The hungry caterpillar: an analysis of how carbohydrates stimulate feeding in Manduca sexta

In 1988, Schoonhoven and Blom developed a model to explain how taste modulates feeding in insects (Schoonhoven and Blom, 1988). They proposed that the size and nature of an insect's feeding response are determined by the algebraic sum of inputs from two functionally distinct classes of taste cell: one responds to nutrients(e.g. carbohydrates or amino acids) and the other responds to unpalatable compounds (i.e. `bitter' taste stimuli). Greater input from the nutrient-sensitive taste cells was hypothesized to elicit an appetitive response, whereas greater input from the bitter-sensitive taste cells was hypothesized to elicit an aversive response. There is widespread support for this model, spanning at least four insect orders: Lepidoptera(Ma, 1972; Blom, 1978; Schoonhoven and van Loon,2002; Omura and Honda,2003; Sasaki and Asaoka,2006), Diptera (Dethier,1976; Wang et al.,2004), Orthoptera (Blaney,1981; Chapman et al.,1991) and Hymenoptera(Scheiner et al., 2001; de Brito Sanchez et al.,2005).

In the present study, we examined an insect (Manduca sextacaterpillars; Sphingidae) that appears to contradict the predictions of the Schoonhoven and Blom model (Schoonhoven and Blom, 1988). Despite displaying a robust aversive response to bitter stimuli (Frazier, 1986; Glendinning et al., 1999; Glendinning et al., 2006), M. sexta shows a weak-to-nonexistent appetitive response to amino acids and carbohydrates (Bowdan,1995; Glendinning et al.,2000). The lackluster appetitive response to carbohydrates is perplexing because the peripheral gustatory system of M. sextaresponds vigorously to carbohydrates. For instance, even though glucose (100 mmol l^–1) and inositol (10 mmol l^–1) each strongly stimulate several classes of taste cell, the caterpillars take as many bites (over a 2-min test) from disks treated with glucose or inositol as from disks treated with water alone(Glendinning et al.,2000).

Given that most plant tissues contain multiple carbohydrates(Hardinge et al., 1965; Somogyi and Trautner, 1974; Clements and Darnell, 1980; Nelson and Bernays, 1998), it is possible that a herbivorous insect such as M. sexta would have evolved a taste system that responds selectively to mixtures of carbohydrates. If so, then this would explain why single carbohydrates (e.g. glucose or inositol) were so ineffective at eliciting an appetitive response in previous studies (Bowdan, 1995; Glendinning et al., 2000). In humans, carbohydrate mixtures often produce synergistic effects, such that the mixture generates a sweetness intensity that is higher than would be predicted based on the sum of the intensities generated by the individual component carbohydrates (Schiffman et al.,1995; Schiffman et al.,2000). Similarly, in rodents, binary mixtures of sweeteners can stimulate intake much more effectively than individual sweeteners(Valenstein et al., 1967; Capretta, 1970; Smith et al., 1976).

Like many other insects (Schoonhoven and van Loon, 2002; Thorne et al., 2004; Wang et al.,2004), M. sexta has a heterogeneous population of carbohydrate-sensitive taste cells. It has three bilateral pairs of carbohydrate-sensitive taste cells, and at least two of these pairs have different response properties(Schoonhoven, 1969). As a result of this peripheral organization, it follows that (1) no single carbohydrate could stimulate all three classes of carbohydrate-sensitive taste cell, and (2) some carbohydrate mixtures could stimulate more carbohydrate-sensitive taste cells than others. We hypothesized that M. sexta monitors the pattern of input from the different classes of carbohydrate-sensitive taste cell, and that it activates an appetitive response only when all three classes of taste cell are activated simultaneously. To test this hypothesis, we performed three experiments. In Experiment 1, we determined which carbohydrates activate the peripheral taste system of M. sexta. In Experiment 2, we ascertained the response properties of the different carbohydrate-sensitive taste cells. In Experiment 3, we asked whether the magnitude (or intensity) of the initial feeding response increases with the number of activated carbohydrate-sensitive taste cells.

We maintained a breeding colony of tobacco hornworms (Manduca sexta; Sphingidae) in our laboratory. This colony was derived from eggs that were donated by the Manduca rearing facility at the ARL Division of Neurobiology, University of Arizona, AZ, USA. The caterpillars were reared on a wheat germ-based artificial diet (Bell and Joachim, 1976), and were maintained in an environmental chamber with a 16 h:8 h L:D cycle at 25°C. All experiments were performed with caterpillars during the first or second day of their fifth larval growth stage (instar). All caterpillars were naive to the taste stimuli prior to testing. To control for any potential differences among caterpillars from different egg batches, individuals from each batch were interspersed randomly across treatment levels, according to a blind procedure. We provide sample sizes in the figure legends.

We tested seven carbohydrates: six simple and one complex. Four of the simple carbohydrates are relatively abundant in the insect's host plants[myo-inositol, sucrose, fructose and glucose(Nelson and Bernays, 1998)],and the other two are known to stimulate taste cells in other caterpillar species [maltose and trehalose(Schoonhoven and van Loon,2002)]. The complex carbohydrate, Polycose™, is a mixture of glucose polymers ranging from G1 (glucose) to G30; it is water-soluble, has an average molecular weight of 1000 Da (provided by the manufacturer), and stimulates feeding in many species of mammal(Sclafani, 1987). We considered polycose to be a representative starch, the most abundant carbohydrate in plants.

