Seeds of Gossypium hirsutum L., variety STAM 59A, which is commonly cultivated in Africa, were provided by CIRAD (French agricultural research and cooperation organization, France). We soaked the seeds in water at 24°C in the dark overnight and then sowed them individually in plastic pots (h: 8.5 cm, d: 6 cm), filled with soil (universal potting soil, Ricoter, Switzerland). Seedlings were kept in phytotrons (GroBanks CLF Plant Climatics, Germany, 16 h light, 28°C, 65 μmol m⁻² s⁻¹ and 8 h dark, 25°C) until they were used for experiments, at the age of 3 weeks. They were watered every 3 days and fertilized with 30 mL of fertilizer (3.3 mL/L, Capito, 7-3-6 NPK, with oligo-elements) once a week. The day before the first damage treatment, we moved the plants to an experimental room under LED lights (SLT-EDK 6400K, Edkfarm, China, 200 μmol m⁻² s⁻¹, 16 h:8 h light/dark, 26 ± 3°C) where they were kept until measurements were made. To minimize potential signaling through induced volatile emissions plants with the different treatments were separated from each other by ~1 m. Their position was changed every day.

Spodoptera exigua (Hübner; Lepidoptera; Noctuidae) is a cosmopolitan generalist pest that feeds on many crops, including cotton. Its rearing and collection of regurgitant were performed as described by Arce et al. (2021) and according to Turlings et al. (1993). Caterpillars were reared from eggs supplied by Entocare (Wageningen, Netherlands), on wheat germ-based artificial diet (Frontier Scientific Services, Newark, United States). Regurgitant was collected from third to fourth instar larvae, that had been fed on cotton leaves for 24 h. We triggered regurgitation by gently pressing the caterpillars’ body toward the head. The regurgitant was collected with a pipette and stored in Eppendorfs protected from light and kept on ice. After each collection, Eppendorfs with regurgitant were kept at −80°C until use.

To dissociate the effect of physical damage and the effect of S. exigua regurgitant, we simulated chewing herbivory by mechanically damaging cotton seedlings without and with applying regurgitant to the wounds. Controls consisted of undamaged plants. Seedlings used had two true leaves and were starting to develop the third one. The earliest stage used were plants whose third leaves just started to emerge (Supplementary Figure S1A) and the latest stage were plants with small third leaves that started to expand, but had not flattened yet (Supplementary Figure S1B). We used the same procedure and timing of damage that was used by Arce et al. (2021) for their analysis of systemically induced volatile emissions. First and second leaves were damaged with serrated forceps a total of four times, over two consecutive days (mornings and evenings). Each damage event consisted of applying the forceps three times with strong manual pressure on each side of a leaf (~1 cm² per leaf side). Right after damaging the leaves, 10 μL of regurgitant were spread on the wounded surfaces (~5 μL per leaf side). A different batch of regurgitant was used every one to two damage events. We started damaging the apical part of the first leaf (Figure 1A), then its more basal part. The second leaf was damaged in the same way on the second day (Figure 1B). Seven days after damage began, the third leaves were collected for measurements and for choice feeding assays.

Third leaves (n = 18 per damage treatment) were collected and their abaxial surface was directly photographed. Pictures were used to count glands manually using the Multipoint tool of the Fiji distribution of ImageJ software (Schindelin et al., 2012). We counted all the glands present in a surface of 1 × 1 cm selected in the upper center of the leaves. Leaves photographed for gland density measurement were then flash-frozen for subsequent terpenes analysis or used for the first experimental repetition of the choice feeding assays.

We measured gland area (n = 11–12 per damage treatment) to estimate gland size, as we could not measure their volume. Gland size was used as a proxy to estimate overall gland filling. For this, pictures of the central part of the abaxial surface were taken under a stereoscope with a magnification of 6.3×. These pictures were taken after the leaves were used for the third replication of the feeding choice assays, were caterpillars ate a maximum of 4.3% of the leaves. This left sufficient space to measure gland size. Using Fiji, pictures were converted to 8-bit. We then used the Threshold tool to select glands and pictures were converted to mask (Binary ➔ Convert to mask). Analyze Particles was used to measure simultaneously the area of each gland, excluding glands touching the edges of the images. This represented between 20 and 96 glands per leaf. Areas were averaged into a single value for each leaf.

