Trade-offs among anti-herbivore resistance traits: insights from Gossypieae (Malvaceae) ^†

Natural history of the Gossypieae

Gossypieae includes perennial shrubs, most of which are native to arid tropical and subtropical regions (Fryxell, 1979; Percival et al., 1999). Gossypium, the clade containing cultivated cotton, includes 45 species. Gossypieae species share several direct and indirect resistance traits, yet vary considerably in the degree to which they invest in these traits.

Intraspecific investment in resistance traits

For G. thurberi, we assessed the resistance traits of naturally occurring individuals chosen at random from three populations: Florida Canyon (31°45′39.8″ N, 110°50′45.8″ W, n = 43) and Agua Caliente Canyon (31°41′32.4″ N, 110°57′22.7″ W, n = 52), both in the Santa Rita Mountains, Arizona, USA, and Molino Basin (32°20′21.5″ N, 110°41′13.6″ W, n = 46) in the Santa Catalina Mountains, Arizona. Prior experiments showed that ants conferred indirect defense in Florida and Agua Caliente canyon populations. In contrast, plants at Molino Basin experienced low visitation by ants and received no protective benefit (Rudgers, 2002, 2004).

Indirect resistance traits included the frequency and size of extrafloral nectaries. In G. thurberi, these traits did not respond to experimental manipulations of herbivory in the field (i.e., were not inducible, Rudgers, 2002). We determined the proportion of leaves with nectaries every 6 wk during the growing season (July–December 2000). Twice, we measured the maximum length of foliar nectaries to the nearest 0.01 mm with digital calipers (1 July, 1 October) on 15 randomly selected, fully expanded leaves per plant. Nectary size was positively correlated with nectar production in the greenhouse, and the availability of nectar increased visitation by ants in the field (Rudgers, 2004).

Direct resistance traits included the density and size of gossypol glands. In G. thurberi, neither character appeared to be induced by herbivory in the field (Rudgers, 2002). Trichomes were not examined because leaf pubescence was sparse and varied minimally. During 20–31 October 2000, five leaves per plant were collected at random and measured with a leaf area meter (LI 3000A, LI-COR, Lincoln, Nebraska, USA). Glands were counted under a microscope within a circular region (area = 2.4 mm²) for three sections of the leaf: adjacent to the midrib, halfway between the midrib and leaf margin, and adjacent to the margin. For three leaves per plant, we measured the maximum diameters of three glands per leaf to the nearest 0.001 mm with image analysis (Scion Image, Scion, Frederick, Maryland, USA). Gland volume was calculated assuming that each gland was spherical.

The mean values of resistance traits were determined across all leaf samples per plant and combined into indices of investment in gossypol glands (mean gland density × mean volume per gland) and extrafloral nectaries (proportion of leaves with nectaries × mean number of nectaries per leaf × mean size per nectary). The indices were particularly useful for the cross-species comparisons (see next) because nectary and gland traits varied markedly in size and number across species.

Within each population, we determined phenotypic correlations between the gossypol gland and extrafloral nectar indices. Phenotypic correlations can be reliable indicators of underlying genetic correlations (Roff, 1995). However, environmental, rather than genetic, covariation can underlie phenotypic relationships and is a limitation of this approach (Mitchell-Olds and Shaw, 1987). We used multiple regression analyses to reduce the influence of environmental covariation as indicated by leaf size and leaf number. The extrafloral nectary index was the dependent variable and the gland index, leaf area, and initial leaf number were independent variables. Gland and nectar indices as well as leaf area were log transformed, and leaf number was square-root transformed to achieve normality and homoscedasticity of residuals. No significant multicollinearity was detected (Proc REG collinoint, SAS Institute, 1999). Standard partial regression coefficients (b′) (Sokal and Rohlf, 1995) were used to assess relationships among traits.