We tested all carbohydrates, except inositol and Polycose, at a 200 mmol l^–1 concentration. We selected 200 mmol l^–1because preliminary studies indicated that it is the lowest concentration that elicits a maximal neural response in both sensilla. We used a 1 mmol l^–1 concentration of inositol because preliminary studies indicated that it elicited an excitatory response similar to that of 200 mmol l^–1 sucrose. We used 100 mmol l^–1 Polycose because this concentration is highly palatable to several species of rodent(Sclafani, 1987; Glendinning et al., 2005). For the taste cell recordings, we presented the carbohydrates in an electrolyte solution (i.e. 100 mmol l^–1 KCl in deionized water). We purchased all carbohydrates, except Polycose, from Sigma-Aldrich (St Louis,MO, USA); we obtained the Polycose from Abbott Laboratories (Columbus, OH,USA).

For those carbohydrates that elicited an excitatory response in at least one class of taste sensillum (i.e. a response significantly greater than that to solvent alone), we generated concentration–response curves. To this end, we selected concentrations that spanned the dynamic range of response.

Following 30 min of food deprivation, we recorded neural responses of each taste sensillum with a non-invasive extracellular tip-recording technique(Gothilf and Hanson, 1994). In brief, we placed a glass electrode containing a taste stimulus solution over the tip of a lateral or medial styloconic sensillum(Fig. 1A,B). To minimize any potential carry-over between successive recordings, we paused for at least 3 min between stimulations. To minimize the effects of solvent evaporation at the tip of the recording/stimulating electrode, we drew fluid from the tip with a piece of filter paper immediately before stimulation. For each caterpillar, we made recordings from a single lateral and a single medial styloconic sensillum.

We recorded extracellular alternating current signals with the Tasteprobe amplifier system (Syntech, Hilversum, The Netherlands)(Marion-Poll and Van der Pers,1996). We preamplified each recording 10×, ran it through a band-pass filter set at 100–1200 Hz, fed it into a computer through a 16-bit analog-to-digital converter board, and then analyzed it off-line with Autospike software (Syntech).

In this experiment, we tallied the total excitatory response (i.e. total number of spikes that occurred during the initial 1000 ms of stimulation)separately for each taste sensillum. To determine which of the carbohydrate solutions elicited a detectable excitatory response, we compared the number of spikes elicited by the electrolyte solution alone with that elicited by each carbohydrate solution, using one-way factorial analysis of variance (ANOVA)and Tukey's honestly significant difference (HSD) test. For those carbohydrates that generated strong excitatory responses, we also obtained concentration–response curves. We analyzed the curve for each carbohydrate and sensillum separately, using one-way repeated-measures (RM)ANOVA. In this and all subsequent experiments, we set the alpha level at 0.05.

We used the tip-recording technique to record excitatory responses of the lateral and medial styloconic sensilla to the following stimulus triads: (1)glucose, sucrose and the binary mixture of both; (2) glucose, inositol and the binary mixture of both; and (3) sucrose, inositol and the binary mixture of both. We used the same concentrations of each carbohydrate as in the previous experiment, and stimulated one lateral and one medial sensillum from each caterpillar with each stimulus triad. To control for order effects, we randomized the presentation sequence of the triads, and the stimuli within each triad.

Because the lateral and medial styloconic sensilla each contain four taste cells, any given neural response could contain spikes from several taste cells. Thus, we developed a spike classification procedure based on earlier studies in our laboratory and by others(White et al., 1990; Bernays et al., 2000; Glendinning et al., 2002). We avoided using waveform algorithms to assign spikes to specific taste cells because spike amplitudes often changed with time, and spikes often interacted to produce irregular waveforms. Instead, our spike classification approach was visual, using relative spike height and shape, and the regularity of firing to separate and assign spikes. When a neural record was dominated by spikes of similar size and shape, which occurred in regular temporal sequence, we assumed that a single cell produced the spikes.

During the analysis, we found that the response of the lateral styloconic sensilla to sucrose always contained a population of spikes that followed a tonic pattern of firing (as indicated by a unimodal distribution of inter-spike intervals); the remaining spikes occurred less frequently and out-of-phase with the tonic pattern. In this case, we assigned the spikes that fit the tonic pattern to the `dominant' category, and the remaining spikes to the `non-dominant' category. However, the response of the lateral and medial styloconic sensilla to glucose or inositol always contained a population of spikes that followed a phasic-tonic pattern of firing (as indicated by a bimodal distribution of inter-spike intervals); the remaining spikes occurred less frequently and out-of-phase with the phasic-tonic pattern. In this latter case, we assigned the spikes that fit the phasic-tonic pattern to the`dominant' category, and the remaining spikes to the `non-dominant' category. Note that we assume that spikes assigned to the dominant category are derived from a single taste cell, and that spikes assigned to the non-dominant category could be derived from up to three taste cells.

We inferred that two carbohydrates (e.g. glucose and inositol) stimulated the same taste cell within a sensillum if the neural responses met two conditions. First, the binary mixture of carbohydrates had to elicit significantly more dominant spikes than each carbohydrate alone. This would indicate that both carbohydrates were acting additively on one taste cell,causing it to fire more rapidly than either carbohydrate alone. Second, the binary mixture of carbohydrates had to elicit the same number of non-dominant spikes as each taste stimulus alone. This would indicate that the non-dominant spikes in each trace reflected the response to the electrolyte solution.

