SEARCH

SEARCH BY CITATION

Keywords:

  • Bignoniaceae;
  • Bignonieae;
  • genetic variance/covariance matrix;
  • morphological evolution;
  • quantitative traits

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Phenotypic integration is essential to the understanding of organismal evolution as a whole. In this study, a phylogenetic framework is used to assess phenotypic integration among the floral parts of a group of Neotropical lianas. Flowers consist of plant reproductive organs (carpels and stamens), usually surrounded by attractive whorls (petals and sepals). Thus, flower parts might be involved in different functions and developmental constraints, leading to conflicting selective forces. We found that Bignonieae flowers have very similar patterns of variance/covariance among traits and that such patterns are uncorrelated with the phylogenetic relationships between species. However, in spite of pattern stasis, our results also indicate that diversification of floral morphology in this group has occurred throughout the evolution of magnitudes of correlation among traits. Thus, we suggest that stabilizing selection has played an important role in phenotypic integration, resulting in the long-term stasis of covariance patterns underlying flower diversification during the ca. 50 Myr of evolution of Bignonieae. This is the first report of long-term stasis in the phenotypic integration of angiosperms, suggesting that patterns of floral morphology can be recognizable as specific attributes of distinct botanical families.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Evolution of form is a key question in evolutionary biology. As such, it is crucial to understand the evolution of complex relationships among traits that characterize different groups of organisms. Such trait relationships are known as morphological integration (Olson & Miller, 1958; Berg, 1960; Clausen & Hiesey, 1960) or phenotypic integration (Cheverud, 1996; Pigliucci, 2003). Usually, phenotypic integration is characterized by discrete groups of highly correlated traits with varying levels of interconnection at different hierarchical levels, which can emerge from conflicting selective agents, as well as from functional and developmental associations (Lande, 1979; Cheverud, 1996; Wagner & Altenberg, 1996).

The pattern of covariation among traits is represented by the additive genetic variance–covariance matrix (G). G is a fundamental parameter in evolutionary theory, as it summarizes the proportion of variance and covariance that are effectively inherited. Strong stabilizing selection could keep G highly integrated over long periods of time, resulting in similar correlations/covariances among traits above the species level (Lande, 1979; Cheverud, 1996; Estes & Arnold, 2007; Revell, 2007). In contrast, short periods of strong directional selection acting towards morphological diversification, such as the input of genetic variation by mutation in a multilocular-pleiotropic system, might disrupt the constraining aspects of an integrated G and thus decrease the correlation among traits (Lande, 1979; Cheverud, 1996; Marroig & Cheverud, 2001; Steppan et al., 2002; Pigliucci, 2004; Wagner, 2010). Apart from the processes that determine G, the evolutionary potential of lineages is also influenced by their structure (Lande, 1979). Clades that share similar variance/covariance patterns are expected to show similar responses to selection. On the other hand, evolutionary changes in G can lead to variable responses to the same selective pressures (Marroig & Cheverud, 2001). In addition, the magnitude of these correlations among traits also influences the response to selection (Lande, 1979; Wagner & Altenberg, 1996). Differences in the pattern and magnitude of phenotypic integration of complex organs have been demonstrated, suggesting that these features may have evolved independently (Marroig & Cheverud, 2001; Oliveira et al., 2009; Porto et al., 2009). In general, stasis of underlying patterns has been highlighted, along with diversification in the magnitude of trait associations, both of which represent important drivers of the direction and speed of morphological evolution (Marroig & Cheverud, 2001; Oliveira et al., 2009; Porto et al., 2009).

While trait association is important, the evolution of phenotypic integration and stability or disparity of G during the evolutionary history of lineages still remain to be investigated through broad-scale empirical comparisons for most organisms (Arnold, 1992; Marroig & Cheverud, 2001; Steppan, 2004; Jamniczky, 2008). The estimation of genetic correlations among morphological traits requires large sample sizes of phylogenetically related individuals, a difficult task for most organisms (Cheverud, 1996). However, pooled within-group variance/covariance phenotypic matrices (W) obtained from simple phenotypic variance/covariance matrices (P) can be used as surrogates for G if P is similar or proportional to its genetic counterparts (Cheverud, 1996; Marroig & Cheverud, 2001; Prôa et al., 2012). Data from multiple taxa have indicated that P might be a good surrogate for G, especially for morphological characters (Waitt & Levin, 1998; Marroig & Cheverud, 2001). This pattern is even more remarkable for plants (reviewed in Waitt & Levin, 1998), but empirical evaluations of trait association at broad phylogenetic scales are particularly scarce in this group of organisms (but see Armbruster et al., 2004; Pérez et al., 2007; Ordano et al., 2008 and Rosas-Guerrero et al., 2011, for more restricted phylogenetic scales).