Interspecific investment in resistance traits

We examined 31 species of Gossypium and related taxa in the Gossypieae that vary considerably in both direct and indirect resistance. On 27 March and 3 April 2001, a minimum of five leaves was collected from plants (approximately 2–10 yr old) that were growing under standard glasshouse conditions (18–26°C, 12-h photoperiod, arthropod-free) at Iowa State University, Ames, Iowa, USA. Leaves were selected to maximize variation within a plant and included the oldest and the youngest fully expanded leaves on each individual. Because plants were grown in the absence of herbivory, our measures represent constitutive, rather than induced, levels of resistance. Whether these traits are inducible and the degree to which they can be induced is not known for the majority of Gossypieae. By using plants grown in a common environment, we controlled for genotype × environment interactions (Blomberg and Garland, 2002) and for unmeasured “third” traits associated with habitat (including herbivory and induced responses) that may correlate with resistance phenotypes (see Price, 1997). These controls confer an advantage over studies that use published data from multiple sources or samples collected from different environments.

In many cases (23 species), we used single individuals to represent a species, owing to a lack of other samples (some species are difficult to propagate). For the remaining species, 2–6 individuals were sampled (sample size = 3.0 ± 0.32 individuals [means ± 1 SE]). While we recognize that having a single individual to represent a species is less than ideal, standard programs designed to account for phylogeny (phylogenetically independent contrasts) do not incorporate the sample size per species and assume that within-species variance is zero (see COMPARE 4.4, Martins, 2001 or Contrast 3.5c, Felsenstein, 1991). Thus, in using single individuals, we only reduced the precision of our estimate of the mean for those species. Phylogenetically independent contrast studies frequently sample single (or few) individuals per taxon (e.g., Vanhooydonck et al., 2000; Summers and Clough, 2001). These papers and ours assume that variation across species is greater than variation within species. Importantly, variation across Gossypieae species was significantly greater than variation within G. thurberi, for which we sampled 141 individuals across a range of ant-free and ant-tending environments and across varying levels of herbivory (Table 1; Rudgers, 2002).

We scored extrafloral nectary, gossypol gland, and trichome traits for each species. The proportion of leaves with nectaries was determined in the glasshouse. All nectaries on each sampled leaf were counted and photographed under a microscope in the laboratory. We measured the area of nectar-producing cells with image analysis (Scion Image) and leaf area with a leaf area meter (LI-COR). To obtain gland density, we counted the number of glands within a circular region (area = 0.4 cm²) at the same location on each leaf, near the distal end of the primary lobe. Gland volume was determined as described earlier for a minimum of three glands per leaf and for five leaves per plant. Using a microscope, we scored trichome cover on a scale of zero (no trichomes) to 10 (trichomes so dense and long that the leaf surface was obscured). For each species, mean trait values were determined across all leaf samples, and indices of investment in gossypol glands and extrafloral nectaries were created as explained previously.

Phylogenetic trees

We used the well-documented phylogenetic hypotheses for the Gossypieae clade, which are based on phylogenetic analysis of molecular data sets derived from numerous chloroplast and nuclear gene sequences (Cronn et al., 2002, and references therein). Most branches of the synthesis shown in Fig. 2 have high internal support and/or are independently inferred from multiple, independent data sets (11 nuclear and four chloroplast loci in Cronn et al., 2002; see also Seelanan et al., 1997; Liu et al., 2001; Wendel et al., 2002). Not all taxa included in the work of Cronn et al. (2002) were used in the present study; thus, neither branch lengths nor internal support statistics are shown here. All phylogenetic analyses of Gossypium are congruent in showing each genome group to be monophyletic, although some areas of uncertainty remain, particularly regarding the placement of the B-genome clade (sister to either C + G or A + F) and the E-genome clade (sister to either C + G or the remainder of the African genome groups, A, F, B). Accordingly, we analyzed alternative trees that reflect this uncertainty: Tree 1 is our best estimate of the relationships among Gossypieae (Fig. 2). Tree 2 has the B-genome clade sister to C + G rather than to A + F, and Tree 3 has the E-genome clade sister to A + F + B instead of sister to C + G. Because it made the tree reticulate, we omitted a small, monophyletic clade of five allopolyploid species (G. barbadense, G. hirsutum, G. darwinii, G. mustelinum, G. tomentosum) that originated following hybridization between an A- and a D-genome diploid (Wendel, 1989).