By contrast, we inferred that two carbohydrates (e.g. glucose and sucrose)stimulated different taste cells within a sensillum if the neural responses met a different set of conditions. First, the binary mixture of carbohydrates could not elicit significantly more dominant spikes than either carbohydrate alone. This would indicate (1) that the binary mixture of carbohydrates was not acting additively on a single taste cell, and (2) that the dominant spikes from the binary mixture reflected the response to one of the carbohydrates in the mixture (e.g. glucose). Second, the binary mixture of carbohydrates had to elicit significantly more non-dominant spikes than each carbohydrate alone. This would indicate that the non-dominant spikes from the binary mixture reflected the response to the second carbohydrate in the mixture (e.g. sucrose), whereas the non-dominant spikes from each of the carbohydrates reflected the response to the electrolyte solution.

We sought to determine how many carbohydrate-sensitive taste cells existed in both the lateral and medial styloconic sensilla, and the response properties of each of these taste cells. To this end, we measured the neural response of the dominant and non-dominant taste cells, separately for each taste stimulus (see above for details). For a given neural response, we tallied the number of spikes that occurred across the initial 1000 ms of stimulation. We used one-way RM-ANOVA to compare the number of dominant (or non-dominant) spikes generated by each triad of taste stimuli (i.e. two carbohydrates alone and in the binary mixture of both). If the ANOVA was significant, then we ran a post-hoc test (Tukey's HSD test, adapted for a within design). We used this information, together with a detailed analysis of the neural traces (e.g. spike shape, amplitude and temporal pattern), to ascertain which of the carbohydrates (i.e. sucrose, glucose and inositol)stimulated the same or different taste cells within each taste sensillum.

We dissolved each carbohydrate in deionized water and then pipetted the test solution onto a glass-fiber disk (Whatman GF/A, 4.25 cm diameter;henceforth, `disk'). We presented the disk to the caterpillar immediately after adding the test solution so as to reduce evaporative water loss and to make the results comparable to the electrophysiological recordings. We used the following test solutions (at room temperature): water alone, sucrose (200 mmol l^–1), glucose (200 mmol l^–1), inositol(1 mmol l^–1) and binary mixtures of each solution (i.e. 200 mmol l^–1 sucrose plus 200 mmol l^–1 glucose;200 mmol l^–1 sucrose plus 1 mmol l^–1inositol; and 200 mmol l^–1 glucose plus 1 mmol l^–1 inositol). We selected the indicated concentration of each carbohydrate because they were used in the previous experiment.

We recorded the timing of all bites taken over the course of an entire meal. Our biting test consisted of four steps. (1) We placed a caterpillar in the `food-deprivation arena', which consisted of a clean (inverted) Petri dish covered by a clear plastic cylinder (7.5 cm in diameter, 10 cm tall). We fasted the caterpillar in this arena for 30 min to standardize its `hunger'state. (2) Then, we transferred the caterpillar to the `test arena', which was identical to the food-deprivation arena in all respects except that a piece of cork (1 cm in diameter, 3–4 mm high) had been taped to the middle of the inverted Petri dish. Immediately before each test session, we pinned a disk to the piece of cork, and then moistened it with 400 ml of the test solution (see above for details). (3) Next, we placed the caterpillar on the edge of the glass-fiber disk, positioning it so that its legs and prolegs grasped the edge of the glass-fiber disk securely. (4) Once the caterpillar had brought its mouthparts into contact with the surface of the glass-fiber disk and appeared to be tasting it (i.e. drumming the surface of the glass-fiber disk with its chemosensilla), we began the biting test. To be included in the experiment,the caterpillar had to initiate biting on the disk within 8 min of the beginning of the trial. We recorded the timing of all bites until the `meal'ended. We defined a `meal' as a period of relatively continuous biting, which ended once the caterpillar ceased biting for 2 min(Bowdan, 1988). To determine the amount of disk area eaten during the meal, we let the disks dry overnight and used SigmaScan (SPSS, Chicago, IL, USA) to quantify the two-dimensional area of disk (in mm²) that the caterpillar removed during its meal.

We developed three procedures to ensure that the observer could record the timing of each bite easily, accurately and without bias. First, by positioning the test arena on a turntable-like device, we could rotate the caterpillar and keep its mandibles clearly visible as it fed (this rotation did not disrupt the feeding of the caterpillar). Second, we recorded the timing of each bite with a software-based event recorder; this was possible to do accurately because the caterpillars emitted bites at a relatively low frequency (0 to 3 Hz), and because their black mandibles contrasted with the white experimental disks. Third, we kept the observer blind to the identity of the carbohydrate solution in the experimental disk.

Each caterpillar was run through a single meal. We analyzed several features of its biting response: (1) latency to initiate feeding (i.e. the time that elapsed between initially tasting the surface of the disk and initiating biting); (2) total number of bites exhibited during the initial 10 s, 120 s and then across the meal; (3) duration of the meal; (4) disk area consumed (in mm²) across the meal; and (5) mean bite size (i.e. disk area consumed divided by the total number of bites). For each biting measure, we made comparisons across taste stimuli, using the Kruskal–Wallis test and Dunn's multiple-comparison test. We used nonparametric tests because the data were not normally distributed.

Finally, we determined the molar density of carbohydrates in each type of disk (i.e. moles of carbohydrate per mm² of disk). Given that we applied 400 μl of test solution to each disk, and that the area of the disk(at the beginning of each test) was 1418.6 mm², it follows that there was 0.282 μl of carbohydrate per mm² of disk. Because we knew the molar concentration of carbohydrate in each test solution, we could determine the moles of carbohydrate in each mm² of disk.

