Can mechanism help explain insect host choice?


John P. Cunningham, School of Biological Sciences, University of Queensland, Brisbane, Qld 4072, Australia.
Tel.: +617 33657995; fax: +61 7 3365 1655; e-mail:


Evolutionary theory predicts that herbivorous insects should lay eggs on plants in a way that reflects the suitability of each plant species for larval development. Empirical studies, however, often fail to find any relationship between an adult insect’s choice of host–plant and offspring fitness, and in such cases, it is generally assumed that other ‘missing’ factors (e.g. predation, host–plant abundance, learning and adult feeding sites) must be contributing to overall host suitability. Here, I consider an alternative theory – that a fitness cost inherent in the olfactory mechanism could constrain the evolution of insect host selection. I begin by reviewing current knowledge of odour processing in the insect antennal lobe with the aid of a simple schematic: the aim being to explain the workings of this mechanism to scientists who do not have prior knowledge in this field. I then use the schematic to explore how an insect’s perception of host and non-host odours is governed by a set of processing rules, or algorithm. Under the assumptions of this mechanistic view, the perception of every plant odour is interrelated, and seemingly bad host choices can still arise as part of an overall adaptive behavioural strategy. I discuss how an understanding of mechanism can improve the interpretation of theoretical and empirical studies in insect behaviour and evolution.


Sometimes animals appear to behave in a way that does not make evolutionary sense. In the study of herbivorous insects, this is a familiar problem (Berdegue et al., 1998; Cronin & Abrahamson, 2001; Mayhew, 2001); insects are attracted to particular plant species, but often the prediction that good hosts should be preferred over poor hosts and poor hosts over non-hosts is not upheld by empirical data (Stephens & Krebs, 1986; Mayhew, 1997; Ballabeni et al., 2001; Scheirs & De Bruyn, 2002; West & Cunningham, 2002).

Most explanations for this mismatch between data and theory have revolved around the hypothesis that host quality (the suitability of each plant species for offspring survival) is not simply determined by a plant’s nutritional quality (Jaenike, 1978; Courtney et al., 1989) and additional ‘missing’ factors must contribute strongly to the overall fitness of the ovipositing female; these include host–plant abundance (Jaenike, 1978; Rausher, 1980; West & Cunningham, 2002), adult feeding sites (Scheirs & De Bruyn, 2002), insect learning (Cunningham & West, 2008), larval movement (Thompson, 1988; Cunningham et al., 2001) and predator avoidance (Bjorkman & Larsson, 1991; Ohsaki & Sato, 1994; Ballabeni et al., 2001).

An alternative explanation, however, is that nutritional quality is a good predictor of offspring survival, but the insect’s host selection behaviour is limited by its ability to process sensory information – that the behaviour is somehow constrained by the mechanism. This theory has been referred to as the information processing hypothesis (IPH) (Levins & Macarthur, 1969; Bernays, 2001; Egan & Funk, 2006) with its key prediction being that generalist insects (having more information to process) should be less efficient in their responses towards host plants compared to specialists. The IPH has achieved some support from empirical studies (Janz, 2003; Egan & Funk, 2006), but to date, there has been little discussion of what these mechanisms might be or how these processing constraints might arise.

The aim of this study is to explore how a mechanistic view of insect behaviour can be used to explore evolutionary theory. It focuses on olfaction, which plays a key role in host finding and recognition in most herbivorous insects (Bruce et al., 2005; Dudareva et al., 2006) and on the mechanism of the insect antennal lobe (AL), where odour information from the environment is translated into an ‘odour code’ and delivered to the higher centres of the insect brain.

Perhaps one of the greatest hurdles of a mechanistic theory is that it involves complex neurophysiological processes, and for many behavioural scientists this means stepping into a daunting and alien field. Olfactory processing is certainly no exception and from the outset, it is worth clarifying that the elusive odour code is likely to be an integration of many different coding mechanisms within the AL (Kuebler et al., 2011). My conceptual model is a simplification of this complex process, and focuses on the most widely studied coding mechanism – spatial patterns of excitation. My aim is to explain this single coding mechanism in a way that highlights its importance in understanding insect host selection behaviour, without requiring prior knowledge of olfactory neuroscience. I begin by reviewing a number of important features of olfactory processing in the AL. I then use a simple schematic to explore how these features might influence the way host and non-host odours are perceived, and lead to constraints in the evolution of insect olfactory responses.