Third leaves (n = 6 per damage treatment) were flash-frozen in liquid nitrogen and stored at −80°C until they were ground into powder.

We conducted dual choice feeding assays to investigate the relative defensive effects of induction by physical damage and elicitation by regurgitant. S. exigua larvae were offered pairs of leaves of the three possible combinations of damage treatment (n = 14–16 for each combination). Late third to early fourth instars were fed with cotton for 24 h and starved for 3 h before each choice assay. Leaves were placed in pairs in transparent Petri dishes of 14 cm diameter, with their petioles wrapped in wet cotton wool to prevent desiccation. Because gland-associated traits vary with leaf area (McAuslane et al., 1997; Opitz et al., 2008), leaves of similar size were paired (Supplementary Figure S2). We released a single larva in the center of each dish. Petri dishes were placed under artificial light in random position and orientation. The assay was replicated three times with n = 5–6 for the first and third repetition and n = 3–4 for the second one. Data of the three repetitions were analyzed together. Because of difference in feeding between the three repetitions, caterpillars fed for 10 h (10:30 to 20:30) the first repetition and for 24 h (12:00 to 12:00 + 1D) the second and third repetitions. Leaf surface consumed was analyzed from pictures using Fiji.

Statistical analyses were performed with R 4.2.2 (R Core Team, 2022). A threshold of α = 0.05 was used. We plotted predictions from the statistic models and their 95% confidence intervals with the function ggeffect() of the package ggeffects (v. 1.1.3, Lüdecke, 2018) and ggplot2 (v. 3.3.6, Wickham, 2016).

Both gland density and gland size depended on leaf area (Figure 2). While gland density decreased with leaf area, gland size increased. Gossypol leaf concentration decreased with leaf size, but there was no significant relationship for heliocides content (Supplementary Table S1). Similarly, volatile terpenes concentrations either decreased with leaf size or showed no significant correlation (Supplementary Table S1).

Damaged cotton seedlings increased their gland density and gland size, whether regurgitant was applied to their wounds or not (Figure 3). A week after damage began, the third leaves of mechanically damaged plants had on average 27% more glands per cm² than leaves of undamaged plants (Figure 3A). In addition, their glands were on average 41% larger (Figure 3B). For both traits, there was no detectable difference between plants that received damage only or damage with regurgitant application.

We quantified gossypol and heliocides. Their leaf concentration followed the same pattern as gland density and gland size: damaged plants accumulated more of these compounds than undamaged plants, regardless of regurgitant application (Figure 4; Supplementary Table S1). A week after the start of induction, the third leaves of mechanically damaged plants had more than doubled their gossypol content, with an average increase of 128%. Heliocides increased by 86%. There was again no observable difference in their concentrations between mechanical damage alone and mechanical damage with regurgitant application.

Ten monoterpenes and nine sesquiterpenes were quantified from leaf extracts (Supplementary Table S1). The leaf concentration of all monoterpenes and six sesquiterpenes was significantly affected by damage treatment (Supplementary Table S1). Several compounds showed similar patterns to that observed for the terpenoid aldehydes: leaf accumulation increased with wounding, with no distinct differences between physical damage only and damage with regurgitant. This was the case for the sesquiterpenes β-caryophyllene, humulene, β-elemene, α-guiaene and γ-bisabolene, and the monoterpene β-ocimene (Figure 4; Supplementary Table S1). β-ocimene levels were most affected by induction, with a nearly fourfold increase for both types of damage. The other monoterpenes also increased with mechanical damage (e.g., β-phellandrene) or tended to do so (e.g., α-pinene), but to a lesser extent than β-ocimene or the above mentioned sesquiterpenes (Figure 4; Supplementary Table S1). Unexpectedly, plants treated with regurgitant did not accumulate more of these monoterpenes: leaf concentrations were similar to those of undamaged plants and significantly lower than with mechanical damage alone. This was also the case of the sesquiterpene α-copaene (Figure 4; Supplementary Table S1).