We sought to determine which carbohydrates stimulate the peripheral taste system of M. sexta. To this end, we focused on the medial and lateral styloconic sensilla (Fig. 1A)because they are the only ones with taste cells that generate excitatory responses to carbohydrates (Schoonhoven and Dethier, 1966; Glendinning et al., 2000). Previous studies demonstrated that carbohydrates do not elicit excitatory responses in any of the other taste sensilla [i.e. the epipharyngeal and maxillary palp sensilla(de Boer et al., 1977; Glendinning et al., 2000)(J.I.G., A.J. and A.T.R, unpublished data)].

The medial styloconic sensillum responded to the carbohydrates with the following relative magnitude of response:inositol>glucose>sucrose=trehalose=fructose= maltose=Polycose=solvent alone (Fig. 2A). However, the lateral styloconic sensillum responded to the carbohydrates with different relative magnitudes of response: inositol=sucrose>glucose>fructose=Polycose=maltose=trehalose=solvent alone(Fig. 2B). Because inositol,sucrose and glucose were the only carbohydrates that elicited responses significantly greater than solvent alone (in at least one class of styloconic sensillum), we limited all subsequent experiments to these three carbohydrates.

We found that the neural responses of each taste sensillum to inositol,sucrose and glucose all increased significantly with concentration (for each concentration–response curve, P<0.05; one-way RM-ANOVA), but the magnitude of the response varied considerably across carbohydrates(Fig. 2C,D). The medial styloconic sensilla exhibited robust concentration-dependent increases in response to inositol, but only modest increases to glucose and sucrose. By contrast, the lateral styloconic sensilla exhibited robust concentration-dependent increases in response to inositol and sucrose, but a modest increase to glucose. The most notable difference between the lateral and medial styloconic sensilla concerned the sucrose response – e.g. 200 mmol l^–1 sucrose elicited an excitatory response of approximately 75 spikes s^–1 in the lateral, but only 8 spikes s^–1 in the medial styloconic sensillum.

According to Schoonhoven (Schoonhoven,1972), the lateral styloconic sensillum of M. sextacontains one taste cell that responds to sucrose and glucose, and another taste cell that responds to inositol. The medial styloconic sensillum, on the other hand, contains one taste cell that responds to glucose and another that responds to inositol. Here, we assessed this peripheral organizational plan.

We examined the neural response of the carbohydrate-sensitive taste cells in the lateral styloconic sensillum to sucrose (200 mmol l^–1), inositol (1 mmol l^–1) and sucrose+inositol. We found that the binary mixture elicited the same number of dominant spikes as sucrose, but significantly more dominant spikes than glucose (Fig. 3B). By contrast,the binary mixture elicited significantly more non-dominant spikes than sucrose or glucose alone (Fig. 3C). These results, together with analyses of the neural traces(e.g. see Fig. 3A) and temporal patterns of discharge (Fig. 7B,C), indicate (1) that sucrose caused one taste cell to generate a tonic discharge pattern, (2) that inositol caused one taste cell to generate a phasic-tonic discharge pattern, and (3) that the binary mixture of sucrose+inositol caused two taste cells to discharge out-of-phase with one another (one taste cell exhibited a tonic discharge pattern, and the other elicited a phasic-tonic discharge pattern). Based on these findings, we propose that sucrose and inositol each activated different carbohydrate-sensitive taste cells within the lateral styloconic sensillum.

If the two carbohydrate-sensitive taste cells functioned independently of one another, then the binary mixture of sucrose+inositol should have elicited∼70 dominant spikes s^–1 and ∼45 non-dominant spikes s^–1 (Fig. 3B,C). This was not the case, however. We measured ∼70 dominant spikes s^–1, but ∼25 non-dominant spikes s^–1. The low number of non-dominant spikes indicates that inhibitory interactions occurred between the two carbohydrate-sensitive taste cells.

Next, we examined the neural response of the carbohydrate-sensitive taste cells in the lateral styloconic sensillum to sucrose (200 mmol l^–1), glucose (1 mmol l^–1) and sucrose+glucose. We found that the binary mixture elicited the same number of dominant spikes as sucrose, but significantly more than glucose(Fig. 4B). By contrast, the binary mixture elicited significantly more non-dominant spikes than did either component alone (Fig. 4C). These findings, together with analyses of the neural traces (e.g. see Fig. 4A) and temporal patterns of discharge (Fig. 7B,C),indicate (1) that sucrose caused one taste cell to generate a tonic discharge pattern, (2) that glucose caused one taste cell to generate a phasic-tonic discharge pattern, and (3) that the binary mixture of both carbohydrates caused two taste cells to discharge out-of-phase with one another (one taste cell exhibited a tonic discharge pattern, and another a phasic-tonic discharge pattern). Based on these findings, we propose that sucrose and glucose each activated different carbohydrate-sensitive taste cells in the lateral styloconic sensillum.

If the carbohydrate-sensitive taste cells functioned independently of one another, then the binary mixture of sucrose+glucose should have elicited∼70 dominant and ∼40 non-dominant spikes s^–1(Fig. 4B,C). Our results roughly match this prediction, indicating that the two carbohydrate-sensitive taste cells functioned independently of one another when stimulated by the mixture of sucrose and glucose.

