Facing herbivory on the climb up: Lost opportunities as the main cost of herbivory in the wild yam Dioscorea praehensilis

Abstract Plants with simple architecture and strong constraints on their growth may offer critical insights into how growth strategies affect the tolerance of plants to herbivory. Although Dioscorea praehensilis, a wild yam of African forests, is perennial, both aerial apparatus and tuber are annually renewed. Each year, the tuber produces a single stem that climbs from the ground to the forest canopy. This stem bears no leaves and no branches until it reaches optimal light conditions. Once in the canopy, the plant's production fuels the filling of a new tuber before the plant dies back to the ground. We hypothesized that if deprived of ant defense, the leafless growth phase is a vulnerable part of the cycle, during which a small amount of herbivory entails a high cost in terms of loss of opportunity. We compared the growth of stems bearing ants or not as well as of intact stems and stems subjected to simulated or natural herbivory. Ants reduce herbivory; herbivory delays arrival to the canopy and shortens the season of production. Artificially prolonging the stem growth to the canopy increased plant mortality in the following year and, in surviving plants, reduced the stem diameter and likely the underground reserves produced. Tuber size is a key variable in plant performance as it affects both the size of the aerial apparatus and the duration of its single season of production. Aerial apparatus and tuber are thus locked into a cycle of reciprocal annual renewal. Costs due to loss of opportunity may play a major role in plant tolerance to herbivory, especially when architectural constraints interact with ecological conditions to shape the plant's growth strategy.


Monocotyledonous plants differ from most other vascular plants in
some basic features of anatomy and development (Tomlinson, 1995), such as the absence of a vascular cambium and thus of secondary growth. This trait constrains growth strategies, tolerance to herbivory and the evolution of defenses (Tomlinson, 1980). First, the absence of secondary vascular cambia severely limits the number of meristems and the repair of damaged tissues (Tomlinson, 1980). Because of the limitation of branching possibilities, many monocots are highly vulnerable to herbivores that destroy meristems. As an extreme example, palms incapable of branching due to their single apical meristem are completely intolerant of such herbivory (De Stevens & Putz, 1985;Tomlinson, 1980). Secondly, because monocots usually develop solely by primary thickening, the construction of any sizable stem requires a phase of initial thickening termed « establishment growth » (Holttum, 1955;Tomlinson, 1980;Tomlinson & Esler, 1973). In this study, we define as establishment growth the slow phase of initial stem thickening and as exploratory growth the subsequent phase of rapid height elongation.
We are unaware of any study that examines how all these traits combine to affect the tolerance of the plant to herbivory at different stages of development, and to shape its adaptations against herbivores. We undertook such a study on a monocotyledonous vine of African seasonal rainforests, the wild yam Dioscorea praehensilis Benth.
(Dioscoreaceae). We hypothesized that the growth strategy of this plant, shaped by ecological and developmental constraints, should lead to a weak point in the cycle, where a small amount of herbivory could entail a very high cost.
Dioscorea praehensilis is a perennial geophyte whose aerial apparatus and underground storage organ are both annually renewed. During the driest part of the year, the plant consists of an underground tuber, whose stored reserves fuel the growth of a single stem to the canopy when growth conditions become favorable. This stem usually does not branch and bears no foliage leaves, only reduced leaves termed cataphylls, until it reaches sunlit conditions. The plant then deploys a crown of leaves covering the canopy vegetation, sometimes situated at 30 m or more above the ground. Production by these leaves over several months then fuels the filling of a new tuber, replacing the plant's exhausted underground reserves. Finally, after the end of the long rainy season, the entire aerial system dies. This annual cycle of renewal of the vine's crown and of the stored underground reserves is the central feature of the morphology, physiology, and phenology of the plant.
A reduction in the efficiency of this cycle of reciprocal renewal could be costly to the plant. We hypothesized that rapid height growth during the climb to the canopy should be crucial to the plant's success, because photosynthetic production starts only after reaching the canopy and because the plant's aerial system has a limited lifespan.
