Active touch in orthopteroid insects: behaviours, multisensory substrates and evolution

Abstract

Orthopteroid insects (cockroaches, crickets, locusts and related species) allow examination of active sensory processing in a comparative framework. Some orthopteroids possess long, mobile antennae endowed with many chemo- and mechanoreceptors. When the antennae are touched, an animal's response depends upon the identity of the stimulus. For example, contact with a predator may lead to escape, but contact with a conspecific may usually not. Active touch of an approaching object influences the likelihood that a discrimination of identity will be made. Using cockroaches, we have identified specific descending mechanosensory interneurons that trigger antennal-mediated escape. Crucial sensory input to these cells comes from chordotonal organs within the antennal base. However, information from other receptors on the base or the long antennal flagellum allows active touch to modulate escape probability based on stimulus identity. This is conveyed, at least to some extent, by textural information. Guidance of the antennae in active exploration depends on visual information. Some of the visual interneurons and the motor neurons necessary for visuomotor control have been identified. Comparisons across Orthoptera suggest an evolutionary model where subtle changes in the architecture of interneurons, and of sensorimotor control loops, may explain differing levels of vision–touch interaction in the active guidance of behaviour.

1. Introduction

Active touch most typically has been studied in mammalian systems. The majority of these studies have been conducted on primate palpation or manipulation (often clinically [1], but sometimes under more natural conditions [2]) and on rodents in the act of whisking [3]. However, there have been enough recent studies of other touch-sensory systems that it is possible to suggest a truly comparative picture of mammalian active touch is emerging. For example, whisking by insectivores (e.g. Etruscan shrews) [4,5], tactile searching and foveation of prey items in specialized insectivores (e.g. star-nosed moles) ([6]; and see [7]), and hydrodynamic trail following using mystacial vibrissae in pinnipeds (e.g. California sea lions; [8] and see [9]) have all been analysed mechanistically. In some of these mammalian examples, the ability to relate behaviour to central neural circuitry is quite extensive (e.g. [10]).

In order to understand active touch at the most general neurobehavioural level, it would be important to analyse examples obtained from higher animals that have different details of neural organization. Arthropods are advanced animals that have sometimes provided models that can be related to mammalian neurobehavioural issues. For example, spiders have inspired some experiments on spatial learning in mammals [11] and analysis of visual sensory processing in insects has provided abundant parallels with the complexities of processing in vertebrates (e.g. [12,13] for moving stimuli and [14,15] for configural stimuli).

In this paper, we will focus on insect model systems that are beginning to provide data on active touch. Our goal is to highlight how ‘touch’ in insects can take some slightly different forms than it does in mammals, and that touch systems often are part of a larger multisensory network. Our own work has used orthopteroid insects and we will concentrate on this group. The term ‘orthopteroid’ is an old one in insect taxonomy and it loosely refers to several orders that include crickets, grasshoppers, locusts, stick insects, praying mantises and cockroaches (see Tree of Life; http://tolweb.org/Neoptera/8267). The term is useful in that it roughly bounds a group from which much behavioural and neurobiological data have been mined ([16]; and see [17] for examples from stick insects).

First, we will outline the general methods used in our studies on multisensory guidance of behaviour and of active touch. Additional details are contained in the electronic supplementary materials. Next, we will briefly review the organization of the cercal wind-mediated escape system of orthopteroids. It is a convenient nucleus for this paper. From work on cockroaches, we know that this classic wind-mediated escape circuit is embedded in larger networks of visual and touch-sensory processing (figure 1 summarizes the form of the wind-mediated escape circuit in relation to the touch and visual pathways as a diagrammatic framework for the discussion in this paper).

Figure 1.

Schematic of the classic cercal, wind-sensory escape pathway (black arrows) in relation to the more recently described pathways for antennal touch control of escape (red arrows) and for visual guidance of the antennae during active touch (blue arrows). The diagram was originally developed from information on cockroach, but it may apply to cricket as well. Pathways where neurons have been individually identified are represented by solid lines; dashed line and open arrow are pathways inferred from behavioural studies. Am, antennal motor neurons; DMIs, descending mechanosensory interneurons; DVIs, descending visual interneurons; GIs, giant interneurons; Lm, leg motor neurons; M, mechanosensory pathway from basal antennal receptors; M′, hypothesized pathway from antennal flagellum. See text for details.