We examined the neural response of the carbohydrate-sensitive taste cells in the lateral styloconic sensillum to glucose (200 mmol l^–1), inositol (1 mmol l^–1) and glucose+inositol. We found that the binary mixture elicited significantly more dominant spikes than did glucose or inositol alone(Fig. 5B). By contrast, the binary mixture elicited the same number of non-dominant spikes as glucose or inositol alone (Fig. 5C). These results, together with analyses of the neural traces(Fig. 5A) and temporal patterns of discharge (Fig. 7B),indicate that the glucose, inositol and glucose+inositol solutions each caused a single taste cell to generate similar phasic-tonic discharge patterns. The only difference between these discharge patterns was that the binary mixture elicited a higher rate of firing across the entire 1000 ms of response. Based on these findings, we infer that inositol, glucose and the binary mixture each activated the same carbohydrate-sensitive taste cell in the lateral styloconic sensillum.

If glucose and inositol each acted independently on the same carbohydrate-sensitive taste cell, then the binary mixture of glucose+inositol should have elicited ∼85 dominant spikes s^–1(Fig. 5B). Instead, we observed∼65 dominant spikes s^–1. This indicates that inhibitory interactions occurred between glucose and inositol.

We also examined the neural response of the carbohydrate-sensitive taste cells in the medial styloconic sensillum to glucose (200 mmol l^–1), inositol (1 mmol l^–1) and glucose+inositol. We found that the glucose+inositol solution elicited significantly more dominant spikes than did the glucose or inositol solution(Fig. 6B). Further, the glucose, inositol and glucose+inositol solutions each elicited a low and statistically equivalent number of non-dominant spikes(Fig. 6C). These findings,together with analyses of the neural traces(Fig. 6A) and temporal patterns of discharge (Fig. 7A),indicate that the glucose, inositol and glucose+inositol solutions each caused a single taste cell to generate a phasic-tonic discharge pattern. Based on these findings, we propose that inositol, glucose and the binary mixture each activated the same carbohydrate-sensitive taste cell in the medial styloconic sensillum.

If glucose and inositol each acted independently on the same carbohydrate-sensitive taste cell, then the binary mixture of glucose+inositol should have elicited ∼105 dominant spikes s^–1(Fig. 6B). This was approximately what we observed, indicating that no inhibitory interactions occurred between glucose and inositol in the medial styloconic sensilla.

We asked whether sucrose stimulates the same carbohydrate-sensitive taste cell as glucose or inositol in the medial styloconic sensillum. We could not answer this question, however, because the neural response to 200 mmol l^–1 sucrose was indistinguishable from that to the electrolyte solution.

Taken together, these results show that the carbohydrate solutions (i.e. sucrose, glucose, inositol and the three binary mixtures) activated different populations of carbohydrate-sensitive taste cells. Whereas sucrose activated one class of taste cell in the lateral styloconic sensillum, glucose+inositol both activated another taste cell in both the lateral and medial styloconic sensillum.

In many species of mammal (Spector,2000) and insect (Ma,1972; Blaney, 1975; Dethier, 1976; Scheiner et al., 2001; Omura and Honda, 2003; Wang et al., 2004; Sasaki and Asaoka, 2006), the presence of single carbohydrates elicits an immediate concentration-dependent increase in feeding responsiveness. Previous work in M. sextarevealed a paradoxical effect of carbohydrates on feeding, however. Although carbohdyrates appear to have a limited (or non-existent) effect on immediate biting responses (Bowdan,1995; Glendinning et al.,2000), they nevertheless stimulate intake in long-term feeding studies (Yamamoto and Fraenkel,1960; Städler and Hanson,1978; Glendinning et al.,2000).

Here, we addressed three inter-related issues. First, we attempted to explain the discrepancy between the results from short- and long-term feeding tests. To this end, we tested the hypothesis [derived from a study of the terrestrial slug, Limax maximus(Reingold and Gelperin, 1980)]that carbohydrates stimulate intake by causing M. sexta to bite for more extended periods of time, without increasing biting rate or bite size. We defined feeding stimulation as biting activity greater than that elicited by water alone. Second, we compared the ability of different carbohydrates (alone or in binary mixture) to stimulate feeding. Third, we tested two hypotheses to explain why some carbohydrate solutions stimulated larger meals. If meal size is determined by negative feedback from the midgut or hemolymph [owing to the accumulation of nutrients or osmotically active compounds(Simpson and Raubenheimer,1993; Timmins and Reynolds,1992)], then meal size should decrease with increasing molar density of nutrients in the ingested food. Alternatively, if meal size is determined by excitatory input from carbohydrate-sensitive taste cells, then meal size should increase with (a) the total number of spikes elicited across all carbohydrate-sensitive taste cells, or (b) the number of different classes of carbohydrate-sensitive taste cell that are activated.

When presented with a water-treated disk, the caterpillars took approximately 12 s to initiate biting, but when presented with any of the carbohydrate-treated disks, they took significantly less time (2–3 s) to initiate biting (K–W statistic=26.0; P⩽0.05; Fig. 8A). All of the carbohydrate solutions, however, were equally effective at reducing the latency to initiate biting.

Once biting commenced, the caterpillars exhibited the same number of bites on the water-treated disk as on the carbohydrate-treated disks during the initial 10 s (K–W statistic=5.2; P>0.05) and 120 s(K–W statistic=4.3; P>0.05) of the meal(Fig. 8B,C). This shows that carbohydrates did not modulate initial biting rate. To examine more protracted biting responses, we examined three intake parameters across the first meal:mean bite size, total number of bites and meal duration. There were no significant differences in mean bite size (K–W statistic=3.6; P>0.05; Fig. 9A)across the carbohydrate solutions, but there were large and significant differences in total number of bites (K–W statistic=27.8; P⩽0.05; Fig. 9B)and meal duration (K–W statistic=26.4; P⩽0.05; Fig. 9C). A post-hoc analysis revealed that despite a trend for all carbohydrate solutions to stimulate feeding, only two carbohydrate solutions, sucrose+glucose and sucrose+inositol, caused the caterpillars to take significantly larger meals than water alone. These findings show that specific mixtures of carbohydrates are required to stimulate feeding, and that these mixtures do so by increasing the amount of time spent biting.