Growth in height is maximized by the production of long internodes and by the absence of branching. The climbing phase to the canopy thus appears to be the most vulnerable to herbivory. Damage to the single apical meristem necessitates the activation of a lateral bud to take over vertical growth. To produce such a ramification could delay reaching the canopy, reduce the length of the single season available for production, and result in loss of opportunity for photosynthesis by the plant's entire crown. Should herbivory occur during this phase, the high cost in terms of lost opportunities should have long-term consequences. First, a smaller window of photosynthetic production could reduce the amount of reserves stored in the tuber. Second, this lowered storage could in turn reduce both the rate of growth to the canopy and the photosynthetic capacity (both the duration of the productive season and the photosynthetic surface that can be deployed) of the following year's shoot. Once the cycle is weakened, the plant could be caught in a downhill spiral.
The cost of herbivory during this phase of growth has two components, cost of lost material and cost of lost time. Because construction cost of the apical meristem is probably small relative to the amount of reserves stored in the tuber, the material cost of replacing the small amount of meristem tissue destroyed by herbivores is likely to be small. In contrast, the cost of meristem herbivory in terms of lost time for photosynthetic activity and replenishment of reserves is likely to be considerable and long lasting. Herbivory that delays arrival to the canopy, where the plant can accumulate enough reserves to repeat the trip in the following year, could greatly reduce the plant's fitness.
The height of the canopy and the amount of accumulated reserves in the tuber should thus be crucial variables affecting the overall success of the strategy. Failure to replenish the tuber reserves should result in increased mortality, decreased growth performance, or both, thus extending and even amplifying the effects of herbivory over time.
One of the traits of D. praehensilis supporting the hypothesis of a high cost of herbivory during the phase of rapid growth through the understory is the presence of both direct (saponins) and indirect antiherbivore defenses (biotic defense via extrafloral nectaries). The latter type of defense appears to be specific to this growth phase. The extrafloral nectaries, solely active on cataphylls near the apical meristem, attract more than 30 species of opportunistic ants. The effectiveness of this biotic defense varies from year to year due to the opportunistic nature of the protection mutualism and the efficiency of the anti-ant defenses of the principal herbivore in our study sites, Lilioceris latipennis Clark (Chrysomelidae: Criocerinae) (Di Giusto, Anstett, Dounias, & McKey, 2001). Indeed, larvae of this beetle subvert the plant's chemical defenses, incorporating them into a fecal shield in their own defense against ants (Di Giusto et al., 2001). The fact that a single specialist phytophagous insect strongly predominates among the herbivores we observed on the plant is consistent with well-developed defenses, likely driven by a high cost to the plant were herbivory to occur.
We studied the effect of herbivory on the time required by growing stems to reach the canopy. We compared the growth of intact stems with that of stems subjected to natural or simulated herbivory.
Simulation of herbivory allowed us to control sources of variation (in the timing, location, and severity of attack) in actual attacks. We also simulated the effects of a higher canopy by artificially prolonging the time required by stems to reach the canopy where they could begin to replenish stored reserves. Our observations and experiments were designed to address the following questions: (1) Does herbivory affect the time required for the stem apex to reach the canopy? (2) If so, do ants minimize the impacts of herbivory on the plant's growth? (3) Is the onset of production delayed and the productive period shortened after herbivory? (4) What is the relative importance of herbivory for establishment versus exploratory growth? (5) Does the quantity of reserves stored in the tuber affect the time required by the stem to reach the canopy? (6) How would a delay in reaching the canopy affect the replenishment of tuber reserves, probability of survival, and growth of surviving individuals in the following year?
Vegetation in this region is typically semideciduous forest (Letouzey, 1985). Annual rainfall is 1760 mm (average over 6 years from 1988 to 1993; data from the weather station of the logging company Société d'Exploitation Forestière des Bois du Cameroun (SEBC) located about 30 km from the study site). The climate is characterized by four distinct seasons; fieldwork was conducted between January and June, that is, from shortly before to shortly after the short rainy season (March to May) in 1999, 2000.
