Maplike Representation of Celestial E-Vector Orientations in the Brain of an Insect

See allHide authors and affiliations

Science  16 Feb 2007:
Vol. 315, Issue 5814, pp. 995-997
DOI: 10.1126/science.1135531


For many insects, the polarization pattern of the blue sky serves as a compass cue for spatial navigation. E-vector orientations are detected by photoreceptors in a dorsal rim area of the eye. Polarized-light signals from both eyes are finally integrated in the central complex, a brain area consisting of two subunits, the protocerebral bridge and the central body. Here we show that a topographic representation of zenithal E-vector orientations underlies the columnar organization of the protocerebral bridge in a locust. The maplike arrangement is highly suited to signal head orientation under the open sky.

Many animals, including birds, fishes, cephalopods, and arthropods, share the ability to perceive linearly polarized light (1, 2). The plane of polarization (E-vector) varies systematically across the blue sky and depends on the Sun's position. For a variety of insects this pattern has been shown to guide spatial orientation (2). In locusts, polarotactic orientation depends on a specialized part of the compound eye, the dorsal rim area (3), and involves several central processing stages, including the central complex (47). The central complex (CC) is a group of neuropils spanning the midline of the insect brain. Substructures are the protocerebral bridge (PB) and the upper and lower divisions of the central body. An outstanding anatomical feature of the CC is its regular and highly sophisticated internal neuroarchitecture (810). In simplified terms, it consists of stacks of arrays, each composed of a linear arrangement of 16 columns with topographic interhemispheric connections between columns both within and between different arrays. Hypotheses on the functional roles of the CC range from a control center for motor coordination (11) to a recently demonstrated involvement in visual pattern learning and recognition (12). In the locust, several cell types of the CC are sensitive to the orientation of zenithal E-vectors (5), but the correspondence of cell morphology and E-vector tuning has remained obscure. In this study, we have used intracellular recordings combined with dye injections to analyze E-vector tuning in CC neurons of the locust with columnar arborization domains.

Two major classes of polarization-sensitive (POL) neurons were encountered regularly when we recorded from CC neurons: (i) a particular type of tangential neuron of the PB and (ii) several types of columnar neurons. Tangential neurons of the PB, termed here TB1 neurons, have not been described previously in the locust or any other insect. A total number of 18 of these cells were analyzed. Their morphology was revealed by iontophoretic tracer injection, histological processing, and camera lucida reconstruction (13). TB1 neurons provide a connection between a posterior brain region, the posterior optic tubercle, and the PB (Fig. 1, A and B). Each TB1 neuron had two domains of varicose and, thus, putatively presynaptic arborizations confined to a single column in each hemisphere of the PB. When PB columns are numbered as L1 (lateralmost column in the left hemisphere of the PB) to L8 (most medial column in the left hemisphere) and, accordingly, from R1 to R8 in the right hemisphere of the bridge, TB1 neurons with varicose ramifications in columns R1/L8, R2/L7, R4/L5, R5/L4, R6/L3, and R7/L2 were encountered. Varicose processes were always eight columns apart, with processes ipsilaterally in one of the four outer columns and contralaterally in one of the four inner columns (Fig. 1E). The columns neighboring those with varicose processes were free of ramifications, and six to eight other columns were invaded with fine, smooth arborizations. Furthermore, all TB1 neurons had varicose arborizations in the posterior optic tubercle, a brain area connected to a small neuropil in the optic lobe, the accessory medulla (14).

Fig. 1.

Morphological and physiological properties of TB1 neurons. (A and B) Frontal reconstructions of two TB1 neurons; inset shows frontal view of the locust midbrain. Scale bars, 100 μm. (C and D) Circular plots of mean firing rate during E-vector rotations for the neuron in (A) [(C)] and (B) [(D)] (n = 4, bin width 10°; error bars: SD). Solid lines indicate background activity. (E) Wiring scheme of the TB1 neuron system. Each line represents one TB1 neuron in the PB (black squares: columns with varicose arborizations; gray squares: columns with smooth arborizations; white squares: columns without arborizations). Asterisks indicate TB1 neurons shown in (A) and (B). (F) Correlation between location of varicose columns in the PB and E-vector tuning (Φmax) of TB1 neurons (n = 18). Because each neuron has varicose arborizations in single columns in the right and left PB hemisphere, Φmax values were plotted twice for the right hemisphere (open circles) and the left hemisphere (filled circles). Linear regression shows significant correlation (r = 0.81, SD = 35.8°, P < 0.0001, two-tailed t test, y = 24.2x + 77.0). Also shown are 95% confidence bands (13). AL, antennal lobe; CC, central complex; MB, mushroom body; PB, protocerebral bridge; POTu, posterior optic tubercle.