Phylogenetically independent contrasts

The COMPARE program (Martins, 2001) was used to calculate standardized independent contrasts for the three alternative topologies, assuming a Brownian motion model of evolution. Because information was not available on branch lengths, all branch lengths were set equal to one. A simulation study found that the assumption of equal lengths produced lower inflations in error than did alternative methods (Ackerly, 2000).

We used least squares regression to examine relationships among phylogenetically independent trait values (i.e., contrasts) (Proc REG, SAS Institute, 1999). Gland and nectary indices as well as leaf area were log transformed, and trichome cover was square-root transformed to achieve proper standardization of the contrasts (see Garland et al., 1992). We report standard regression coefficients (b) (Sokal and Rohlf, 1995) for regression lines through the origin (see Appendix 1 in Garland et al., 1992). For the extrafloral nectary index, multiple regression was used to control for variation in leaf area (a trait that was significantly correlated only with the extrafloral nectary index), and standard partial regression coefficients (b′) are reported. In addition, we performed regressions for data that were not corrected for phylogeny. In all analyses, the data met assumptions of univariate normality, linearity, no multicollinearity, and homoscedasticity of residuals as required for regression techniques. Standard outlier detection methods resulted in the exclusion of two outliers from the phylogenetically independent correlation between gland index and trichome cover (Proc REG influence, SAS Institute, 1999). Confidence limits (95%) were determined for regression coefficients by bootstrapping (3000 iterations) using the bias-corrected bootstrap (Efron and Tibshirani, 1993; Jackboot macro, SAS Institute, 1999). Percentile and bias-corrected accelerated bootstraps gave similar results. If the assumption of equal branch lengths is incorrect, these confidence limits can lead to inflated type I error (Ackerly, 2000). Thus, we also present P values from a table of adjusted critical values from Ackerly (2000), which provides a correction for this error.

Phenotypic correlations in G. thurberi

Direct and indirect resistance traits were not correlated for any of the G. thurberi populations: Florida Canyon (Fig. 3A; standard partial regression coefficient b′ = −0.003, t = −0.02, P = 0.98, n = 43), Agua Caliente Canyon (Fig. 3B; b′ = −0.046, t = −0.34, P = 0.74, n = 52), Molino Basin (Fig. 3C; b′ = −0.095, t = −0.55, P = 0.59, n = 46). The absolute values of the standard partial regression coefficients |b′| needed to be greater than 0.33, 0.28, and 0.35, to achieve significance of a two-tailed t test at P < 0.05 in Florida, Agua Caliente, and Molino populations, respectively (following Zar, 1999). The lack of a relationship was consistent across populations, despite the fact that populations differed significantly in the levels of protection afforded by ants (Rudgers, 2002, 2004).

Correlations across the Gossypieae

In our examination of Gossypieae, correlations between direct and indirect resistance traits mirrored those documented within G. thurberi. Neither the gossypol gland index nor trichome cover was significantly negatively correlated with the extrafloral nectary index (Fig. 4A, B; Table 2). In fact, the standard partial regression coefficient for the relationship between extrafloral nectaries and glands was positive for all trees examined, although marginally significant only for Tree 3 (see 95% CL, Table 2). For Tree 1, to achieve significance at P < 0.05 using a two-tailed t test, the absolute value of the standard partial regression coefficient |b′| must be greater than 0.371 for the extrafloral nectary index vs. the gland index and 0.369 for the extrafloral nectary index vs. the trichome index; the power of these tests was 0.65 and 0.52, respectively (following Zar, 1999).