Dioscorea praehensilis grows in the forested zones of western and central Africa (Burkill, 1985;Hladik, Bahuchet, Ducatillon, & Hladik, 1984). The growth cycle of this geophyte has been described by McKey, Di Giusto, Pascal, Elias, andDounias (1998) andDi Giusto et al. (2001). The tuber head is perennial, but the bulk of the tuber (up to 28 kg fresh weight) is used up and replaced each year (B. Di Giusto and E. Dounias, unpublished data) (Figure 1a). Each year, to renew the plant's photosynthetic apparatus, reserves stored in an underground tuber fuel the growth of a single stem to the canopy. The plant is lightdemanding, and its crown of leaves is usually found in fully exposed canopy vegetation. Dioscorea praehensilis is not found in primary forest with very high canopies, where it is replaced by other yam species with perennial tubers and stems that are renewed at biennial or longer intervals, such as D. mangenotiana (Hladik et al., 1984).
At the end of the dry season (January-February), the plant consists solely of the tuber and some buds located at the perennial uppermost head of the tuber. In all 3 years of study, the first stems to emerge from the ground were observed in the first 2 weeks of January. Stem emergences peaked between 15 February and 15 March. Most plants had reached the canopy by the end of April, although a few stems were still emerging at this period.
During establishment growth, stems are rigid and self-supporting for the first 1.0-1.5 m of their length, vertical growth is slow, and internodes produced are short. This establishment growth provides the base required for exploratory growth. The latter is characterized by rapid vertical growth (mean 22.4 cm/day in the absence of herbivory; see below), less robust viny stems, long internodes, and the use of the surrounding vegetation to climb to the canopy (Figure 1b,c). In our study site, the canopy was situated at about 12-15 m. Plants required on average more than 2.5 months to reach the canopy. Once the plant's cohort of leaves has been put in place (late April to early May), the flow of energy is reversed to replenish a new tuber. Flowering and fruiting occur in July and August, after which the aerial shoot gradually withers (over a period of 2 months) and dies by the end of October (B. Di Giusto and E. Dounias, unpublished data).
During its growth through the understory, the apex is frequently attacked by phytophagous insects (Di Giusto et al., 2001). These include grasshoppers (Acrididae) and lepidopteran larvae, but the most frequent herbivore in our study site was the leaf beetle L. latipennis.
Larvae and adults feed on the tender young stem and cataphylls, often killing the apical meristem ( Figure 1d). Adult beetles first appear within 10-15 days after the first rains of the short rainy season (mid-March), and continue reproductive activity over the entire period of stem emergence. Duration of larval development varies between 1 and 2 weeks from egg hatching to pupation when conditions are favorable (Di Giusto, 2002). An imago can repeatedly lay eggs on the growing area of the same stem or other individuals may add their clutches of eggs to a pre-existing clutch (B. Di Giusto, unpublished data).  We randomly divided 74 plants into three groups, all placed on stakes (2.5 m-high supports that maintained stem apices at eye level) to facilitate observations. All stems were maintained in near-vertical position to avoid stems from bending or breaking. A first group of controls (N = 25) was naturally patrolled by ants. A second group ("ant- The ant-exclusion treatment allowed an estimate of the effect of ant patrols on the probability of herbivory and, when herbivory did occur, its effect on the growth of stems. The simulated herbivory treatment allowed us to increase the total sample size of attacked stems. Standardized treatment also allowed for elimination of extraneous sources of variation that occur when real herbivory intervenes (Tiffin & Inouye, 2000), due to factors such as (in D. praehensilis) the type of insect attacking, the precise site of the attack and how much time the insect spent consuming tissues before fully severing the apex.