Understanding the olfactory mechanism: odour processing in the AL

The characteristic odour of a plant species is made up of a blend of individual compounds (volatiles). These volatiles are common to many plant species, genera and families (see reviews by Bruce et al., 2005; Dudareva et al., 2006; Pichersky et al., 2006; Raguso, 2008).

If an herbivorous insect is to distinguish between plant odours, its olfactory system must therefore be capable of telling apart blends of chemicals that share common volatile elements. In its simplest form, an insect could distinguish between two plant odours by identifying volatiles that are present in only one of the plants (often termed ‘key volatiles’). Given the enormous diversity of plant life, however, such unique, identifying volatiles are rare occurrences in nature (Bruce et al., 2005), which leaves the insect’s olfactory system with the more complex task of recognizing blend structure itself, i.e. the combinations of volatiles that co-occur in each plant odour.

Although odour identification is undoubtedly a product of the entire olfactory system, from the antennal receptors to the higher centres of the insect brain, a key hub of olfactory processing, often regarded as the ‘primary centre’, is the insect AL. Its function has been brought to light through pioneering research on the structure and neural wiring of the AL (Christensen et al., 1991; Hansson et al., 1992; Gao et al., 2000; Vosshall et al., 2000) and neurophysiological responses to insect pheromones and plant odours (e.g. Joerges et al., 1997; Christensen et al., 2000; Galizia & Menzel, 2000a,b; Carlsson et al., 2002; Christensen & Hildebrand, 2002; Hansson et al., 2003; Carlsson et al., 2005; Menzel et al., 2005; Deisig et al., 2010; see Lei & Vickers, 2008 for a detailed review).

A simple schematic to conceptualize the mechanism of the AL is presented in Fig. 1. From the complex blend of many volatile compounds that make up a plants odour, a subset of volatiles (Riffell et al., 2009) trigger sensory neurons [olfactory receptor neurons (ORNs)] in the insect antennae. These ORNs are volatile-specific, and each class of ORN relays information to a specific region within the AL, called a glomerulus. In this way, chemical information, relating to the volatile structure of a plant odour, is translated into spatial patterns of excitation within the AL (see Lei & Vickers, 2008). These excitation patterns are then relayed to higher centres of the insect brain via another set of neurons (projection neurons) and ultimately lead to behavioural responses, such as flying upwind towards the odour source.

Figure 1.

 How plant volatiles form spatial patterns within the antennal lobe (AL). (a) Elemental patterns. Out of the complex blend of volatile compounds that make up a plant’s unique odour, a subset of volatiles (in this example, the four coloured volatiles) are detected by receptors on sensory neurons [olfactory receptor neurons (ORNs)] in the insect antennae. Different ORN classes bear different receptor types (and are thus triggered by different volatiles), and each class of ORN relays information to specific regions (called glomeruli) in the AL. For example, in this schematic, the volatile represented by the green dots activates specific (green) ORNs, which relay information to glomerulus C. As a result, blends of volatiles are translated into patterns of excitation in the AL (yellow hexagons). The activated glomeruli evoke synchronized firing in output neurons (broader arrows), which send information to higher centres of the insect brain. Further processing then leads to behavioural responses such as upwind flight towards the odour source. (b) Pattern sharpening (blend-specific patterns). When a glomerulus is activated, interneuron activity (black arrows) can influence the level of excitation in neighbouring glomeruli by increasing (+) or decreasing (−) output activity. The global response of interneurons can be represented as a processing algorithm (the example here is A excites B, A inhibits C, B inhibits D), which sharpens output firing patterns. Red shading denotes an increased level of glomerular activity (increased output strength) evoked by excitation from both ORNs and interneurons.