Spodoptera exigua larvae were offered pairs of leaves of all three possible combinations of damage treatment. Larvae preferred leaves of undamaged plants to those of induced plants, regardless of regurgitant application (Figure 5). However, their preference for uninduced leaves was stronger when the choice was with leaves from plants treated with regurgitant than with leaves from plants that were damaged only (Supplementary Figure S3). Larvae did not distinguish between leaves of plants mechanically damaged only and those damaged with regurgitant application (Figure 5).

Cotton gland density is highly dependent on leaf area, or age: the larger a leaf grows, the more space there is between the glands that it carries (McAuslane et al., 1997). Therefore, leaf area is an important variable to consider when studying the induction of glands in cotton. As expected, we observed a negative correlation between gland density and leaf area. In contrast, gland size had an opposite relationship, with larger leaves having larger glands to remove. Larger glands suggest that they were likely more filled with chemical content. Gland filling is thus a process occurring during leaf maturation. Mint, for instance, has glands that fill as the leaves age, with monoterpene synthesis peaking in middle-aged leaves (Gershenzon et al., 2000; Turner et al., 2000). Gland-stored compounds differed in their relationship to leaf area, indicating that they likely contributed to gland filling differently and that leaf chemical composition varied with leaf maturation. For instance, gossypol and α-pinene decreased with leaf area, whereas heliocides and β-ocimene stayed at similar levels. Eisenring et al. (2017) described that gossypol decreases with leaf aging, while heliocides remain constant, an effect hypothesized to be due to differences in compound stability and synthesis over time (Bell et al., 1978; Eisenring et al., 2017).

Cotton seedlings responded to repeated mechanical damage on their first and second leaves by accumulating more gossypol, heliocides and volatile terpenes in their developing undamaged third leaves. Volatile terpenes differed in their levels of induction, with β-ocimene showing the strongest increase, followed by some of the sesquiterpenes. These results are in accordance with those of Opitz et al. (2008). They described mechanical damage to increase foliar content of the gland-stored compounds in immature leaves, within the same time frame. They also observed a particularly strong induction of β-ocimene, as well as a stronger response from sesquiterpenes than from monoterpenes. Induction of gland-stored compounds is a result of both greater gland density and increased gland content (McAuslane et al., 1997; Opitz et al., 2008). We found that leaves of mechanically damaged plants had indeed higher gland density and they also had larger glands, suggesting that they were more filled with chemical content. The induction triggered by mechanical damage apparently deterred Spodoptera exigua larvae from feeding: they strongly preferred to feed on leaves of undamaged plants in the dual choice assays.

Adding S. exigua larvae regurgitant to artificial wounds did not elicit a stronger response than mere physical damage in terms of gland density, gland size and accumulation of gland-stored compounds. In accordance with this, caterpillars fed equally on leaves from plants induced with mechanical damage and leaves from damaged plants with addition of regurgitant. Keeping damage constant, our experiments strongly suggest that systemic induction of gland number and filling is not affected by regurgitant-derived factors. It adds evidence to the hypothesis that systemic induction of cotton glands is an unspecific response to physical damage, as proposed by Opitz et al. (2008).

Using a similar induction method, Arce et al. (2021) observed that the application of S. exigua regurgitant to repeatedly mechanically damaged plants induces a stronger systemic release of some volatile compounds than mechanical damage alone. This contrasts with what we observed for gland induction and implies that involvement of regurgitant-derived HAMPs in eliciting cotton defenses is trait-dependent. Glands provide cotton with a broad-spectrum direct defense: they are not only effective against various insects, but also against pathogens and non-ruminant animals (Hagenbucher et al., 2013). Therefore, the non-specificity in regulating their induction may be adaptive. On the other hand, systemic volatile emissions are involved in the more specific interactions with natural enemies of herbivores and can act as signals between or within plants. In those cases, specificity may be necessary to convey reliable information (Turlings and Erb, 2018). Traits-dependency of the elicitation by HAMPS likely explains why S. exigua had a stronger preference for uninduced leaves when the choice involved leaves from plants treated with regurgitant than when the assay was done with leaves from plants that were only damaged. They were probably deterred by other defenses, such as volatile emissions that are more strongly induced when regurgitant is applied. Indeed, it is known that S. exigua prefers to feed on uninduced glandless plants to induced ones, indicating an effect of other inducible traits than glands on their choice (McAuslane and Alborn, 1998).

