable insights about avian visual systems. N.J.S. received a grant from the NIH National Center for Research Resources (RR08688), and M.W. received a Feodor Lynen stipend from the A. v. Humboldt Foundation, as well as small grants from the University of Arizona’s NSF Integrative Graduate Education and Research Traineeship Program and NSF PlantInsect Interactions Group. LITERATURE CITED Baldus K. 1926. Experimentelle Untersuchungen über die Entfernungslokalisation der Libellen (Aeschna cyanae) Z Vergl Physiol 3:475– 505. Bell RA, Joachim FA. 1976. Techniques for rearing laboratory colonies for rearing tobacco hornworms and pink bollworms. Ann Ent Soc Am 69:365–373. Buchner E. 1976. Elementary movement detectors in an insect visual system. Biol Cybern 24:85–101. Buschbeck EK, Strausfeld NJ. 1997. The relevance of neural architecture to visual performance: phylogenetic conservation and variation in dipteran visual systems. J Comp Neurol 383:282–304. Christensen TA, Hildebrand JG. 1987. Male-specific, sex pheromonesensitive-selective projection neurons in the antennal lobes of the moth Manduca sexta. J Comp Physiol A 160:553–569. Collett TS. 1978. Peering—a locust behaviour pattern for obtaining motion parallax information. J Exp Biol 76:237–241. Collett TS, Land MF. 1975. Visual control of flight behaviour in the hoverfly, Syritta pipiens L. J Comp Physiol (A) 99:1– 66. Douglass JK, Strausfeld NJ. 1995. Visual motion detection circuits in flies: peripheral motion computation by identified small field retinotopic neurons. J Neurosci 15:5596 –5611. Douglass JK, Strausfeld NJ. 1995. Visual motion detection circuits in flies: parallel direction- and non-direction-sensitive pathways between the medulla and lobula plate J Neurosci 16:4551– 4562. Eriksson ES. 1980. Movement parallax and distance perception in the grasshopper (Phanlacridium vittatum) (Sjostedt) J Exp Biol 86:337– 341. Farina WM, Varju D, Zhou Y. 1994. The regulation of distance to dummy flowers during hovering flight in the hawk moth Macroglossum stellatarum. J Comp Physiol 174:239 –247. Farina WM, Kramer D, Varju D. 1995. The response of the hovering hawk moth Macroglossum stellatarum to translatory pattern motion. J Comp Physiol 176:551–562. Franceschini N, Riehle A, Le Nestour A. 1989. Directionally selective motion detection by insect neurons. In: Stavenga DG, Hardie RC, editors. Facets of vision. Berlin: Springer. p 360 –390. Gabbiani F, Krapp H, Laurent G. 1999. Computation of object approach by a wide-field motion-sensitive neuron. J Neurosci 19:1122–1144. Gilbert C, Strausfeld NJ. 1991. The functional organization of male-specific visual neurons in flies. J Comp Physiol A 169:395– 411. Glantz, RM. 1974. Defense reflex and motion detector responsiveness to approaching targets: the motion detector trigger to the defense reflex pathway. J Comp Physiol (A) 95:297–314. 375 Graziano, MSA, Andersen RA, Snowden RJ. 1994. Tuning of MST neurons to spiral motions. J Neurosci 14:54 – 67. Gronenberg W, Strausfeld NJ. 1992. Premotor descending neurons responding selectively to local visual stimuli in flies. J Comp Neurol 316:87–103. Hausen K. 1982. Motion sensitive interneurons in the optomotor system of the fly. II. The horizontal cells: receptive field organization and response characteristics. Biol Cybern 46:67–79. Hausen K, Egelhaaf M. 1989. Neural mechanisms of visual course control in insects. In: Stavenga DG, Hardie R, editors. Facets of vision. New York: Springer. p 391– 424. Holmqvist MH, Srinivasan MV. 1991. A visually evoked escape response of the housefly. J Comp Physiol (A) 169:451– 459. Horikawa K, Armstrong WE. 1988. A versatile means of intracellular labeling: injection of biocytin and its detection with avidin conjugates. J Neurosci Methods 25:1–11. Ibbotson MR, Maddess T, Du Bois R. 1991. A system of insect neurons sensitive to horizontal and vertical image motion connects the medulla with the midbrain. J Comp Physiol (A) 169:355–367. Judge SJ. 1991. Vergence. In: Carpenter RHS, editor. Eye movements, vol 8. Vision and visual dysfunction (Cronly-Dillon JR, series editor). Boca Raton, FL; CRC Press. p 157–174. Julesz B. 1972. Cyclopean perception and neurophysiology. Invest