The insect central complex and the neural basis of navigational strategies

Abstract

Oriented behaviour is present in almost all animals, indicating that it is an ancient feature that has emerged from animal brains hundreds of millions of years ago. Although many complex navigation strategies have been described, each strategy can be broken down into a series of elementary navigational decisions. In each moment in time, an animal has to compare its current heading with its desired direction and compensate for any mismatch by producing a steering response either to the right or to the left. Different from reflex-driven movements, target-directed navigation is not only initiated in response to sensory input, but also takes into account previous experience and motivational state. Once a series of elementary decisions are chained together to form one of many coherent navigation strategies, the animal can pursue a navigational target, e.g. a food source, a nest entrance or a constant flight direction during migrations. Insects show a great variety of complex navigation behaviours and, owing to their small brains, the pursuit of the neural circuits controlling navigation has made substantial progress over the last years. A brain region as ancient as insects themselves, called the central complex, has emerged as the likely navigation centre of the brain. Research across many species has shown that the central complex contains the circuitry that might comprise the neural substrate of elementary navigational decisions. Although this region is also involved in a wide range of other functions, we hypothesize in this Review that its role in mediating the animal's next move during target-directed behaviour is its ancestral function, around which other functions have been layered over the course of evolution.

Keywords: Insect brain; Motor control; Navigation; Neuroanatomy; Sensory integration.

Figure

Fig. 1. Anatomy of the central complex (CX).

(A) Location of the CX (colour) in the insect brain (sweatbee Megalopta genalis, from Stone et al., 2017). Image source: www.insectbraindb.org. (B) The CX is conserved across a wide range of insects. Data from el Jundi et al. (2009) (locust), Wei et al. (2010) (cockroach), Stone et al. (2017) (sweat bee), Immonen et al. (2017) (dung beetle), de Vries et al. (2017) (Bogong moth), Jenett et al. (2012) (Drosophila). Images from www.insectbraindb.org for all species except cockroach and Drosophila. (C) Input pathways to the different CX-components. Input to the PB, EB, and noduli (NO) is shown on the left side, while FB-input is shown on the right. (D) Columnar neurons form highly stereotypical intrinsic connections between the PB and the FB/EB. Note that different cell types form projection patterns that are shifted with respect to one another, so that identical PB-columns are mapped to different FB/EB-columns. Based on locusts and Monarch butterfly (Heinze and Homberg, 2008; Heinze et al., 2013).

Fig. 2. Insect navigation behaviours.

(A) Ball-rolling dung beetles use celestial cues for choosing a random travel direction. Left: illustration of behaviour; right: rolling directions are random with respect to the Sun (data from Byrne et al., 2003). (B) During a circular dance on the dung ball these beetles take a snapshot of the celestial cue constellation to decide on a rolling direction. Beetle images in A/B courtesy of B. el Jundi and M. Dacke. (C) The Monarch butterfly and the Bogong moth migratory routes. Both insects fly over 1000 km from their breeding grounds to the wintering/summering sites. (D) When navigating by celestial cues, migrating insects need to compensate for the time of day. (E) A honeybee foraging flight with navigation-relevant visual cues used to compute a vector representation of the hive’s position (path integration). Triangle: hive. Looping pattern at the end of inbound flight: systematic search to locate nest entrance. Flight track adapted from Degen et al. (2016). (F) Skylight polarization is an ambiguous directional cue. Bees communicate their directional knowledge from the outbound flight (top panels) via the waggle dance (bottom panels). Dance distributions are bidirectional when only polarized light cues are available; adapted from Evangelista et al. (2014). (G) Image matching, as performed by ants in rich visual environments (Collett et al., 2013), is used for following habitual routes. A series of snapshots of the visual panorama are compared to current views and both are brought to a best match via body rotations (Zeil, 2012). (H) Landmark based navigation uses salient visual features independent of the surrounding panorama in a similar way to overall image matching. G/H adapted from Heinze, 2017.

Fig. 3. Compass encoding in the insect CX.

(A) Brain of tropical bee Megalopta genalis. Highlighted are neuropils involved in compass encoding (CX, lateral complex, AOTU) as well as the pathway for processing polarized skylight (data from Zeller et al. 2015, Pfeiffer and Kinoshita, 2012). (B-E) Brains with highlighted navigation relevant regions for dung beetle (B), Monarch butterfly (C), Bogong moth (D), and desert locust (E). Data from Immonen et al. (2017), Heinze et al. (2013), de Vries et al. (2017), el Jundi et al. (2009). Images from www.insectbraindb.org. (F) Neurons involved in processing of compass cues in the dung beetle (images courtesy of B. el Jundi). (G) Top: Representation of the angles of polarized skylight in the PB-columns of the locust (Heinze and Homberg, 2007). Bottom: Corresponding mapping of body orientation of Drosophila with respect to surrounding landmarks (Seelig and Jayaraman, 2015). (H) Tracking of heading-directions in Drosophila: Asymmetrical activity in the P-EN neurons (CL2 in other insects) of the right and left PB hemispheres during turning shifts the E-PG-neuron encoded activity bump in the EB, allowing angular integration of body rotations. Arrowheads: directional tunings of E-PG neurons; adapted from Green et al. (2017).