In contrast, there was a strong negative correlation between investment in gossypol glands and investment in trichome cover across Gossypieae (Fig. 4C, Table 2, P < 0.01). The gossypol gland index explained approximately 34% of the variation in trichomes (Tree 1, r² = 0.344, P < 0.01; Tree 2, r² = 0.338, P < 0.01; Tree 3, r² = 0.339, P < 0.01).

Our cross-species results were not sensitive to uncertainty in the topology; patterns of association were similar for all three trees examined (Table 2). In addition, phylogenetic corrections did not significantly influence the correlations among resistance traits. The direction of the correlations and their significance levels were similar for both the phylogenetically independent contrasts and the species correlations (Table 2). In our case, accounting for phylogeny appeared to enhance the strength of the correlation between glands and trichomes, although not significantly (see 95% CL, Table 2).

Direct resistance

The direct resistance traits of leaf glands and trichomes were negatively correlated. Species with dense trichome cover had reduced investment in gossypol glands. This pattern was robust as it was maintained across three alternative trees that accounted for uncertainty in the phylogenetic hypotheses for Gossypieae. The negative correlation may reflect several underlying mechanisms. First, plants' investment in both glands and trichomes may be constrained by costs. Costs may include both the structural investments (such as trichome cells or terpenoid aldehydes) as well as nondefensive functions of the traits (e.g., trichomes may reduce water loss [Ehleringer and Mooney, 1978], and glands may protect against pathogens [Bell, 1986]). Second, glands and trichomes may be negatively genetically correlated. Third, glands and trichomes may deter similar herbivores and therefore be redundant in ecological function. However, given that these traits deter different types of pests of cultivated cotton species (Jenkins et al., 1967; Bell, 1986; Matthews, 1989; Smith, 1992; Summy and King, 1992), the redundancy hypothesis seems unlikely. An examination of pleiotropy as well as an assessment of fitness costs may help isolate the mechanisms underlying the negative correlation between glands and trichomes.

Direct resistance vs. indirect resistance

Both within and across species, indirect resistance (extrafloral nectar investment) was not correlated with either of the direct resistance traits (gossypol glands and trichomes). For each of three populations of Gossypium thurberi, there were no significantly negative phenotypic correlations between glands and extrafloral nectaries. The lack of significance is unlikely to be due to a lack of power because standard partial regression coefficients would only have to be larger than 0.35 to detect a significant relationship, given the sample sizes and variances for each population. The evidence is consistent with the hypothesis that gland and extrafloral nectary traits evolve independently; however, phenotypic correlations are only a first step toward understanding correlated characters, and genetic studies are needed to determine the basis of these patterns.

Paralleling the results from G. thurberi, investment in direct resistance traits across the Gossypieae did not negatively correlate with allocation to indirect resistance for any phylogeny that we examined. The absence of a negative correlation between direct and indirect resistance traits stands in contrast to previous studies (see Introduction), which suggest that negative correlations occur more commonly between direct and indirect resistance traits than among types of direct resistance. Differences in methodology may explain the discrepancy between our results and those of previous studies, including the prior focus on obligate, rather than facultative, ant–plant mutualisms, the failure of previous studies to incorporate phylogeny in the analysis, or the prior use of ant exclusions in lieu of the quantification of plants' actual investment in indirect resistance traits. Phylogenetically corrected explorations of indirect resistance traits in both facultative and obligate mutualisms are needed to distinguish among these hypotheses. In addition, because correlations across species do not necessarily reflect patterns within species, individual species should be examined (as with G. thurberi) to determine whether patterns of investment hold within species.

When are negative correlations expected?