Extending analysis of escape behaviour from its elicitation by wind to its multisensory guidance led us to an appreciation of the role of active touch in this critical behaviour, predator detection and avoidance. We thus will move quickly from wind-evoked escape to the tactile sensory circuits related to the antennae, and then to the active use of the antennae for tactile exploration under visual guidance. Most of the data reviewed—and some new data—will be from cockroaches and crickets, but relevant data from other orthopteroids will be included. We will conclude with an evolutionary perspective that is made possible by the comparative information on escape, vision and antennal sensory function that is available at the cellular level in orthopteroid insects. The summary also will consider some broad analogies with the mammalian whisking system.

2. Approach and methods

Our studies of wind-evoked escape in cockroaches [18], and of visual guidance of the antennae, and touch-evoked escape [19] have followed a consistent methodology. First, we studied the behaviour of cockroaches interacting under unrestrained conditions with various predators (spiders, mantids, toads and mice) and sometimes with conspecifics. More recently, we have begun parallel studies on crickets and locusts [20,21] to provide a comparative picture of orthopteroid behavioural capacities and neural circuitry.

For behavioural analysis, we have used videographic recording (at both standard and high-speed imaging rates) of insects interacting with real predators. We then generated ways to test the responses of the insects under partially restrained or tethered conditions with less natural, but more controlled stimuli. Finally, we moved to electrophysiological recordings of neurons within the relevant circuits. In most cases, extracellular multi-unit recordings were followed up with intracellular recordings and dye injection; this adds a cellular, neuroanatomical dimension to the behavioural story. While the overall research programme is not yet complete, the ultimate goal is to have circuit details of the sort which insects so readily supply—a description at the level of uniquely identifiable neurons. For details of the particular species used, neurophysiological and neuroanatomical methods, see the electronic supplementary material.

A substantial measure of naturalistic observation has been common at the initiation of research programmes on mammalian active sensing systems, and then studies have moved to analysis of central nervous system (CNS) mechanisms where the ecological validity is lessened. With programmes of study that make this transition, it is crucial to ask: (i) are cellular analyses in the CNS followed up by confirmative studies that return to more ecologically relevant levels, and (ii) is the gain in precision at the cellular level worth the loss in ecological validity as one moves away from naturalistic behavioural study to work on reduced preparations with abstracted stimuli? It is not necessary to sacrifice ecological validity in order to pursue studies of CNS physiology. However, it is technically demanding to balance good physiology with naturalistic behavioural study. There are successful examples of this balance from studies of mammalian active touch [22]. Relative ease of balancing naturalistic behavioural analysis with rigorous single-cell physiology is an attribute that insect model systems share with the more common rodent models of active touch.

3. Background: identified neurons and escape behaviour

Cockroaches possess multiple mechanosensory channels for detecting predators. The earliest known channel was that for detection of wind cues generated by predator movement—and sensed via filiform hairs on the cerci. The system is now a classic, having been described long ago by Roeder [23]. He characterized the first individually identifiable interneurons in orthoptereroids—the so-called ‘giant’ interneurons (or GIs) that ascend in the ventral nerve cord. They carry wind information from the caudally located cerci to leg motor circuitry in the thoracic ganglia (figure 1, circuit with black arrows). The system also has been elegantly analysed in naturalistic terms by Camhi (e.g. [24]) and it has been used in recent textbook summaries to illustrate general principles of neural coding and behavioural control [25].

The GIs are directionally selective in responding to wind, and information from at least four bilateral pairs of GIs is integrated to specify the appropriate direction of an evasive turn away from a lunging predator. The evidence linking the GIs to mediation of directional escape in cockroaches is extensive and quite direct [26–28]. Similarly, organized GI systems also have been described in cricket [29–31], locust [32] and praying mantis [33].

The GIs will not be described in any more detail here, because reviews of the GI system and escape behaviour are available (e.g. [24]). However, awareness of this system is important to the information presented here, because in all of the more recent studies of direct touch sensation mediated by the rostral antennae, it has been necessary to continually verify the independence of the rostral (touch) system from the caudal (wind) system. This has been done by including controls, in virtually all experiments, in which the cerci are removed or covered to make sure that the very sensitive wind system is not confounding interpretations about touch or vision.