Application of regurgitant triggered a reduction in the leaf concentration of some volatile terpenes, to levels similar to undamaged plants. It happened despite a higher gland density than undamaged plants, implying that the additional glands that were formed contained a different blend of compounds. Insect regurgitant can also contain effectors, that inhibit the induction of plant defenses (Acevedo et al., 2015). However, the fact that this suppression is restricted to only some of the gland-stored compounds rather suggests that the differences are due to the reallocation of the substrates used in their synthesis to other inducible traits, such as the systematically emitted VOCs.

The strength of induction by mechanical damage and regurgitant exposure and their relative importance likely depends on several factors. We measured induced traits at a single time point, in a specific leaf and at a specific plant stage. However, there is likely temporal variation in the induction of plant defenses. As mentioned, accumulation of gland-stored compounds depends on leaf development (McAuslane et al., 1997; Eisenring et al., 2017) and plant age is another ontogenic factor to take in account (Anderson et al., 2001; Boege and Marquis, 2005; Quintero and Bowers, 2011). There is also spatial heterogeneity within the plants: the levels of cotton glands induction depend on the position of a leaf relative to the damage (Anderson and Agrell, 2005; Eisenring et al., 2017). Hence, measurements in other leaves, other time points and other plant developmental stages could lead to different outcomes for both the effects of damage alone and addition of regurgitant.

Also, induction of cotton glands depends on damage intensity (Agrawal and Karban, 2000; Opitz et al., 2008; Eisenring et al., 2017). Eisenring et al. (2017) described that in cotton seedlings gland density and terpenoid aldehydes content strongly increase with the surface damaged by herbivores over 2 days, before reaching a plateau around 2.5 cm². The area that we damaged in this study was much higher, 8 cm² in total. This might have triggered induction to its full strength. It is possible that elicitation by HAMPs may be visible only if the surface damaged is not high enough to trigger a full response. Another point that should be considered is how resources availability affects a trade-off between allocation to growth vs. defense (Züst and Agrawal, 2017). Increasing nitrogen for instance diminishes the effect of S. exigua herbivory on terpenoid aldehyde production (Chen et al., 2008). Thus, nutrient availability likely impacts how cotton plants integrate both damage and insect-derived cues in their response.

The characteristics of cotton glands make Gossypium hirsutum a great model to teach the chemical ecology of plant defenses against insect herbivores. Cotton glands are a nice example of a constitutive as well as inducible direct defense. The unspecific response of cotton seedlings to physical damage allows to work without the need of having insects available. The timing of the response is ideal for working with classes, as a week can separate damage treatment and measurements. Gland density and gland size can be easily measured, without requirement of chemical analyses. On the other hand, the quantification of terpenoid aldehydes and volatile terpenes can be carried out with more advanced students to introduce HPLC-DAD and GC-MS as analytical tools. The possibility to perform a simple bioassay is also interesting. Choice assays can be readily performed with generalist caterpillars such as S. exigua and Spodoptera littoralis, if available. The above-mentioned factors that could influence induction should be taken in account. With low nitrogen availability, induction could be stronger (Chen et al., 2008), but we advise to use sufficient fertilization to ensure proper plant growth. We indeed observed that undamaged plants had bigger third leaves than damaged plants under a lower fertilization regime (data not shown). This effect on growth was not observed in the present study with well-fertilized plants. Having leaves of similar size makes the interpretation of the choice tests considerably easier. Indeed, if leaves of undamaged plants are bigger, their gland density is lower than damaged plants, making it difficult to disentangle the effect of leaf size and induction on caterpillar feeding behavior.