Because of the difficulty of estimating the exact quantity of tissue consumed by insects, we recorded only two states for the apices, "cut" versus "not cut". Attacks that cut the apex always entailed destruction or necrosis of the apical meristem and a section, usually about 20 cm in length, of the apical portion of the stem. Our method does not take into account minor attacks that did not kill the apical meristem, but often caused a slight and temporary reduction of growth. In general, the greater the diameter of the stem, the greater the likelihood that the apical meristem will survive an attack (B. Di Giusto, unpublished data).

| Global analysis of stem growth over 20 days
We used a General Linear Model (R Core Development Team 2013) to estimate the effects of natural herbivory, simulated herbivory, and the absence of patrolling ants on stem growth. We assessed the effects of various explanatory variables [simulation of herbivory ("cut" vs. "not cut", number of times the apex was cut by natural herbivory (0 = intact, 1 = cut once, 2 = cut twice), the presence of ants ("antexclusion" vs. "ant presence"), the presence of larvae ("absence" vs. "presence"), cross-sectional stem area (at 1 m above ground), and initial length of the plant on the total stem length produced in 20 days.
The presence of larvae (main source of natural herbivory) at any point during the experiment was checked every other day on all stems. The presence of ants was similarly checked on both control and clipped stems; the absence of ants was confirmed on ant-excluded stems. The model with the best fit was chosen using the AIC criterion, adjusted R-square and comparing the different models among themselves using ANOVAs ("Car" package, R Development Core Team, 2013).

| Detailed analysis of stem growth
For this analysis, measures of stem length were carried out every 2 days (10 measures) on the control and simulated herbivory groups only. Ant-excluded plants were not included in this experiment, as frequent measurement of stems risked knocking off some of the numerous L. latipennis larvae feeding on those stems. To estimate the time lost in reaching the canopy after real or simulated herbivory, we compared the growth of intact and damaged stems using a repeatedmeasure ANOVA. We then used our data over the 20-day period to estimate by extrapolation the total time lost in reaching the canopy.
This lost time includes three potential components: (1) the time required to activate lateral buds after destruction of the apical meristem, (2) the time required to complete a new round of establishment growth, and (3) the effect of herbivory on the rate of exploratory growth.
First, the number of days required for activation of at least one lateral bud following real or simulated herbivore attack was recorded for control and simulated-herbivory stems. We considered the activation to be complete when the bud reached 1-1.5 cm in length (resting buds are 1-6 mm in length, depending on diameter of the stem bearing them).
We used the number of days necessary for bud activation as dependent variable and the initial length and diameter of the stem, the presence/ absence of larvae (at one or more daily observation), and type of damage (real or simulated herbivory) as explanatory variables. We then conducted an ANOVA using a generalized linear model (R Core Team, 2013).
Second, we estimated the time required to complete a new round of establishment growth, which, in the sense of Tomlinson and Zimmermann (1966), includes bud activation, during which the meristem of monocots begins to thicken. In D. praehensilis, this phase includes both a slow elongation and a thickening of the stem to provide the rigid base required for the self-supporting stage. The two phenomena, one peculiar to monocots, the other typical of vines with an initial self-supporting phase, are difficult to separate.
We used an arbitrary criterion to mark the end of establishment growth: The mean minimum growth rate (13.8 cm/day) of 15 healthy individuals that had begun exploratory growth and bore no trace of herbivory. Among the nine control stems subjected to real herbivory and the 25 stems subjected to simulated herbivory, only a subset reached our arbitrary criterion [real herbivory: N = 5, i.e. four control stems plus one simulated-herbivory stem) and simulated herbivory: For each stem of this subset, we recorded the number of days after bud activation required to reach the threshold growth rate.
Analysis of establishment growth using a generalized linear model was also conducted, using the same explanatory variables as in the preceding analysis. Repeated damage to the apex, a longer time to bud activation, a prolonged delay in leaving the self-supporting phase, or a combination of these causes may explain why some of the stems did not reach the threshold minimum growth rate of 13.8 cm/day (marking the end of establishment growth).