The GI system of cockroaches is not an active touch system (for example, the cerci are not moved within wind-sensory fields) and the output is generally a contraversive turn and run (although see [34,35] for interesting descriptions of a certain amount of patterning within the seemingly unpredictable angles of turn generated during escape). Cricket cercal wind sensation continues to be a rich source of information on how the directional selectivity of the GIs is established and how the entire spectrum of wind sensation is displayed in a computational map in the terminal abdominal ganglion [36,37].

4. Escape in a different modality

In cockroaches with all of the GIs deleted from the ventral nerve cord, some evasive responses to wind are still present, but the character of the responses indicates that wind is not the adequate stimulus for ‘non-GI’ escape. Such responses require very intense wind puffs interacting with the rostral antennae suggesting that the antennae may be displaced or vibrated by intense wind, as they are during direct touch. The information is carried to thoracic leg motor centres on a separate set of interneurons, distinct from the GIs [38]. Once these distinct cells were identified, we termed them descending mechanosensory interneurons (DMIs; [39]; and see figure 1, circuit shown with red arrows).

These cells, like the GIs, have large calibre axons and lead to the same behavioural response: turn and run. The behaviour is triggered at very short latencies (about 25 ms on average), whereas the GIs trigger escape about 50 ms after a wind puff stimulates the cercal filliform hairs. The appropriate stimulus for the DMI system is indeed not wind, but touch delivered to one of the antennae [40,41]. In more ecologically valid studies, one can see that a range of natural predators are detected by antennal touch if they strike from close range and/or do not generate an appropriate wind cue [19]. So, cockroaches have two ‘giant fibre’ pathways: one for wind and one for direct touch, and they differ in terms of which sorts of predators they are likely to detect. Spiders are a good example of the type of predator detected by the antennal touch-sensory system.

Spiders have provided an entry point into understanding some of the complexities of the antennal touch-sensory system. In behavioural studies, cockroaches respond differently to controlled contact with an antenna from (i) wolf spiders versus (ii) conspecifics [42]. When cockroaches are allowed to approach and actively explore (palpate) the surface of a spider with the long antennal flagellum (for approx. 5 s), they produce escape reliably in response to subsequent controlled contact from the spider. (In order to have control over this interaction, the spiders were freshly killed and mounted in a holder that could be abruptly advanced against an antenna; figure 2a.) Cockroaches produce escape responses significantly less often when palpating another cockroach and then receiving controlled contact from that cockroach. This ‘discrimination’ is not observed when animals are touched with these two stimuli without any period for active touch prior to contact [42] (figure 2b).

Figure 2.

Antennal touch-evoked escape and the influence of ‘palpating’ the stimulus. (a) Reconstruction of antennal touch-evoked escape trial from video records. Spider stimulus (grey) moved towards cockroach (black) by solenoid. Frame numbers for cockroach turn are based on frame 0 = contact with antenna. Direction of subsequent run shown by arrow. (b) Percentage of trials with evasive turns in response to touching an antenna with a spider stimulus (black bars) versus a cockroach as stimulus (unfilled bars). Under normal conditions (Nm) the cockroach being tested was allowed several seconds to actively ‘palpate’ the stimulus before it was abruptly advanced against the antenna. Under these conditions, the percentage of responses was significantly different (asterisks) for the two types of stimuli. When no prior active touch was allowed, there was no significant difference (n.s.) in the percentage of responses to each stimulus.

The nature of the sensory cues obtained during palpation would most likely be of two general types: surface chemical or mechanical. Arthropods have a variety of compounds on the surface of their cuticles and the wolf spiders used here have numerous surface hairs that impart a texture to the surface. The cockroach flagellum is richly covered with mechanoreceptors and chemoreceptors. We conducted tests to try and distinguish which class of cues/receptors might be involved in distinguishing benign from threatening stimuli under the conditions we used.

When cuticular surface chemicals are extracted from cockroach and spider test stimuli, and animals are retested under conditions where they can palpate the surface of the test stimulus before the antenna is tapped, animals still discriminate between spiders and conspecifics [38]. So the surface chemicals of a wolf spider are not necessary for triggering escape turns at a relatively high frequency, nor are the surface chemicals of a cockroach necessary for triggering escape at significantly lower frequencies.

Might surface chemical cues be sufficient to allow stimulus discrimination via the antennae if differences in textural cues are eliminated? This can be tested with identical stimulus dummies, scented with cockroach or spider cues. Cockroaches that are stimulated with these objects, under conditions of prior exploration with the flagellum, respond with escape turns at somewhat lower frequencies overall, but they do not respond differentially to the objects marked as ‘spider’ versus those marked as ‘conspecific’. Thus, objects with similar textures, differing only in the chemical cues of the two species, are not discriminated under the conditions we tested.