To explain the large variation in meal size in Fig. 9, we tested two hypotheses. The first was that meal size is determined by negative feedback from the gut or hemolymph. Accordingly, meal size should vary inversely with the molar density of nutrients in the ingested food. To test this hypothesis,we regressed meal size (i.e. mean number of bites emitted over meal; Fig. 9B) on the molar density of carbohydrates in the test solutions(Table 1). This regression was not significant (slope=19.0; r²=0.06; F_1,4=0.27; P>0.05), indicating that meal size varied independently of the molar density of accumulated carbohydrates in the gut and/or hemolymph.

This second hypothesis was that excitatory input from the carbohydrate-sensitive taste cells determined meal size. We examined two possible mechanisms by which this could occur. One is that the insect's brain monitors excitatory input from all carbohydrate-sensitive taste cells, and increases meal size in direct proportion to the algebraic sum of this afferent input. To test this possibility, we regressed meal size on the total number of spikes s^–1 elicited by each carbohydrate solution(Table 1). This regression was not significant (slope=0.3; r²=0.003; F_1,4=0.02; P>0.05), indicating that meal size varied independently of total excitatory input. A second mechanism is that the insect's brain monitors the pattern of activity across the three classes of carbohydrate-sensitive taste cell (rather than total excitatory input per se), and generates the largest meal when all three classes of carbohydrate-sensitive taste cell are activated strongly. We found clear support for this latter mechanism. Even though the glucose+inositol solution elicited a high total firing rate (166 spikes s^–1), it activated only two carbohydrate-sensitive taste cells and failed to generate meals larger than those on water alone(Table 1, Fig. 9B). By contrast, the two other binary mixtures (sucrose+inositol and sucrose+glucose) each elicited an intermediate firing rate (i.e. 126–154 spikes s^–1), but nevertheless activated three carbohydrate-sensitive taste cells and caused the caterpillars to take meals significantly larger than those on water alone. In fact, the meals on the sucrose+inositol disks were significantly larger than those on the glucose+inositol disks.

The peripheral taste system of M. sexta responded vigorously to three carbohydrates (sucrose, glucose and inositol) and weakly to one carbohydrate (fructose); it failed to respond altogether to three others(maltose, trehalose and polycose). Because we used high concentrations of each carbohydrate, it is unlikely that the weak responses to some carbohydrates reflect the use of sub- or peri-threshold concentrations. This finding agrees with prior reports (Schoonhoven,1972; Glendinning et al.,2000), and confirms that the taste system of M. sextadetects a limited number of carbohydrates, as compared with other species of Lepidopteran (Schoonhoven and van Loon,2002) and Dipteran (Dethier,1976). It is notable that the taste system of M. sextaalso responds to a limited number of bitter taste stimuli(Wrubel and Bernays, 1990)(J.I.G., unpublished data), amino acids(Glendinning et al., 2000) and acidic compounds (Bernays et al.,1998). Comparative studies are needed to explain why M. sexta evolved such an insensate taste system.

Although several reports have shown that taste cells within the lateral and/or medial sensilla of M. sexta exhibit excitatory responses to sucrose, glucose and/or inositol (Frazier,1986; Lam and Frazier,1991; Gothilf and Hanson,1994; Glendinning et al.,2000; Schoonhoven and van Loon, 2002), we are aware of only one systematic attempt to determine the response properties of each carbohydrate-sensitive taste cell– it was performed by Schoonhoven and Dethier(Schoonhoven and Dethier,1966). These authors examined representative neural responses of the lateral and medial styloconic sensilla to binary mixtures of carbohydrates. For instance, they show a response of the lateral (their fig. 4, trace 12) and medial(their fig. 3, trace 9)styloconic sensilla to a binary mixture of glucose (100 mmol l^–1) and inositol (100 mmol l^–1); the traces appear to illustrate a single taste cell firing rapidly. In addition, the authors show a response of the lateral styloconic sensilla to a binary mixture of sucrose (100 mmol l^–1) and glucose (100 mmol l^–1) (their fig. 3, trace 11); the trace appears to illustrate two taste cells firing out-of-phase with one another. These findings corroborate the more extensive results presented herein, and lead us to conclude that (1) each lateral styloconic sensillum contains one carbohydrate-sensitive taste cell that responds to sucrose and another that responds to glucose and inositol;whereas (2) each medial styloconic sensilla contains a single carbohydrate-sensitive taste cell, which responds to glucose and inositol.

In a comprehensive review of caterpillar taste, Schoonhoven and van Loon reported that many species possess taste cells that respond exclusively to sucrose or inositol, but they did not cite any species with taste cells that respond both to glucose and inositol(Schoonhoven and van Loon,2002). It would appear, therefore, that the presence of a sucrose-sensitive taste cell is common among caterpillars, but that the presence of glucose/inositol-sensitive taste cells is unusual.