For the purpose of this paper, which aims to understand how mechanism may constrain the evolution of insect host selection, two key features of the AL are summarized in the schematic: first, different plant volatiles lead to activation of different glomeruli within the AL (Joerges et al., 1997; Lei & Vickers, 2008) (Fig. 1a) and second, interactions within the AL lead to blend-specific patterns of output firing (Joerges et al., 1997; Pinero & Dorn, 2007; Deisig et al., 2010; Kuebler et al., 2011) (Fig 1b).

Blend-specific patterns are important because they convey contextual information – the whole (the pattern created by all volatiles together) is different from the sum of the parts (the combined patterns from each individual volatile). A recent study by Deisig et al. (2010) has elegantly demonstrated how spatial patterns of activity evoked by different plant odour blends are sharper and more distinct from one another than would occur if these patterns were simply the summed responses from individual volatiles. The key to how these blend-specific patterns are formed lies in the activity of a network of local interneurons, which connect glomeruli (Reisenman et al., 2005; Olsen et al., 2007; Root et al., 2007; Silbering & Galizia, 2007; Olsen & Wilson, 2008; Chou et al., 2010; Huang et al., 2010; Seki et al., 2010) (Fig. 1b), allowing activity within one glomerulus to influence activity in another. The way in which this network of interneurons achieves pattern sharpening has crucial implications towards our understanding of insect odour responses and their evolution.

Rules within the mechanism: the processing algorithm

Figure 2 presents a simple schematic for how the AL might process the odours of six plants, which differ in their suitability as hosts (for example in terms of nutritional quality); Plants 1 and 2 are good hosts, Plants 3 and 4 are poor hosts and Plants 5 and 6 are non-hosts. Each plant odour is comprised of a blend of four volatiles, from a total possible pool of nine volatiles (A–I) (n.b. only volatiles that trigger ORNs, are considered here). The schematic represents the way, in nature, plants share individual volatiles, but have specific blends. As can be seen, Plants 1 and 2 have similar volatile profiles (for example they could be related species, or even interplant differences in a single species – such as with and without damage by herbivory). Poorer hosts (3 and 4) and non-hosts (5 and 6) also share the same set of volatiles (i.e. from A to I), but in different combinations.

Figure 2.

 How an interneuron network could influence plant odour perception in the antennal lobe (AL). The schematic represents spatial patterns in the AL evoked by odours from six plant species. Each plant odour is comprised of four volatiles from a possible 9 (A–I), and each volatile increases excitation in (activates) the corresponding lettered glomeruli. Plants 1–4 are all host species (Plants 1 and 2 are good hosts and Plants 3 and 4 are poor hosts), whereas Plants 5 and 6 are non-host species. Level of excitation: clear = activity below a threshold for recognition by higher centres of the insect brain, yellow = moderate (behaviourally relevant) excitation, red = strengthened excitation. (a) In the absence of interneuron effects (elemental representation of volatiles), host plants have overlapping patterns, which cannot be simply categorized into good, poor and non-hosts. (b) Interneuron effects detailed in Algorithm 1 give rise to sharper, more distinct patterns, which can be more easily categorized (e.g. good host = red C, poor host = red G, non-host ≤ 3 active glomeruli). (c) A new interneuron algorithm (Algorithm 2) gives rise to a different perception of host and non-host odours. (d) Simulating an adaptive change in Algorithm 1, such that Plant 6 is now perceived as a host. In Algorithm 1x, interneurons from glomerulus I now excite glomerulus A. The change in perception of Plant 6 carries the ‘cost’ (or constraint) of changing perception of Plant 4 (poor host to good host) and Plant 5 (non-host to host).

In the schematic, each plant volatile activates a single glomerulus (hexagon) denoted by its corresponding letter. Imaging studies on the AL have shown that single volatiles often activate a number of glomeruli to different levels of excitation (Joerges et al., 1997; Galizia & Menzel, 2000a; Deisig et al., 2010; Carlsson et al., 2011; Kuebler et al., 2011), but the schematic has simplified this for the purpose of investigation. Similarly, in this schematic, each glomerulus can have only three levels of excitation: level 0 (clear) is activity below a behaviourally significant threshold, and activation level 1 (yellow) and the higher level 2 (red) lead to output to the higher centres of the insect brain. Here, level 2 can only be reached by stimulation by both incoming peripheral neurons (ORNs) and additional excitatory interneuron activity.