These observations suggest that (i) surface textural cues obtained during active touch are behaviourally significant to the insect, (ii) this information must be retained for at least a short period of time to influence subsequent behaviour, and (iii) it implies that receptors on the flagellum contribute to obtaining information on stimulus identity (dashed line pathway in figure 1). However, this last point is still uncertain as it has not been tested directly, and we know that there are many additional receptors in the basal antennal segments that influence escape (see below).

How sophisticated is the discrimination of stimulus identity in this system? Our data suggest that a very simple algorithm may underlie it. Escape is elicited with a relatively high probability when the antennae are tapped without the chance to palpate the stimulus first, and also with a relatively high probability by spider stimuli after a period of palpation. However, when a conspecific taps the antenna after palpation (cockroaches are sometimes found at high densities under ‘natural’ conditions) escape probability is lower. A simple explanation would be that the ‘default condition’ would be for abrupt antennal contact to elicit escape with a high probability, but that sensory inputs associated with conspecifics tend to lower the probability. This would seem to make some sense in terms of simplifying the decision process into an interaction, at some level, between afferent inputs [43]. It also would not require processing information about the many potential predators that might be detected by antennal contact. This set of ideas warrants further study.

5. The antennal mechanosensory pathway

If antennal information derived from palpation can alter escape probabilities, then what specifically is the triggering circuit where this influence is expressed? When the long antennal flagellum is amputated and a plastic fibre (of the same length and mass as the flagellum) is attached with surgical cement to the antennal base, deflection of such a ‘prosthetic flagellum’ is still able to activate the DMIs and produce escape responses [42]. However, escape is dramatically reduced by constraining the movement of the two basal segments (the scape and the pedicel) with wax. This prevents the intact flagellum from causing deflections or vibrations of the basal segments when it is tapped. Thus, the crucial triggering receptors for escape are not on the flagellum.

The types of receptors at the antennal base include hair plates and campaniform sensillae on the surface of the scape and pedicel [44,45] and chordotonal organs (COs) below the cuticle (e.g. [40]). Scapal hair plate receptors have been implicated in antennal positioning, and object localization during walking and searching movements [46]. Scapal hair plate receptors have phasic-tonic discharge [45]—a characteristic that would be appropriate to a role in localizing objects [46], but perhaps not a characteristic to be expected for triggering escape. We have recorded from scapal CO afferents in cockroaches and found that some have phasic discharges. Additionally, when the afferents were labelled they could be seen to enter the brain and closely overlap the main neuritic arbours of identified DMIs [47] (figure 3a). Recordings from DMIs in crickets have demonstrated input from scapal COs and perhaps other pedicellar receptors [48].

Figure 3.

Characteristics of the antennal touch pathway in cockroach and cricket. (a) An afferent from the antennal chordotonal organ (red) in relation to the dendrites of neuron DMI a-1 in cockroach. (b) Mechanical tuning of response to touching the distal flagellum recorded with a transducer at the antennal base. Recording is from cricket, the tuning curve for cockroach is virtually identical. (c) Behavioural responses of cockroaches (left) and crickets (right) to antennal stimulation in intact animals (int.) compared with animals with hair plate receptors (hair) shaved off at the antennal base versus the chordotonal organ (CO) surgically cut. Coloured bars give % of escape turns, white bars give % of trials on which animal stepped back, but did not turn and run. Only the change following CO removal was statistically significant.

More recently, we studied the ability of touch stimuli applied to the antennal flagellum to be transmitted mechanically to the antennal base. Small deflections in the frequency range below100 Hz were most efficiently transmitted to basal segments (figure 3b) in both cockroaches and crickets. Single unit recordings showed that this is the region of peak physiological sensitivity for scapal chordotonal receptors and for the DMIs [21]. Moreover, selective transection of the CO-type receptors within the scape led to a significant disruption in the ability of unrestrained cockroaches and crickets to escape in response to antennal touch. However, shaving the hair plate receptors from the basal segments did not have this effect (figure 3c).