The pattern of varicose arborizations in the PB corresponded to physiological properties of the TB1 neurons. For intracellular recordings, the animals were fixed in the recording setup and stimulated from the zenith with a rotating E-vector. Recordings were obtained from their main neurite in the PB. Each TB1 neuron showed polarization opponency, i.e., E-vector orientations leading to an increase in spiking activity (excitation) were oriented perpendicularly to E-vectors, leading to a decrease in spiking activity (inhibition) (Fig. 1, C and D). The E-vector tuning (E-vector orientation resulting in maximum excitation, Φmax) was determined for each neuron by circular statistics [Rayleigh test (15)]. E-vector tuning of TB1 neurons showed a linear relation to the position of their varicose ramifications in the PB (Fig. 1F). Thereby, a range of Φmax tunings of 182° ± 71° extends through the eight columns of each hemisphere of the PB and, thus, corresponds to the whole range of possibly occurring E-vector orientations.

Because the E-vector map in the PB corresponds with the proposed output regions of the TB1 neurons, we asked whether candidate post-synaptic neurons show a similar representation of E-vector tuning. Columnar neurons have arborizations in single columns of the PB and send axonal projections to an area outside the CC, the lateral accessory lobe (LAL). Several cell types have additional arborizations in distinct columns of the central body. Three types of these neurons are polarization-sensitive (Fig. 2). We evaluated the data from 19 recordings from these cell types, termed CPU1, CP1, and CP2 neurons. CPU1 neurons have smooth endings in single columns of the PB, in columns of the dorsalmost layer I of the upper division of the central body (CBU), and an axonal fiber with varicose endings in the LAL (Fig. 2A). Each neuron connects a single column of the PB with two neighboring columns of the CBU, following a wiring scheme described for Drosophila (9) (Fig. 2G). All CPU1 neurons (10 recordings) showed polarization opponency and background spiking activity of 20 to 40 Hz. Comparison of the innervated columns in the PB and Φmax tuning again revealed a spatial representation of E-vector orientations across the PB, which covered a range of 228° ± 73° through the 8 PB columns in one brain hemisphere (Fig. 2I). The slope of the regression lines of TB1 and CPU1 neurons did not differ significantly [P = 0.32, analysis of covariance (16)], indicating that the tuning range through the 16 columns of the bridge matched for both cell types. However, the CPU1 map was shifted by 101° (equivalent to –79°) relativetothe mapof TB1 neurons (significantly different elevation of regression lines, P < 0.0001).

Fig. 2.

Morphology and physiology of columnar neurons. (A to C) Camera lucida drawings of a CPU1 neuron (A), CP2 neuron (B), and CP1 neuron (C) projected onto three-dimensional reconstructions of the central complex. Scalebars, 100 μm. (D to F) Circular plots of mean firing rate during E-vector rotations for the neurons shown in (A) to (C) (n = 4, bin width 10°; error bars: SD). (G and H) Wiring schemes of the CPU1 neuron system and the CP2 neuron system. Asterisks indicate the CPU1 neuron shown in (A) [(G)] and the CP2 neuron shown in (B) [(H)]. The CP1 neuron system (not shown) projects to the MO but is otherwise identical to the CP2 system. (I and J) Linear correlation between the location of the columnar arborization domain in the PB and the E-vector tuning (Φmax) of columnar neurons. (I) CPU1 neurons (r = 0.93, SD = 36.5°, P = 0.0001, two-tailed t test, y = 30.4x – 76.2). Midline crossing occurs at 182°. (J) CP1 neurons (open circles) and CP2 neurons (filled circles) have been combined for statistical analysis (r = 0.97, SD = 38.0°, P < 0.0001, two-tailed t test, y = 27.4 x – 61.8). Midline crossing occurs at 171°. Confidence bands are shown at 95%. CBL, CBU, lower and upper divisions of the central body; LAL, lateral accessory lobe; LT, lateral triangle; MO, median olive; PB, protocerebral bridge.