Figure 2a shows the spatial patterns of glomerular activity evoked by odours from the six different plants, when no interneuron effects are present. The key things to note are: first, the blend pattern is the sum of the individual volatile elements (elemental representation); second, all plant odours activate the same number of glomeruli (four) to the same level (yellow); and third, individual glomeruli cannot distinguish between plants (i.e. no one volatile, or activated glomeruli, signifies good host, poor host or non-host).

In Fig. 2b, interactive effects between glomeruli, simulating interneuron activity have been added. Activated glomeruli can further excite (increase by one level) and inhibit (decrease by one level) other glomeruli. For example, when glomerulus A is excited, interneuron activity increases excitation in glomeruli B and C and inhibits glomerulus E (if both excitatory and inhibitory effects occur in the same glomerulus, inhibition takes precedent – i.e. the glomerular excitation remains below a behavioural threshold).

The global AL response – the combined influence of all interneurons connecting glomeruli – can be expressed as an algorithm (i.e. if A = excited, then increase B and C and decrease E; if B = excited, then decrease E and F, etc.) This has been named Algorithm 1 (Fig. 2b). In Fig. 2c, interneuron activity has been modified by applying a new algorithm (Algorithm 2) to the schematic.

In Fig. 2d, I have considered how Algorithm 1 could be modified to include Plant 6 within the insect’s range of host plants; this could be seen as an adaptive change in host selection if, for example, the insect had evolved to detoxify the defence chemicals of this plant species. The change in perception of Plant 6 is achieved by changing the influence of glomerulus I on glomerulus A (from inhibitory to excitatory). Under Algorithm 1x, volatiles from Plant 6 now evoke a ‘host response’, but as a consequence of this change, non-host Plant 5 is also perceived as a host and Plant 4 has moved categories from poor host to good host.

What does the schematic tell us about the benefits and constraints of the processing algorithm?

  • 1Pattern sharpening. In Fig. 2b, the patterns coded for by different plant odours are more clearly defined when compared with Fig. 2a: (i) there is less overlap in activated glomeruli in the good, poor and non-host categories, (ii) the strength of activation (red glomeruli) has increased in the hosts and differs in good and poor hosts and (iii) the number of activated glomeruli has been reduced (to two glomeruli) in the non-hosts.
  • 2Categorization. A simple set of rules can now identify host categories in Fig. 1b; Four or more activated glomeruli = host, less than three = non-host. Additionally, glomerular activation can define host type (activated A = good host, activated E = poor host), as can glomerular firing strength (strong red B or C = good host, strong G = poor host)
  • 3Different algorithms, different patterns.Figure 2c shows how a different algorithm can significantly change the pattern formation. Under Algorithm 2, Plants 2 and 4 now evoke weak AL responses (they are perceived as ‘non-hosts’) and Plants 5 and 6 evoke strong AL responses (they are perceived as ‘hosts’). Thus, interneuron activity alone, without any change in glomerular number or arrangement, or incoming neuronal activity from the peripheral nervous system, can modulate the categorization of host and non-host odours. An insect with an AL that processes under Algorithm 2 has vastly different glomerular patterns evoked by the six plants and consequently perceives plant odours very differently from an insect with an AL that processes under Algorithm 1. Thus, interneurons could be a key site for evolutionary change.
  • 4Volatiles common to hosts and non-host odours are still important. From a behavioural perspective, it might appear that volatile A does not play a major role in influencing host selection – it is present in the odours of both good hosts but also in the odour of one of the non-hosts (Plant 5). From a mechanistic perspective, however, volatile A has a major influence on host recognition, through its effect on rules within the processing algorithm.
  • 5Evolutionary constraints in odour perception. In Fig. 2d, a change in perception of one non-host odour (Plant 6) is brought about by changes in interneuron activity connecting two glomeruli (I and A). Under this modification (Algorithm 1x), non-host Plant 5 is now perceived as a host and the perception of Plant 4 has changed from being perceived as a poor host (red glomerulus G) to a good host (red glomerulus C). Consequently, evolutionary change to recognize Plant 6 is constrained – it comes at the cost of changing the perception of Plants 4 and 5.
  • 6Constraints and odour similarity. The schematic also shows how the constraints that prevent Plant 6 from being perceived as a host without including Plant 5 are not merely a consequence of odour similarity. Plant 5 shares two volatiles with Plant 6 but also with Plants 1, 2 and 4. Nor are they a consequence of a particular volatile within the blend (e.g. Plants 5 and 6 both share volatile I, but so does Plant 4). Instead, constraints to evolutionary changes in perception can only be understood by considering the entire processing algorithm.