Our behavioral studies offer several insights into how gustatory input from the carbohydrate-sensitive taste cells modulate feeding in M. sexta. First, the fact that all of the carbohydrate solutions significantly reduced the latency to initiate biting (as compared with water) demonstrates that carbohydrates can incite feeding. It is notable, however, that there were no significant differences in latency to initiate biting across the different carbohydrate solutions. This may reflect the fact that we used relatively high concentrations of most carbohydrates (e.g. 200 mmol l^–1glucose). In support of this possibility, we reported previously that a lower concentration of glucose (100 mmol l^–1) did not reduce the latency to initiate biting (Glendinning et al., 2000). Given that 200 mmol l^–1 glucose generates a substantially stronger peripheral gustatory response than 100 mmol l^–1 glucose (Fig. 2), it would appear that there is a threshold level of input from the carbohydrate-sensitive taste cells required to incite feeding, and that input from 200 mmol l^–1 glucose (and the other carbohydrate solutions tested herein) surpassed this threshold.

In many species of insect, the taste of sugars stimulates concentration-dependent increases in feeding. For instance, increasing the concentration of sucrose will cause a proportional increase in (1) the strength of the swallowing response in Bombyx mori caterpillars(Sasaki and Asaoka, 2006); (2)the rate of biting in Pieris brassicae caterpillars(Ma, 1972); and (3) the probability that flies and bees will extend their proboscis(Dethier, 1976; Scheiner et al., 2001) and butterflies will initiate feeding (Omura and Honda, 2003). Based on these findings, it is surprising that M. sexta caterpillars emitted the same initial rate of biting on disks treated with water as on disks treated with high concentrations of carbohydrates, both here and elsewhere(Bowdan, 1995; Glendinning et al., 2000). Even though Bowdan reported that M. sexta caterpillars tend to bite more avidly from disks treated with 100 mmol l^–1 than 10 mmol l^–1 sucrose during the first chewing bout(Bowdan, 1995), the median number of bites on each type of disk did not differ significantly(Mann–Whitney U-test, P>0.05).

We found that only two of the carbohydrate solutions (sucrose+inositol and sucrose+glucose) significantly increased meal size, relative to water. To explain this observation, we asked whether meal lengths on the different carbohydrate solutions correlated with differences in the (1) amount of postingestive feedback, (2) magnitude of the total peripheral response, or (3)number of activated carbohydrate-sensitive taste cells. Our analysis indicated that meal length correlated with differences in the number of activated carbohydrate-sensitive taste cells, but not with the other two measures. This leads us to propose that meal length on the carbohydrate-treated disks was not modulated by the algebraic sum of inputs from the carbohydrate-sensitive taste cells (Schoonhoven and Blom,1988) or the accumulation of carbohydrates in the mid-gut or hemolymph (Timmins and Reynolds,1992; Simpson and Raubenheimer, 1993).

Insects use several coding mechanisms to identify and discriminate taste stimuli, including labeled-line (Marella et al., 2006), temporal(Glendinning et al., 2006) and ensemble (Dethier and Crnjar,1982; Glendinning et al.,2002) codes. Our findings indicate that M. sextaidentifies preferred foods based on an ensemble code – i.e. the pattern of activity generated across the entire population of carbohydrate-sensitive taste cells. Accordingly, the ensemble code for a preferred carbohydrate-rich food would be strong and simultaneous activation of all three classes of carbohydrate-sensitive taste cell. A crucial prediction of this coding mechanism is that those carbohydrate solutions that activate only one or two pairs of carbohydrate-sensitive taste should fail to stimulate feeding,irrespective of the magnitude of the peripheral taste response. We found clear support for this prediction. For instance, even though the mixture of glucose+inositol elicited a vigorous peripheral gustatory response (i.e. 166 spikes s^–1), the spikes were limited to two classes of carbohydrate-sensitive taste cell and the caterpillars failed to exhibit significant feeding stimulation (Table 1, Fig. 9). However, the mixture of sucrose+inositol elicited a comparable peripheral response (i.e. 154 spikes s^–1), but activated all three classes of carbohydrate-sensitive taste cell and stimulated significantly larger meals than water alone.

There are three interpretive limitations of our study. The first is that we are relating taste cell responses that span 1 s to biting responses that span several minutes. As we illustrate in Fig. 7, adaptation processes can cause large reductions in the absolute firing rates of taste cells over periods as short as 1 s. Before we can address this caveat, however, we need to determine the temporal pattern of stimulation that taste cells experience over the course of the meal. This is because caterpillars repeatedly retract their styloconic sensilla from the surface of food items as they are feeding(Devitt and Smith, 1985). Whenever a sensillum is retracted, the taste cells contained within it should experience partial or complete disadaptation(Blaney, 1975). If so, then our 1 s neural responses would accurately reflect what caterpillars experienced over the course of a meal.

The second interpretive limitation is that we focused on the first meal,and ignored consumption across successive meals. We did so to maximize our chances of measuring taste-mediated feeding responses. It is notable, however,that both sucrose and inositol alone have been reported to stimulate feeding in M. sexta during feeding tests that span several hours(Yamamoto and Fraenkel, 1960; Städler and Hanson, 1978; Glendinning et al., 2000). We can propose two non-mutually exclusive explanations for the discrepancy between the results of these protracted feeding tests and the one presented herein. First, in the present study, there was a trend for sucrose and inositol to stimulate feeding, but the difference was not significant. It is possible that the feeding stimulation would have become significant had we performed the test over a longer period of time. Second, it is possible that post-oral response mechanisms indirectly stimulate feeding during long-term intake tests. This possibility is based on the observation that the positive postingestive actions of carbohydrates alone can stimulate robust intake in mammals (Sclafani, 2001; Sclafani and Glendinning,2005).