The lack of a negative relationship between direct and indirect resistance traits in Gossypieae may be explained by a number of non-mutually exclusive hypotheses. First, resistance traits may pose little cost to plants, although recent reviews of the fitness costs of resistance suggest otherwise (Koricheva, 2002; Strauss et al., 2002). Second, resistance traits may work in conjunction to reduce herbivory, and thus plants may require multiple traits to mount an effective defense (Futuyma, 2000). Some studies have found positive rather than negative correlations between resistance traits (e.g., Zangerl and Berenbaum, 1990; Ruuhola et al., 2001), a result that is consistent with the hypothesis that resistance traits may exhibit synergisms. For example, experimental evidence from Acacia drepanolobium showed that the combination of ants (indirect) and thorns (direct) synergistically reduced browsing by herbivores (Stapley, 1998). Similarly, a volatile sesquiterpene (caryophyllene oxide) in cotton that had no independent effect on herbivory, interacted synergistically with the terpenoid aldehyde, gossypol, to reduce feeding by Heliothis caterpillars by 33% compared to gossypol alone (Gunasena et al., 1988).

Multiple mechanisms of resistance may be particularly crucial in Gossypieae species because extrafloral nectaries can directly attract some herbivore species. Extrafloral nectaries in cultivated cotton can enhance the abundance of or damage by several crop pests including lepidopterans and plant bugs (Trelease, 1879; Lukefahr and Rhyne, 1960; Lukefahr et al., 1960; Benschoter and Leal, 1974; Schuster et al., 1976; Wilson and Wilson, 1976; Henneberry et al., 1977; Adjei-Maafo et al., 1983; Beach et al., 1985; Smith, 1992; Summy and King, 1992), mainly because adults of these taxa consume extrafloral nectar. Other traits, such as gossypol glands and trichomes, may be needed to deter herbivores attracted to nectar. Therefore, in contrast to predictions made by current theories on plant defense, cotton species investing in extrafloral nectar may require alternative defenses due to the ecological cost of nectar in attracting some herbivore species. However, this ecological cost may be more apparent in cultivated cotton systems, where natural communities of predators are likely to be disrupted by routine tilling and pesticide application. Further studies in natural systems are needed.

The absence of a correlation between extrafloral nectary and gland/trichome traits does not preclude the possibility that correlations occur with other, as yet unmeasured, plant resistance characters. These traits include condensed tannins (Bell, 1986), leaf toughness (Coley, 1983), volatiles that function in indirect resistance by signaling predators and parasitoids of herbivores (Loughrin et al., 1995; Pare and Tumlinson, 1997), and induced forms of direct or indirect resistance to herbivory (Karban and Baldwin, 1997). An analysis encompassing additional resistance traits would provide an even broader picture of allocation to resistance in Gossypieae.

Finally, the robust lack of a correlation may reflect the facultative nature of indirect resistance in Gossypieae. Facultative mutualisms are prevalent in nature and highly conditional (Bronstein, 1998). Negative correlations with direct resistance traits may be most likely when indirect resistance is obligate, rather than facultative, in nature. For example, Turner (1995) showed that Macaranga plants engaged in obligate relationships with ants had lower levels of tannins than species with non-obligate associations. In addition, leaves from obligately ant-defended acacias often had lower concentrations of cyanogenic glycosides and less resistance to caterpillars than leaves from non-obligate acacias (Janzen, 1966; Rehr et al., 1973; Seigler and Ebinger, 1987). Exploring a range of mutualisms from unspecialized and facultative to specialized and obligate may shed light on the conditions under which negative correlations among resistance traits occur. In such comparisons, controlling for phylogeny is essential, as plants in obligate associations may be more closely related to each other than to species in facultative associations (e.g., Davies et al., 2001).

In summary, explorations in the cotton tribe, Gossypieae, demonstrate patterns of investment in anti-herbivore resistance traits that differ from the majority of other studies on trade-offs among plant defenses. Negative correlations, as occurred between trichomes and gossypol glands in Gossypieae, may reflect costs or redundant functions of these constitutively expressed, direct resistance traits. In contrast, the lack of a negative correlation between these direct resistance traits and extrafloral nectar suggests that these traits evolve independently over evolutionary time scales. Finally, our results lead to the prediction that indirect defenses associated with unpredictable, facultative mutualisms (such as those in Gossypieae) will be less likely to be negatively correlated with other resistance traits than are indirect defenses that mediate obligate protective mutualisms.