A mechanistic view of insect olfaction could improve our understanding of why insects respond to different plant species in the way we observe. Based on current neurophysiological research, the schematic presented here is used to show how insect olfactory processing in the AL may be governed by a set of rules, or algorithm. This algorithm is inherent in the activity of spatially distinct neurophysiological units (glomeruli) and the global network of interneurons that connect them. The key implication of this view of odour processing is that changes to the perception of one odour, brought about by modification to the algorithm, will influence the processing and perception of all plant odours; in short, that insect responses to host and non-host odours are constrained by the olfactory mechanism.

The schematic is used to explain conceptually how glomerular activity patterns within the AL, evoked by different plant odours, can be sharpened by interneuron activity. This pattern sharpening allows odours to be more easily categorized into good hosts, poor hosts and non-hosts. The schematic also shows how the algorithm determines the way hosts and non-hosts are categorized; different algorithms give entirely different patterns, implying that this may be a key site for evolutionary change in odour perception.

A single-step ‘evolutionary’ change in the processing algorithm is used to demonstrate how modification of the response towards a single plant odour can incur the cost of re-categorizing other host and non-host species. This suggests that evolutionary biases and constraints to behaviour can exist within the processing algorithm, and that the adaptive value of a change in perception towards one plant odour would depend on concurrent changes to processing (and thus perception) of other ecologically important odours. Inter- and intraspecific differences in interneuron wiring could lead to similar observed behavioural responses towards certain host–plant species (within and between species) with differing costs of adaptation towards recognition of a new host plant. In other words, adaptation would depend more on internal mechanism than on behavioural observations of ‘host preference’.

If natural selection acts upon the network of interneurons within the AL, we would predict that adaptation is towards the best processing algorithm – one that maximizes the total number and quality of all offspring laid on all plants and thus the best response to all hosts (Stephens & Krebs, 1986). The best algorithm may not, however, be one that can distinguish all host odours from non-host odours, nor one that can identify hosts relative to their individual quality (preference–performance correlations). When insects show attraction to odours of plants that do not appear to support larval development, or show stronger responses to certain ‘poor’ host species relative to other more suitable ‘good’ host species, this may occur as a constraint of the processing mechanism and does not necessarily imply that other missing factors that influence offspring fitness (such as predation and adult feeding) (Mayhew, 2001) must be sought after.

This mechanistic view of odour perception proposes that current neurophysiological evidence is in support of the IPH – neural processing may be constraining insect responses towards plant odours. The original hypothesis focuses on accuracy in host selection in generalist insects compared to specialists (Levins & Macarthur, 1969; Bernays, 2001); clearly, using this model for olfactory processing, the greater the number of plant odours that need to be distinguished among, the more constrained processing will become. A highly polyphagous insect may only be capable of broadly classifying odours, such that the majority of odours are responded to as good, poor and non-hosts. More finely tuned responses, such as preference–performance correlated responses within (e.g. decreased attraction to nutritionally deficient, or herbivore damaged plants) or between host–plant species, may only be possible in insect species with a restricted host range (Gripenberg et al., 2010), where a more narrowly defined set of odours is processed.