In angiosperms, phenotypic integration among floral parts is thought to result from functional trade-offs among characters involved in pollinator attraction, sexual functions, antagonistic avoidance and efficient pollen donation/reception (Berg, 1960; Stebbins, 1974; Ordano et al., 2008). Indirect selective forces caused by developmental pathways and genetic architecture, coupled with lineage history, can also result in complex floral morphologies (Berg, 1960; Stebbins, 1974; Riedl, 1977; Diggle, 1992; Herrera et al., 2002; Armbruster et al., 2004; Ordano et al., 2008; Harder & Johnson, 2009). Consequently, the level of integration among floral traits might be taxon specific, as determined by the functional role of each floral trait, developmental/genetic correlation among those traits and history of each lineage (Armbruster et al., 2004; Ordano et al., 2008). In this sense, clades with diverse floral morphologies and well-supported phylogenies are particularly interesting for a better understanding of the evolutionary patterns of phenotypic integration of floral traits. The tribe Bignonieae (Bignoniaceae) is a monophyletic group (Lohmann, 2006) that fits such characteristics.

Bignonieae comprises ca. 400 wood species widely distributed throughout the Neotropics (Lohmann, 2006). Most Bignonieae species are lianescent, representing the most diverse clade of Neotropical lianas, and the tribe seems to have originated in the Eocene, at approximately 50 Myr ago (Lohmann et al., 2013). The floral morphologies (Fig. 1) represented in Bignonieae are thought to be associated with shifts in pollinator guilds (Gentry, 1974; Alcantara & Lohmann, 2010). These Bignonieae flower ‘types’ (as defined by Gentry, 1974) seem to have appeared by homoplastic evolution from the basic Anemopaegma-type morphology (Alcantara & Lohmann, 2010). Moreover, individual quantitative floral traits exhibit opposing phylogenetic signals and differ in their evolutionary rates, suggesting that complex evolutionary dynamics may have influenced the evolution of floral form in this group (Alcantara & Lohmann, 2011). In addition to the labile evolution of floral types associated with different pollination strategies, most species of Bignonieae so far evaluated seem to lack post-zygotic barriers to interspecific mating (F. Firetti, pers. comm.), and the nectar robbers observed in several species (compiled in Alcantara & Lohmann, 2010) provide examples of the putative conflicting pressures acting on these flowers.

image

Figure 1. Floral morphology diversity observed in the tribe Bignonieae (Bignoniaceae). Morphological types were named as classified by Gentry (1974), with the addition of the ‘mixed floral types’ (Alcantara & Lohmann, 2010). (a) Anemopaegma-type, Anemopaegma chamberlaynii (Sims) Bureau & K. Schum. (b) Martinella-type, Pyrostegia venusta (Ker Gawl.) Miers. (c) Amphilophium-type, Amphilophium paniculatum (L.) Kunth. (d) Pithecoctenium-type, Amphilophium crucigerum (L.) L. G. Lohmann. (e) Cydista-type, Bignonia corymbosa (Vent.) L.G. Lohmann. (f) Tynanthus-type, Tynanthus cognatus (Cham.) Miers. (g) Tanaecium-type, Tanaecium jaroba Sw. (h) ‘Mixed-type’, Stizophyllum inaequilaterum Bureau & K. Schum. Modified from Alcantara & Lohmann (2010).

Download figure to PowerPoint

In this study, we address whether different taxa and different floral morphologies of Bignonieae share a common variance/covariance structure among traits or whether changes in floral morphologies are associated with changes in this underlying structure. More specifically, we use a phylogenetic context to evaluate the patterns and magnitudes of phenotypic integration among floral traits in order to understand the evolutionary dynamics of floral morphology in Bignonieae.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Specimens and traits sampled

To sample the morphological diversity of Bignonieae, we performed flower measurements in 1085 herbarium specimens of 102 species (Appendix S1) for which the phylogenetic relationships are known (Lohmann, 2006). A complete list of the examined materials is provided at Appendix S1. Specimens were obtained from the following institutions: Herbário Barbosa Rodrigues (HBR), Museu Botânico Municipal de Curitiba (MBM), Missouri Botanical Garden (MO), New York Botanical Garden (NY), Jardim Botânico do Rio de Janeiro (R), Instituto Botânico de São Paulo (SP), Herbário da Universidade de São Paulo (SPF) and Herbário da Universidade Estadual de Campinas (UEC). We examined only one well-preserved flower with completely expanded petals per specimen. Because of the difficulty of evaluating three-dimensional shapes from herbarium specimens, only two-dimensional measurements were taken, using a ruler and millimetre paper (Fig. 2).