This is an example of a situation where we returned to more naturalistic behaviour studies after delving into cellular details in order to confirm that we were still on the right track. From these sorts of studies (in our laboratory and others), we believe that receptors sensitive to overall displacement of an antennal flagellum (scapal CO-type receptors) are involved in triggering escape. The only receptors at the base for which we currently lack a direct test of involvement in escape triggering are the campaniform sensillae; so at present we cannot rule out that they may play some role.

6. Visual guidance of touch

Orthopteroid antennae are highly mobile. Cockroaches spontaneously move their antennae about the sensory field, and so do crickets—although the degree of movement varies from species to species (Y. Baba 2010, unpublished data). The antennae of orthopteroids also are moved in patterns related to locomotion. A good description of this exists for antennal position during walking in stick insects [49] and cockroaches [50]. In rapidly running cockroaches, the antennae are held pointing forward, so as to facilitate object detection and crash avoidance [51]. In a more slowly walking insect, the antennae play a role in assisting with leg placement and navigation over or around objects [52]. These are additional examples of active touch in orthopteroids (see [17]).

Any model of antennal-mediated behaviour must take the mobility of the antennae into account. It has been known since early work by Honegger [53] that crickets can visually locate moving objects, as indicated by antennal movements. This tracking motion is spatially accurate, fast and saccade-like, and may involve one or both antennae [53,54]. Robust visual guidance of the antennal flagellum towards novel objects immediately raises the question of influences from vision and active touch on escape triggered through the antennae.

Observations of free-ranging cockroaches suggested that their antennal orienting movements are under visual guidance. When the antennae were positioned at a typical point within their range (60–80° from the midline), an object introduced into the visual periphery (and at a distance from the animal of approximately one antennal length) reliably caused movement of the ipsilateral antennal flagellum towards the object. When the compound eyes were covered with opaque paint, this response was essentially eliminated [55]. We then replicated the observation of Honegger from crickets so that we could directly compare cricket visually triggered exploration with that of cockroaches. Under our testing conditions, crickets responded in the same way as cockroaches and showed the same dependence on the compound eyes (figure 4a, and see the electronic supplementary material). Thus the orthopteroid visual system can play a role in guiding an antenna to the location of a novel object entering the visual field. In these animals with long antennae (greater than one body length) and immobile eyes, this is how they ‘look at’ and examine new or approaching objects. This behaviour provides a crucial affordance for active touch in the sensory world of cockroaches and crickets, and it adds an active touch loop to the full circuit that must be taken into account when considering the total control network for escape (figure 1, arrows in blue).

Figure 4.

Visually triggered antennal guidance to explore novel objects. (a) Testing of cockroaches (Periplaneta americana, white bars) and crickets (Gryllus bimaculatus, black bars) under similar conditions (inset at top) showed reliable orienting of the antennae to an object in peripheral visual field, that was blocked by covering the compound eye. (b) Trajectories from six individual trials where stimulus was presented ‘out of reach’ from the antenna. Horizontal distance is scaled so that 0 = middle of antennal movement field or about 45° from directly ahead. Circle outlines the visible edges of the target stimulus on the lateral field. See text for details.

Given this important role for vision at the outset of behavioural interactions that can lead to escape, it is important to note that vision, by itself, cannot trigger escape. When vision is occluded, there is no effect on responsiveness to antennal touch, no change in the average latency of touch-evoked escape and no change in any typical performance measures for the initial turn away from the triggering stimulus [55]. By contrast, occlusion of vision has an effect on the performance of the run phase that follows the initial turn [55], but this is not germane to the present review.

7. The circuit for visually guided active touch

The sensorimotor circuit for guidance of the antennae has been approached first from the motor side. In cockroaches [56], crickets [57] and stick insects [49], the muscles moving the antennae and the pool of motor neurons that control them have been fully described. All of these species have seven antennal muscles (five in the head and connecting to the scape and two in the scape connecting to the pedicel). There are no muscles within the long flagellum. The reported number of motor neurons is from 14 to 18 on each side of the brain (in the deuocerebrum) and there are in addition one or two dorsal unpaired median neurons in the suboesophageal ganglion that send processes bilaterally and are probably modulatory [58]. Thus, one has a motor system here with a very tractable number of elements.