In a recent study, Carlsson et al. (2011) have provided empirical evidence that perceptual differences, related to host range, may occur within the AL. These researchers measured glomerular activity patterns evoked by different plant odours, in two butterfly species – one of which was a specialist (Aglais urticae) and the other a generalist (Polygonia c-album). In general, the activity patterns in the AL were similar between the two butterfly species, but the specialist had a more specific response towards the odour of its preferred host species, compared to the generalist. In this study, however, the patterns mainly reflected input activity to the AL (i.e. before interneuron processing). A comparative study along these lines, but reflecting output activity (i.e. after processing), could generate interesting data that further support the IPH and the predictions of the theory presented here.

With the mechanism of the AL in mind, the IPH could be extended from its generalist–specialist approach to the prediction that all odour processing is constrained. Evolutionary history – host plants on which an insect species evolved (Wint, 1983; Walter & Benfield, 1994) – would play an important role in setting the algorithms coding for host odour recognition, and subsequent evolutionary change towards utilizing new hosts would depend on the fitness costs of changes to the algorithm. Adaptation involving fewer neurological changes (less genes involved) would be expected to occur more rapidly than more complex changes (Matsubayashi et al., 2010). For any given insect species, processing algorithms conferring greater fitness returns may be possible (i.e. there may be algorithms that better categorize host species), but adaptation to the ‘fitter’ algorithm may require a substantial reorganization of neural processing (the interneuron network), with intermediate changes incurring high fitness costs. Thus, adaptive landscapes in olfactory responses, with fitness ‘valleys and peaks’ (Wright, 1932), would lead to host responses in polyphagous insects being relatively conservative across populations (Thompson, 1993), forming a barrier to behavioural adaptation.

The mechanistic view of olfactory behaviour has implications for the design and interpretation of empirical studies on insect–plant olfactory responses. It supports the growing body of behavioural research that has shown context, in the form of the co-occurrence of volatiles in odour blends, is crucial to understanding insect olfactory responses (Pinero & Dorn, 2007; Riffell et al., 2009; Beyaert et al., 2010; Tasin et al., 2010; Webster et al., 2010). From a mechanistic perspective, the ‘role’ of a volatile in host recognition is inherent in the way it influences the global response, or processing algorithm. Volatiles should not be seen as attractants or deterrents in themselves. Empirical studies may show that certain volatiles are common to a host or non-host species, but this does not imply that these volatiles have an independent effect on behaviour. Similarly, studies that demonstrate the release of particular volatiles by herbivore-damaged plants should be cautious with implications that volatiles have inherent ecological roles.

The validity of the theory presented here depends heavily on the role of the AL in sharpening, categorizing and coding odour information before presenting it to higher centres of the insect central nervous system (CNS). Natural selection may act on any part of the olfactory mechanism, and adaptations in peripheral responses and central responses would also be expected under selection pressures in the environment (Ramdya & Benton, 2010), affecting input to the AL and interpretation of AL activity. Down-streaming effects from the higher centres of the insect brain (e.g. through learning) may also influence processing within the AL (Faber et al., 1999; Denker et al., 2010).

Although much of the structure and functioning of the AL has been uncovered in recent years, precisely how odour blends are translated into an odour code is still an area of intense discussion (Lei & Vickers, 2008; Silbering et al., 2008; Martin & Hildebrand, 2010; Seki et al., 2010). Here, the emphasis is on the most widely examined mechanism in the AL, spatial patterns of excitation, but it should be borne in mind that other coding mechanisms, such as synchrony, frequency and latency of neuron firing, may also carry inherent information on odour composition and quality (Lei & Vickers, 2008; Martin & Hildebrand, 2010; Kuebler et al., 2011). Computer models and simulations capable of integrating these multiple dimensions of coding could be the next important step in understanding odour processing and odour perception in the AL. These simulations could be used to generate predictions, such as expected behavioural responses to different volatile combinations that could then be tested empirically. By bringing together theoretical and empirical studies on mechanism with observed behavioural responses, we may move towards a more complete understanding of insect–plant interactions and their evolution.


I thank Stuart West, Bronwen Cribb, Meron Zalucki, Gimmie Walter, Kim Boyle and two anonymous reviewers for comments on the manuscript. This work was funded by the Australian Research Council.