During establishment growth, the stem is rigid and self-supporting.
To determine whether this trait was associated with an increase in stem diameter that might compromise its growth in length, we com- we also compared the slopes (considered along a straight line) between two consecutive length values of exploratory growth between intact (N = 15) and simulated-herbivory stems (N = 16; the others failed to exceed our criterion for ending establishment growth). We first compared slopes of growth for each 2-day interval between treatments, then compared slopes of growth over the 20-day period for each treatment separately. For this analysis, we considered a type III error as recommended in SAS manual.

| Effect of stem size on the timing of stem emergence and the end of the leafless phase
In 2000

| Global analysis of stem growth over 20 days
Larvae of L. latipennis were responsible for all the attacks on the growing stems of D. praehensilis. Although control stems (seven cut stems of 25) and simulated-herbivory (three cut stems of 25) stems patrolled by ants suffered less herbivory than ant-excluded stems (10 cut stems of 24), the difference was not significant (2 × 3 Fisher exact test, twotailed, p = .06). However, significantly more stems on ant exclusion (15 stems attacked by larvae) bore L. lilioceris larvae than control stems (seven stems attacked by larvae) and simulated-herbivory (five stems attacked by larvae) (2 × 3 Fisher exact test, two-tailed, p = .006).
The effect of simulated herbivory was significant. Experimentally damaged stems (simulated herbivory) grew much less during the experiment (N = 25, mean ± SD: 199.4 ± 84.5 cm) than did control stems patrolled by ants (N = 25, mean ± SD: 448.5 ± 176.0 cm) and stems subjected to ant-exclusion (N = 24, mean ±SD: 352.9 ± 194.5 cm) ( Stem with large cross-sectional area had a significantly higher vertical growth rate (Table 1). Similarly, initial stem length was also positively correlated with the rate of stem vertical growth: as stems elongated, their rate of vertical growth increased (Table 1). However, the interaction cross-sectional area*initial length was also significant, as some stems with large diameter but short initial length had severe growth reduction compared to others with long initial length (Table 1). The presence of larvae on the stem at an early stage of growth increased the risk of repeated attack, as shown by the significant interactions between the presence of larvae*initial length. This could be due to the fact that female L. latipennis repeatedly oviposit on the same stems a few days apart (B. Di Giusto, pers. obs.). Finally, the interaction cross-sectional area*presence of ants was also significant, showing that the effect of cross-sectional area of a stem on its extension growth was especially important when ants were excluded (Table 1). This could indicate that biotic defense is especially important for young, small plants.

| Detailed analysis of stem growth
Of the 25 control stems, we selected 15 whose apices remained intact and that showed no signs of any perturbation of growth (such as the presence of larvae for a period of 1 week or longer, natural mechanical damage, or damage induced by errors of manipulation Simulation of herbivory* initial length 1 0.02 .87 Number of natural herbivory* presence of larvae 1 3.41 .07 Number of natural herbivory* initial length 1 0.32 .57 Presence of larvae* presence of ants 1 0.37 .54 Presence of larvae* initial length 1 5.09 .03 Cross-sectional area* presence of ants 1 6.69 .01 Cross-sectional area* initial length 1 5.37 .02 Significant results are in bold.
mean minimum growth rate of 13.8 cm/day was arbitrarily used as the threshold point marking the end of establishment growth.
If the apical meristem of the shoot was destroyed by herbivores, two lateral meristems, in general the pair situated highest on the plant (usually the distalmost intact node immediately below the site of attack), became active. One of these became dominant, repeating first establishment growth, then exploratory growth.
Our results concerning the three phases in resumption of vertical growth after natural herbivory are presented below. First, after natural herbivory, 2-4 days (N = 10, mean ± SD: 2.6 ± 0.8 days) were required before lateral buds became active and reached a length of 1-1.5 cm.