Ophthalmol 11:540 –548. Kern R, Varju D. 1998. Visual position stabilization in the hummingbird hawk moth, Macroglossum stellatarum L. I. Behavioral analysis. J Comp Physiol 128:225–237. Lee DN, Davies MNO, Green PR, Van der Weel FR. 1993. Visual control of velocity of approach by pigeons when landing. J Exp Biol 180: 85–104. Krapp H, Hengstenberg R. 1996. Estimation of self-motion by optic flow processing in single visual interneurons. Nature 384:463– 466. Livingstone M, Hubel D. 1988. Segregation of form, color, movement, and depth: anatomy, physiology, and perception. Science 240:740 – 749. Luksch H, Cox K, Karten HJ. 1998. Bottlebrush dendritic endings and large dendritic fields: motion-detecting neurons in the tectofugal pathway. J Comp Neurol 396:399 – 414. Milde JJ. 1993. Tangential medulla neurons in the moth Manduca sexta. Structure and responses to optomotor stimuli. J Comp Physiol 173: 783–799. Pfaff M, Varju D. 1991. Mechanisms of visual distance perception in the hawk moth Macroglossum stellatarum. Zool Jb Physiol 95:315– 321. Rind FC. 1984. A chemical synapse between two motion detecting neurones in the locust brain. J Exp Biol 110:143–167. Rind FC, Simmons PJ . 1999. Seeing what is coming: building collisionsensitive neurones. Trends Neurosci 22:215–220. Rossel S. 1986. Binocular spatial localization in the praying mantis. J Exp Biol 120:265–281. Rowell CHF. 1971. The orthopteran descending movement detector (DMD) neurones: a characterisation and review. Z Vgl Physiol 73:167–194. Schmidt M, Ache BW. 1990. Afferent projections to the midbrain of the spiny revealed by biocytin. Soc Neurosci Abstr 16:400. Srinivasan MV, Lehrer M, Zhang SW, Horridge GA. 1989. Honeybees measure their distance from objects of unknown size. J Comp Physiol (A) 165:605– 614. Strausfeld NJ, Bassemir UK. 1983. Cobalt-coupled neurons of a giant fibre system in Diptera. J Neurocytol 12:971–991. Strausfeld NJ, Bassemir UK. 1985a. Lobula plate and ocellar interneurons converge onto a cluster of descending neurons leading to neck and leg neuropil in Calliphora erythrocephala. Cell Tissue Res 240: 617– 640. Strausfeld NJ, Bassemir UK. 1985b. The organisation of giant horizontalmotion-sensitive neurons and their synaptic relationships in the lateral deutocerebrum of Calliphora erythrocephala and Musca domestica. Cell Tissue Res 242:531–550. Strausfeld NJ, Gronenberg W. 1990. Descending neurons supplying the neck and flight motor of Diptera: organization and neuroanatomical relationships with visual pathways. J Comp Neurol 302: 954 –972. Strausfeld NJ, Lee J-K. 1991. Neural basis for parallel visual processing in the fly. Vis Neurosci 7:13–33. Strausfeld NJ, Meinertzhagen IA. 1998. The insect neuron: types, morphologies, fine structure, and relationship to the architectonics of the 376 insect nervous system. In: Harrison FW, Locke M, editors. Microscopic anatomy of invertebrates. Insects, vol 11B. New York: Wiley-Liss. p 487–538. Strausfeld NJ, Kong A, Milde JJ, Gilbert C, Ramaiah L. 1995. Oculomotor control in calliphorid flies: GABAergic organization in heterolateral inhibitory pathways. J Comp Neurol 361:298 –320. Sun H, Frost BJ. 1998. Computation of different optical variables of looming objects in pigeon nucleus rotundus neurons. Nature Neurosci 1:296 –303. Tanaka K, Saito H. 1989. Analysis of motion of the visual field by direction, expansion/contraction, and rotation cells clustered in the dorsal part of M. WICKLEIN AND N.J. STRAUSFELD the medial superior temporal area of the macaque monkey. J Neurophysiol 62:626 – 641. Wang YC, Frost BJ. 1992. Time to collision is signaled by neurons in the nucleus rotundus of pigeons. Nature 19:236 –238. Wang YC , Jiang S, Frost BJ. 1993. Visual processing in pigeon nucleus rotundus: luminance, color, motion, and looming subdivisions. Vis Neurosci 10:21–30. Wicklein M, Varju D. 1999. The visual system of the European hummingbird hawkmoth Macroglossum stellatarum (Sphingidae, Lepidoptera). Motion sensitive Interneurons of the lobula plate. J Comp Neurol 408:272–282. Zeil J, Nalbach G, Nalbach H-O. 1986. Eyes, eye stalks and the visual world of semi-terrestrial crabs. J Comp Physiol (A) 159:801– 811.