The search for cells involved in cockroach visual processing will be facilitated by a comparative framework. Several descending brain interneurons signal a novel visual stimulus entering a cockroach's sensory field. One of these visual interneurons (figure 5) is particularly conspicuous in neural recordings when visual stimuli enter the field contralateral to its descending axon. Recordings of its visual activity with simultaneous video records of antennal movements showed that firing of this interneuron always preceded antennal movements towards novel stimuli [59,60]. This interneuron is similar to the descending contralateral movement detector (DCMD) so well known from locusts (e.g. [61,62]). The cockroach homologue to DCMD was previously described from extracellular recordings [63]. The data we have gathered so far suggest that the cell we have studied physiologically and stained for anatomical reconstruction is probably the cockroach DCMD (figure 5). Additionally, we found a cell with similar physiology and morphology in crickets (V. Leung & C. Comer 2005, unpublished data).

Figure 5.

Identified descending visual interneuron (DVI) of cockroach that may be homologous to the locust DCMD. A wholemount reconstruction of the cell is shown at left. There would be a mirror symmetrical interneuron on the opposite side of the brain. The blue arbour is positioned to receive visual inputs, the red arbour extends into the neuropil where the dendrites of antennal motor neurons are located. At right, a summary of one of its physiological tuning properties as determined from intracellular recordings. When stimuli are presented in the visual field contralateral to the axon, there is a more vigorous response to approaching than to receding stimuli. This physiological tuning is similar to that found in locust DCMD; the presence of a neurite entering the neuropil relating to antennal motorneurons is not seen in the locust DCMD.

In locusts, DCMD has been thought for quite some time to play a role in the escape jumping behaviour of the insect (e.g. [64]). However, it seems that DCMD may assume a slightly different degree of centrality to escape depending on whether an animal is flying or standing, and it may be more specifically related to the occurrence of short latency escapes [65,66]. To be sure, in cockroaches and in crickets there are several cells that reliably discharge in response to escape eliciting visual stimuli [67] and prior to orientation of the antennae [60], and it will take careful study with direct manipulations to determine what each of these descending visual interneurons (DVIs) contributes to antennal guidance. The main point is that the visual control loops of orthopteroid species are tractable—there appear to be fewer key interneurons than there are motor neurons on each side of the brain and so again the circuit should be able to be dissected to the level of uniquely identifiable neurons.

More recently, we have been interested in the control strategies for directing the antennae accurately to a visual target. We have begun addressing this by analysing the trajectories of individual, visually triggered antennal reorientations. By acquiring video records of antennal movements from at least two cardinal directions, one can reconstruct the movements in both the mediolateral and dorsoventral planes (see example in figure 4b). In the figure, movements proceed spatially from left to right (from the frontal sensory field towards the periphery). They are not smooth, but characteristically display transient increases in elevation (sometimes several) as rotation in the mediolateral plane progresses. All the trajectories ended near the apparent edge of the visual target. There was no actual contact because the target was beyond the reach of the antenna. This indicates that the endpoint of the movements may be (i) fully pre-programmed, or (ii) that visual feedback or proprioceptive cues of antennal position may be capable of determining movement metrics in lieu of exteroceptive touch-sensory cues. Of course, these are not mutually exclusive possibilities. It should be possible to thoroughly describe the construction of antennal movement trajectories with further experiments, but these observations serve to indicate that visually triggered antennal movements that can lead to palpation have a purposive, information-seeking character.

8. Design principles and evolutionary possibilities

(a). Separation of functions

One generalization that has been validated in many studies of vertebrate neural systems is that localizing and identifying stimuli are two distinct operations. Indeed, there is strong evidence for this in the vertebrate visual system and it is also evident in the rodent whisker system [10]. It is assumed in vision that processing for object recognition and processing for localization diverge from early stages of cortical processing into separate ‘streams’ [68]. Diamond et al. [10] have speculated that in the whisker sensory system, these two functions might turn out to be distinct even closer to the sensory periphery (vibrissal follicles).

A dichotomy of a similar sort can be made for active antennal processing in cockroaches based on the periphery: directional escape turns can be triggered without an intact flagellum, but stimulus identification seems to require one. Therefore, an immediately appealing model is that flagellar receptors identify a stimulus, but basal receptors provide localization information. However, there are complexities that need to be worked through before such a model can be accepted.