After simulated herbivory, this delay was 2-8 days (N = 25, mean ± SD: 4.6 ± 1.5 days). The difference in the time required between the two types of herbivory was significant (F 1, 34 = 14.47, p < .0006). An average delay of 4.5 days for activation of a lateral bud is, we believe, a realistic estimate of the time lost for bud activation.
None of the explanatory variables tested (type of herbivory, stem cross-sectional area, initial stem length), nor their interactions, significantly affected the number of days necessary to complete a new round of establishment growth.
The cross-sectional area of the ramification was also positively correlated with the cross-sectional area of the principal stem measured at 1 m above ground (R 2 = .37, p = .00001).
Third, this analysis included cut stems that sustained either natural or simulated herbivory. Only stems producing a ramification that attained the threshold (13.8 cm/day) chosen to mark the end of establishment growth were considered. By the end of the 20-day period, the ramifications of the cut stems had not re-attained the mean rate of extension growth of the intact plants. The mean growth rate of the ramifications (10.6 ± 3.4 cm/day) was only half that of the intact stems (22.4 ± 6.2 m/day) (F 1, 36 = 94.28, p = 1.3 × 10 −11 ). Furthermore, the larger the cross-sectional area of the stem (both intact stems and ramifications produced by cut stems), the faster the mean growth rate (F 1, 36 = 11.09, p = .002). The interaction cross-sectional area*herbivory (whether or not stems were cut) was also significant (F 1, 36 = 13.87, p < .0007), as only the mean growth rates of the intact stems showed an effect of cross-sectional area (N = 15; R 2 = .54, p = .002; see Figure 3a).
Indeed, in contrast to the positive effect of cross-sectional area on  In contrast, the cross-sectional area of experimental stems was reduced on average by 12% (from 9.6 ± 2.9 mm² in 2000 to 8.4 ± 3.1 mm² in 2001) (Figure 6). This reduction in stem cross-sectional area was likely due to both the greater material cost of producing the previous year's stem and to the considerable reduction of the period available for photosynthesis and thus in the quantity of tuber reserves stored for the next cycle. had cross-sectional areas of 35.3 ± 13.5 mm 2 (mean ± SD; range 7.5-75.4 mm 2 ), and those of cohorts 2 (32.0 ± 12.7 mm 2 , range 12.6-70.3), 3 (29.5 ± 10.3 mm 2 , range 11.4-56.7), and 4 (21.5 ± 10.9 mm 2 , range 7.5-53.6) were successively smaller (Kruskal-Wallis test, H 3, 275 = 32.28, p < .0001). Pairwise comparisons between cohorts are presented in Table 2. The strong correlation between cross-sectional area of stems and size of the tubers that produce them makes it likely that early-emerging stems came from plants with larger tubers.

| Effect of stem size on the timing of stem emergence and the end of the leafless phase
In contrast, timing of the end of the leafless phase was negatively correlated with stem cross-sectional area. The 22 stems that had already produced foliage leaves by the 24 April census all had small stem cross-sectional areas (mean ± SD 16.0 ± 8.0 mm 2 ). In comparison, the 253 stems that had not produced any foliage by this final census had larger stem cross-sectional areas (33.1 ± 12.8 mm 2 ). This difference was highly significant (Mann-Whitney, Z = 5.85, p < .000001).
We observed that these early-emerging crowns of small-stemmed plants were all located in the shaded understory at heights below 6 m. Over several years of study, this was the first time we observed D. praehensilis individuals developing their foliar crown in suboptimal conditions. Stem size (and tuber size) thus appeared to affect the plant's ability to reach the favorable light conditions of the canopy. reduced by up to a third, resulting in a delay in reaching the canopy.

| DISCUSSION
Furthermore, attacks on short-height stems increased the probability of further herbivory, and so further lengthened the amount of time required to reach the canopy. Finally, the experiment of prolonged stem growth showed a major cost in terms of lost opportunities. A shortened productive period likely reduced the replenishment of tuber reserves, and thus the probability of survival and growth abilities of surviving individuals in the following year. Our study also suggests that the strategy of D. praehensilis to prevent and respond to its specialist herbivore consists in maximizing stem growth speed and precocity through an over-accumulation of underground reserves.