Fig. 4. Convergence of sensory information in the CX as a basis for path integration.

(A) Convergence of speed and direction input in the CX of the bee Megalopta (left). CX-location (green) in the brain is shown alongside polarized-light input pathway (purple arrows) and the likely route for optic flow information (blue arrows). Insert: proposed cellular substrate for convergence for speed and direction inputs are the CPU4 columnar neurons, one of which is depicted (green). At least 18 CPU4-cells exist for each CX-column. (B) Normalized mean activity of three TN2 ‘speed’-neurons in response to different velocities of translational optic flow. Coloured circles, mean activity; grey circles, individual data points; solid lines, background activity ± SD. Figure from Stone et al, (2017), with permission from Elsevier. (C) The four TN-neurons in the Megalopta CX possess four preferred expansion points of translational optic flow (as predicted by local motion tunings), providing the basis for holonomic motion encoding; based on Stone et al. (2017).

Fig. 5. The CX as neural substrate for path integration.

(A) Schematic connections between cell types in the proposed Megalopta path integrator network (Stone et al., 2017). Arrows: excitation; blunt ends: inhibition. Pon, Pontine neurons. (B) Topology of the path integration model circuit. Circles represent neurons, size of neurons indicate activity level, colour-code for model layers and cell groups as in (A). Arrows in the circles: directional tuning (TB1) and integrated direction preference (CPU4); R: right turn; L: left turn; compass rose: current (green) and desired heading (orange). (C) Illustration of the two activity bumps resulting from encoding current heading (green, in PB and EB) and target direction (yellow, in FB). The target direction results from integrating speed and compass information. (D) Steering is induced by comparing PB-activity with FB-activity column by column. The resulting imbalance in CPU1-neuron activity between the right and left side causes steering. (E) Homing behaviour produced by the path integrator model compared to desert ant data (below). Near the nest site (triangle) the model initiates searching behaviour. (F) Comparison of a desert ant (right) and the model negotiating obstacles during the homeward journey. E/F from Stone et al. (2017); inserts in E/F modified from Wehner (2003) (reproduced with permission from Elsevier).

Fig. 6. A common framework for encoding navigational decisions in the insect CX.

(A) Illustration of how compass signals from the PB could be integrated with a speed signal arriving at the noduli to produce a distributed representation of the home vector across FB-columns during path integration, based on the model by Stone et al. (2017). In the model, the home-vector memory resides in CPU4-neurons, which form recurrent connections between PB-columns and the NO. At least 18 individual CPU4-cells per column exist in bumblebees (Stone et al., 2017), suited to form local microcircuits to sustain a lasting, activity-based working memory. CPU4-output in the FB could directly signal the desired heading to steering cells. (B) Similarly, a direction code can be achieved for straight-line orientation in dung beetles (or menotactic orientation in random directions in general) by transferring the current heading signal from the PB to CPU4-neurons. A modulatory signal, e.g. during the dung beetle dance, could trigger the imprinting of the direction code on the CPU4-population. Stability of the memory over the time of the behaviour would require several CPU4-neurons per column. (C) Adjusting synaptic weights between CPU4 and CPU1-cells in a sinusoid manner would allow genetic encoding of migratory directions. Evenly distributed activity across all CPU4-cells (driven by optic flow input) would signal an activity bump to the CPU1-cells according to synaptic weight distribution, guiding the animal towards the migratory heading upon deviation. (D) A similar mechanism could be used for route following based on memorized visual snapshots. If the current view of the animal matches a memorized snapshot, the resulting positive valence output from the memory centres (e.g. mushroom body) could serve as trigger to imprint the current view as temporary desired heading (Collett and Collett, 2018). (E) Based on the path integration model by Stone et al. (2017) we propose a simplified circuit, lacking memory, speed input and the large number of CPU4-cells as possible ancestral circuit, suited to store a copy of the current heading representation in CPU4-neurons for a short amount of time. In case of disturbance, the original heading can be regained via the CPU1 based steering mechanism, using the information stored in CPU4-cells.

Fig. 7. Sensory-motor transformation in the CX as basis for navigational decisions.

(A) Summary of sensory input and possible output pathways of the CX. (B) Detailed information flow within the CX. Several sensory pathways convey different information to different levels of the two main CX-circuits, the FB-circuit (orange) and the EB-circuit (blue). Question marks: proposed input pathways inferred from physiological data lacking anatomical confirmation. Grey arrows: connections verified in Drosophila (Franconville et al., 2018). Dashed grey arrows: proposed connections. Coloured arrows: information flow. (C) Left: A flip-flop LAL-neuron in the moth Bombyx mori responds to pheromone pulses with inversion of activity. Adapted from Mishima and Kanzaki (1998). Right: Illustration of the zig-zagging plume tracking behaviour initiated by the flip-flop neurons.