Basal receptors encode antennal position [45] and this is true widely in the orthopteroids and other insects [69]. Additionally, antennal position is known to affect the direction of an initial escape turn elicited by touching the antennae [55] (symbolized by the open arrow in figure 1 that points from the antennomotor control loop to the leg motor control loop). So in several ways, basal receptors clearly are involved in localization—of an antenna, and of the animal's movements with respect to ‘palpated’ objects. It is important then, to know exactly how cockroaches and related insects acquire basic information on stimulus identity during active touch exploration of surfaces with flagellar receptors. As a flagellum is moved across a textured surface, its many protruding receptive hairs should interact with surface textural features that may stimulate the flagellar mechanoreceptors directly. Alternatively, or in addition, the interaction of the flagellum with the surface may provide affordances for ‘stick–slip’ events. These might lead to activation of some basal receptors, because mechanical vibrations of the flagellum—at least those in the low frequency region—are efficiently transmitted mechanically towards the antennal base. Nonetheless, this is an addressable question at the level of the well-defined receptors and afferents in orthopteroid insects. We suggest that it is worth teasing the details apart to test the generality of ideas emerging from mammalian studies.

(b). Active touch in relation to escape

What we have described of the core escape system (e.g. black and red circuits in figure 1) indicates that there are two distinct ‘giant fibre’ pathways in cockroaches. There are data to support an extension of this model to crickets, especially, in terms of cells that have been found in crickets which appear to be homologues of the DMI described in cockroach [70,71], and we suspect that much of the diagram will apply—with species-specific variations—to other orthopteroids. It is this core system to which an active touch function is connected (blue circuit in figure 1) by way of visual control of the antennae. The degree to which the antennae are guided by vision to ‘look at’ and explore novel objects almost certainly will vary based on the visual capacity of different species, and the scale of their antennae: cockroaches and crickets have long antennae (typically greater than body length) and this is not true of all orthopteroid species.

The sensory and circuit logic that applies to the core escape system of cockroaches and crickets can be readily summarized. The cercal wind mechanosensory system converges on thoracic leg motor circuitry containing many individually identified cells, and the antennal touch mechanosensory system converges on the same general circuitry [72]. The upshot is that directional evasive behaviour can be initiated either on the basis of mechanosensory-based distance cues (wind) generated by certain movements of large predators (e.g. some toads and lizards), or the more intimate cues (touch) generated by smaller or more stealthy predators (e.g. spiders, praying mantises and mice).

It has been demonstrated that a cercal wind-sensory system can be found in insect groups less derived than the orthopteroids [73]. This led to the suggestion that abdominal giant fibres may be the prototype predator evasion system. Given that antennae are ubiquitous in the Insecta, it becomes important to place the large calibre descending mechanosensory system of orthoperoids, and touch—both passive and active—into consideration in evolutionary models. As cockroaches and crickets are not pronounced fliers and possess very long antennae, these attributes suggest that antennal mechanosensation and indeed active antennal touch may be no less fundamental to insects than is the passive mechanosensation provided by the cerci.

(c). Thinking about evolutionary innovation

At a cellular level, the visual nature of antennal guidance suggests a way to think about the addition of new behavioural capacities during neural evolution. As indicated earlier, the visual interneurons that we have studied so far include a prominent cell seen in recordings from both cockroaches and crickets that displays properties suggesting it is the DCMD homologue. Kral & Prete [12] have recorded extracellularly from the putative DCMD in praying mantis and made an interesting point about how it compares with locust DCMD. The sensory coding properties in mantis are strikingly similar to those reported in locust, but it seems to be doing very different things behaviourally in mantis.

We believe that the same is true of the DVIs that we have found in cockroach and cricket. They too have physiology that is very similar to that reported for locust DCMD (figure 5) but we are suggesting that they trigger antennomotor output in place of, or in addition to, any downstream effects on locomotion. One structural difference of note is consistent with this idea. Both the cricket and cockroach DCMD candidates have not only a neurite projecting into the visual neuropils in the brain, but also a second one projecting into the deutocerebral region where antennal motor neurons are located (figure 5). Locusts have relatively shorter antennae than cockroaches and crickets, and the locust DCMD does not have a branch in the antennal neuropil. Is it possible that but for a change in one seemingly minor architectural detail of DCMD, some orthopteroids are able to ‘look at’ and ‘palpate’ approaching visual stimuli with the antennae?

We do not know the answer yet, but we suggest that such a question is precisely the sort at which neurobehavioural studies should be aimed. There are suggestions emerging from genomic studies that small changes, such as a few base substitutions in a few genes, may lead to major adaptive changes [74]. It would be important to gain a sense of scale about how much cellular change in neural circuits is needed to produce new behavioural capacities. Comparative work on insects and rodents will be the place to look.