| Short-term costs
The most vulnerable point in D. praehensilis' cycle is the high cost of herbivory during the annual trip to the canopy. This high cost results from its growth strategy, which is conditioned by the constraints imposed by its physical environment, by the presence of a specialist herbivore, by the vine life form, and by the monocot habit. First, in monocots, the limited number of meristems (Holttum, 1955) restricts the possibility of ramification. Furthermore, because of the greater complexity of their vascularization, "vascular development of a Monocots' branch often takes place over a longer period of time" (Zimmermann & Tomlinson, 1972). Second, the annuality of the stem and the need to reach maximal solar exposure before branching results in a strong dominance of the apical meristem. This precludes the possibility of modifying the plant architecture through the activation of latent axillary meristems following herbivory (Marquis, 1996;Wise & Abrahamson, 2008). Third, both the ecological and architectural constraints of vines and monocotyledons necessitate an establishment growth stage for each new ramification. On the one hand, most monocotyledons require thickening of the meristem to the primary diameter that will be maintained throughout the lifespan of the shoot system (Holttum, 1955). On the other hand, vines need a solid base to reach surrounding plant supports.
These architectural and ecological constraints tend to shape both resistance (the ability to prevent or lessen the herbivore damage) and tolerance (the ability to maximize recovery following herbivory) (Crawley, 1989;Núñez-Farfán, Fornoni, & Valverde, 2007). Although tolerance is likely an effective strategy against specialist herbivores that are relatively immune to the plant's chemical defenses (Strauss & Agrawal, 1999), D. praehensilis showed no overcompensation after herbivory but rather a reduced rate of exploratory growth. This reduction of the growth rate is both potentially the most important and the one for which our estimates are most subject to uncertainty.
However, two findings suggest a lasting reduction in growth rates.
First, by the end of the 20-day experiment, the maximal rates of extension growth of damaged stems were still reduced by about one-third compared to those of intact stems. Second, the slope of the growth rates of damaged stems appeared to cease increasing 4 days before we terminated observations (Figure 4a). Further observations are needed to determine why the reduction in growth rate persists. We hypothesize an imperfect vascularization between the main stem and its ramification due to the absence of secondary cambium in Monocots (Tomlinson, 1980) or an effect of the bigger diameter of the ramification that might compromise the growth in length.
A simulation of growth ( Figure 7) using our previous results led to a total delay of 10 days in reaching the canopy for equal rates of exploratory growth before and after herbivory. In contrast, for a 30% reduction of rates of exploratory growth, the delay in reaching the canopy varied from 21 to 45 days depending on the height of the apex when herbivory occurred (Figure 7). The earlier the attack, the greater the cost imposed. As found in earlier studies (Marquis, 1992b;Marquis et al., 1997;Strauss & Agrawal, 1999;Tiffin, 2002), Our study is also consistent with others (Blundell & Peart, 2001;Fornoni, 2011;Wise & Abrahamson, 2008) (Tomlinson, 1980).
Several authors have discussed how much simulated herbivory tells us about the effects of real herbivory (Tiffin, 2000;Tiffin & Inouye, 2000). In our study, the time required to activate an axillary bud was significantly different between real and simulated herbivory. This is likely due to the gradual destruction of the apex of the stem by larvae in real herbivory (in contrast to an instantaneous event for T A B L E 2 Differences in cross-sectional area between stems that emerged in each of the four census intervals simulated herbivory), which appeared to induce an early release of lateral buds from apical dominance, even before our assessment that the apex was destroyed.