The remaining two types of columnar neurons, CP1 and CP2 neurons, also connected the PB with the LAL (Fig. 2, B, C, and H), but lacked arborizations in the central body. Within the LAL, projections were confined to either of two subcompartments, the median olive (CP1) or the lateral triangle (CP2). Both types of neuron had smooth arborizations in the PB and varicose endings in the LAL. In all recordings (n = 9), CP1 and CP2 neurons showed polarization opponency, but had lower background activity (3 to 15 Hz) than CPU1 and TB1 neurons. As for CPU1 and TB1 neurons, regression analysis of pooled data from CP1 and CP2 neurons revealed a linear correlation between the innervated column in the PB and the E-vector tuning of the neurons (Fig. 2J). The linear regression for CP1 and CP2 covered a range of 206° ± 76° over one hemisphere (eight columns) of the PB and was not significantly different from the linear regression for CPU1 neurons (slope: P = 0.4777, elevation: P = 0.3525). It was, however, phase-shifted by 111° (equivalent to –69°) against the map of TB1 neurons.

The present study shows, independently for three different cell types, that a map of zenithal E-vector orientations underlies the columnar organization of the locust PB. This spatial representation adds a level of complexity to sensory processing in the insect brain, hitherto thought to be achieved only in vertebrates. In most neurons at least half-maximal activation occurred over an E-vector orientation range of about 60°, implying considerable overlap for neighboring columns and the necessity of a population code for retrieving exact information from the firing rates of these cells (17).

Under the open sky, the activity in the POL neurons described here is directly related to the directional orientation of the locust's head, provided that additional mechanisms (color coding, intensity coding) allow the animal to distinguish the solar from the antisolar hemisphere of the sky. Color-coding properties suited to fulfill this requirement have been demonstrated recently for POL neurons at an input stage to the CC, the anterior optic tubercle (18). Neurons encoding head direction have been intensely studied in mammals (19, 20). An important difference of the locust polarization analyzers in the CC is their global nature of signaling. The neurons fire according to zenithal E-vectors, which provide information about the Sun's azimuth and, therefore, these neurons behave like an internal 180° compass. In rats, in contrast, head-direction cells are recalibrated to visual landmarks in each new environment and are apparently not arranged topographically (21, 22).

Although some progress has been made in analyzing the wiring principles from which computational maps arise in vertebrates (23, 24), the simple brains of locusts offer the opportunity to address this question at the level of single identified neurons. Within the PB, the spatial maps of tangential and columnar neurons are out of phase by about 90°. If TB1 neurons are directly connected to the columnar neurons, a phase shift of 90° might most easily result from inhibitory connections of TB1 neurons to the dendritic trees of columnar neurons. Double-labeling experiments showed that serotonin and a Dipallatostatin–related neuropeptide are present in TB1 neurons (fig. S5). For both substances, inhibitory effects have been demonstrated in other systems (25, 26). Whereas azimuthal space is linearly represented in the columns of the PB, the wiring of CPU1 neurons should result in a superposition of the two 180° representations of the PB in the CBU, but with a lateral shift of about 50° (fig. S3). The functional consequences of this shift are presently unknown. If the E-vector tunings of corresponding CPU1 neurons in each of the eight double columns of the CBU are merged, the azimuth representation would be reduced to only about 156° of frontal space.

On the basis of the data presented here and a recent paper by Liu et al. (12), a coherent functional role for the CC is emerging. In Drosophila, tangential neurons innervating specific layers of the central body are essential for recognizing features of visual objects (elevation in the panorama; contour orientation). Columnar neurons, like the CPU1 neurons, are ideal candidates to associate these visual features with information on their azimuthal direction (fig. S4). Liu et al. (12) already hypothesized that the width of the CBU represents azimuthal space. This is strongly supported by our data, but this representation may differ between cell types and CC substructures as pointed out above.

Supporting Online Material

Material and Methods

Figs. S1 to S5


References and Notes

View Abstract

Navigate This Article