The third interpretive limitation of our study is that the sucrose+glucose mixture contained both 200 mmol l^–1 sucrose and 200 mmol l^–1 glucose. Thus, it is possible that the peripheral gustatory response to this binary mixture differed from that to each carbohydrate alone simply because it contained 400 mmol l^–1of total sugar. This possibility is contradicted, however, by the observation that the neural response of the lateral styloconic sensillum to 200 mmol l^–1 sucrose (or glucose) is indistinguishable from that to 400 mmol l^–1 sucrose (or glucose) (J.I.G., unpublished data). That is, each carbohydrate strongly activates one dominant taste cell at both the 200 and 400 mmol l^–1 concentrations.

Previous studies have reported that when insect taste sensilla are stimulated with taste stimulus mixtures, the discharge rate is often less than the sum of the discharge rates from the individual component stimuli. These inhibitory interactions involve stimulus mixtures that activate different taste cells within the same sensillum(Ishikawa, 1967; Schoonhoven, 1978; Mitchell, 1987; Blaney and Simmonds, 1990; White et al., 1990; Chapman et al., 1991; Schoonhoven et al., 1992; Shields and Mitchell, 1995; Bernays and Chapman, 2000; Glendinning et al., 2000; Bernays and Chapman, 2001) or different signaling pathways within the same taste cell(Glendinning and Hills, 1997). Here, we report similar findings. For example, despite activating different taste cells within the lateral sensillum, one of the binary mixtures (i.e. inositol+sucrose) elicited fewer non-dominant spikes than would have been predicted based on the firing rate elicited by the individual component stimuli (Fig. 3). This indicates that inhibitory interactions occurred between the two carbohydrate-sensitive taste cells. Likewise, despite activating the same taste cell, the glucose+inositol mixture elicited fewer dominant spikes in the lateral styloconic sensillum than expected based on the response to each carbohydrate alone (Fig. 5). This indicates that the inhibitory interactions also occurred within the same taste cell.

The mechanistic bases for inhibitory mixture interactions in the lateral sensillum are unclear. For these interactions to occur between taste cells,there could be direct electrical communication via gap junctions(Steinbrecht, 1989; Isidoro et al., 1993) or ephaptic interactions among adjacent taste cells(Jefferys, 1995; Bokil et al., 2001). For inhibitory interactions to occur between taste stimuli that activate the same taste cell, there could be antagonistic interactions at the cell-surface receptor. Alternatively, if the two carbohydrates activate different signaling pathways, then activation of one pathway could inhibit the response of the other (Dethier and Bowdan,1989).

There are at least two reasons why M. sexta should benefit from feeding selectively on leaves containing mixtures of sucrose+inositol (or sucrose+glucose). Whereas many plant tissues contain relatively high concentrations of glucose and low concentrations of sucrose(Somogyi and Trautner, 1974),solanacous plants contain roughly equal concentrations of glucose and sucrose,and relatively high concentrations of inositol(Nelson and Bernays, 1998). Thus, M. sexta could use the simultaneous activation of all three classes of carbohydrate-sensitive taste cell as a sign stimulus for its solanacous host plants. A second reason is that high carbohydrate consumption increases fat accumulation in both the larval(Thompson et al., 2003) and adult (Ojeda-Avila et al.,2003) stages. This extra fat would help pupae undergo successful metamorphosis and adults fuel flight when nectar reserves are limited(Ziegler and Schultz, 1986). Given that carbohydrate levels vary greatly among leaves within solanacous plants (Nelson and Bernays,1998), any gustatory mechanism that helps M. sextaidentify leaves with relatively high concentrations of several carbohydrates(i.e. sucrose, glucose and inositol) should promote fat accumulation, and thus be adaptive.

We found that a specific ensemble code from the peripheral taste system(i.e. activation of all three classes of carbohydrate-sensitive taste cell)was necessary to stimulate feeding in M. sexta. This ensemble code increased the duration of biting, but had no effect on bite size or the rate at which bites were emitted. Further work is needed to determine the mechanisms by which the ensemble code increases meal length. They could involve raising the insect's central excitatory state at the onset of the meal, or delaying activation of the mechanisms that terminate feeding.

Additional work should attempt to identify other ecologically relevant situations that enhance behavioral responsiveness to carbohydrates. For instance, synergistic interactions could occur between gustatory input from carbohydrates and host plant chemicals. This is based on the observation that rearing M. sexta on their host plants (e.g. tomato, potato or tobacco leaves) induces a strong preference for leaf disks treated with host plant extracts (Hanson and Dethier,1973; de Boer,1993; del Campo et al.,2001; del Campo and Miles,2003; Haribal et al.,2006). Another possibility is that taste-mediated responsiveness to carbohydrates increases following starvation or prolonged consumption of a carbohydrate-deficient diet. If so, then this elevated responsiveness would facilitate the re-establishment of normal blood sugar levels. The existence of such a mechanism is supported by the observation that maintaining M. sexta on a carbohydrate-deficient diet for 24 h causes them to prefer a carbohydrate-rich diet over a carbohydrate-deficient diet(Thompson and Redak, 2000). This deprivation-induced preference for carbohydrates could be mediated by enhanced peripheral (Simpson et al.,1990; Simmonds et al.,1992) or central taste responsiveness.

We thank James Frazier and two reviewers for valuable editorial comments. This project was supported in part by research grant number 5 R29 DC 02416 from the National Institute on Deafness and Other Communication Disorders,National Institutes of Health (to J.I.G.), and a grant from the Howard Hughes Medical Institute (to Barnard College).