| Long-term costs
As Koricheva (2002)  Faced with a specialist herbivore able to subvert the plant's chemical defenses and sequester them in its own defense (Di Giusto et al., 2001), an increase in the plant's chemical defenses might be counterproductive by leading to increased effectiveness of the beetles' fecal shields against ants (Baldwin & Preston, 1999). Under these circumstances, the most effective defense may be to amass a larger tuber, enabling the plant to grow earlier and faster and providing the plant with insurance against severe levels of herbivory.
Our results support our hypothesis that the plant's aerial apparatus and underground reserves are tied together in a cycle of reciprocal annual renewal. First, the greater the quantity of reserves stored in the tuber, the larger the cross-sectional area of the stem produced. Larger stems grew more rapidly in length ( Figure 5) and emerged from the F I G U R E 7 Modeling of growth to the canopy by stems of Dioscorea praehensilis with and without herbivory. Using the data from the 20-day experiment, we estimated by extrapolation the effect of herbivory on the total time lost in reaching the canopy. Assuming a canopy height of 15 m above ground, we took into account the sources of delay we measured: time required to activate lateral buds (4 days), time required to complete the new phase of establishment growth (8 days), and a lasting reduction of up to one-third of the maximal growth rate even 20 days later, during exploratory growth. The major uncertainty was the extent to which there was a long-lasting reduction of growth rates. Depending on the type of analysis conducted, the extension growth rates of cut plants had either attained a slope comparable to that for intact stems by the end of the experiment (Figure 6b) or a slope that was still up to one-third lower than that for intact plants (Figure 6a). We, thus, examined both alternatives. The curve with black squares shows the growth of intact stems from ground to canopy (the black line represents the tangent of the curve). We compared the effects of herbivory on the apex at 200 and at 1,200 cm height and calculated the growth of the ramification. Curves drawn with white circles show the growth of the new ramification with a rate of exploratory growth equivalent to that of intact stems after the activation of its lateral buds and a new establishment growth. Curves drawn with black crosses are similar to the previous curves, but with a rate of exploratory growth reduced by 30% relative to that of intact stems (the dashed line represents the tangent of the curve). Greater water reserves may enable larger plants to initiate growth well before the end of the dry season, while smaller plants (and L. latipennis imagos) might have to await the onset of the rains. Water accounts for 70%-75% of the fresh weight of tubers; plants with larger tubers have more of this resource, as well as more energy. They may also have deeper and more extensive root systems, enabling them to tap water reserves deeper in the soil than can smaller plants.
Missed opportunities may drive the plant into a downward spiral.
As in other heliophiles, its seedlings establish only in light gaps. But unlike other heliophiles, which maintain their place in the canopy by growing up continuously as the gap closes, this geophyte must make a longer trip from the ground to the canopy with each successive year until the gap closes (McKey et al., 1998). Thus, a steady increase in the rate of growth from year to year is a necessity, given the regeneration niche of D. praehensilis. Plants that fail to reach the canopy in time for a full season of photosynthetic activity are more likely to fail to accumulate sufficient reserves to minimize the effects of herbivory and reach the canopy in the following year -by which time the canopy will be higher. Despite several reviews showing an impact of herbivory on growth, flowering, fruiting, or seeding within the same growth season (Strauss & Agrawal, 1999;Wise & Abrahamson, 2008), our study is the first that we are aware of that documents costs persisting across growth seasons.
However, annual renewal of the entire aerial apparatus is not the only solution to the problems of viny monocots. Some wild yams closely related to D. praehensilis, and sympatric with this species in forests of central Africa, possess aerial systems that last two or more years (Hladik et al., 1984).

| CONCLUSION
Dioscorea praehensilis has evolved a strategy, shaped by the ecological and architectural constraints it faces, that centers on the annual and reciprocal renewal of both the aerial apparatus and underground tuber.
The plant's aerial apparatus is highly vulnerable to herbivory occurring during the annual trip to the canopy. Herbivore attacks that destroy the apical meristem greatly reduce stem growth rate to the canopy.
By temporarily halting growth, herbivory reduces the advantage to the