image

Figure 2. Quantitative traits sampled. Traits are coded as follows: 1. CoL: corolla length (total); 2. CoD: corolla diameter (total); 3. CTL: corolla tube length; 4. CDBC: corolla diameter at the base of the calyx; 5. CDM: corolla diameter at the mouth openness; 6. CaL: calyx length; 7. CaM: calyx midpoint diameter; 8. LISS: location of the insertion of the small stamens (measured from the base of the ovary); 9. LILS: location of the insertion of the large stamens (measured from the base of ovary); 10. LSS: length of the small stamens; 11. LLS: length of the large stamens; 12. AnW: anther width (measured at midpoint); 13. AnL: anther length; 14. StyL: style length (measurement including ovary length); 15. StW: stigma width (measured at midpoint); 16. StL: stigma length.

Download figure to PowerPoint

The 16 floral traits analysed reflect the variation in the four floral whorls: sepals, petals, androecium and gynoecium (Fig. 2). The same traits were used to evaluate the phylogenetic signal in flower morphologies in a previous study (Alcantara & Lohmann, 2011). All specimens were measured twice, with the analysed values representing the average of repeated measurements. Measurement repeatability (i.e. the proportion of total variance based on individual differences rather than measurement error) was estimated following the procedure described by Lessells & Boag (1987). All trait values were ln-transformed to fit normality.

Phylogeny and taxonomy of Bignonieae

A combined phylogeny including 104 species of Bignonieae that used chloroplast (ndhF) and nuclear (PepC) DNA sequences was used as a basis to elaborate a new generic classification that recognizes 21 monophyletic genera diagnosed by morphological synapomorphies (Lohmann, 2006). This tree was time-calibrated with fossil data, resulting in a tree with branches proportional to time (Lohmann et al., 2013). The phylogeny used here was identical to the combined molecular phylogeny from Lohmann (2006) except that Pyrostegia dichotoma was treated as a synonym of Pyrostegia venusta, whereas Perianthomega vellozoi was included in the tree following the relationships recovered by the ndhF phylogeny to represent this monotypic genus of Bignonieae not included in the combined phylogeny (Lohmann, 2006). Therefore, the present study uses this time-calibrated tree and follows the generic classification based on this phylogeny (Lohmann, 2006; Appendix S1; Appendix S2). Herbarium specimens of 102 species included in the dated phylogeny had flowers that could be analysed morphologically. Thus, the two species missing from the original chronogram were removed, with branch lengths kept proportional to time.

Data analyses and hypotheses tested

Estimation of variance/covariance and correlation matrices among floral traits

Our main data set included all 16 floral characters sampled (Fig. 2), which are representative of the overall floral morphology of Bignonieae, as well as 16 of the 21 recently recognized genera of this group. Genera with a small number of sampled specimens were not included in the analyses. Another data set, including only sepal and petal measurements (seven traits), was also evaluated to increase taxonomic sampling by inclusion of genera not included in the main data set. This partial sampling also allowed for comparisons of covariance structure at two different taxonomic levels: the 19 genera evaluated and the 77 individual species out of which we sampled at least eight individuals.

Some sources of variation encountered in our data set were not of immediate interest for this study. The proper representation of the covariance structure for any biological group is one that removes sources of variation and covariation that are not directly related to the genotype–phenotype map per se. For example, if one were to estimate a P-matrix at the genus level without taking into account species-level differences, a substantial proportion of the correlation observed between traits might result from differences between the two species and not directly from the underlying genetic architecture. Therefore, species-level differences were explored through multivariate analysis of variance tests based on Wilk's lambda statistic and considered significant at P < 0.05. As a result, we identified and controlled for significant sources of variation when estimating the residual pooled within-group phenotypic variance/covariance and correlation matrices for each taxa. This procedure was carried out using the general linear model routine in SYSTAT Software (2000).

Patterns of phenotypic variance/covariance and correlation among traits

We used a matrix analysis approach combined with the random skewers method (following Marroig & Cheverud, 2001) to assess the evolutionary pattern of phenotypic covariation and integration among floral traits in Bignonieae. As previously noted, pooled within-group phenotypic matrices (W) obtained from simple phenotypic matrices (P) can be used as G surrogates under specific assumptions, especially whenever those matrices do not differ significantly across large monophyletic groups (Cheverud, 1996; Marroig & Cheverud, 2001; Oliveira et al., 2009; see 'Discussion'). Thus, to test the hypothesis of similarity on the phenotypic variance/covariance structure across Bignonieae taxa and their implication for G assessment (Waitt & Levin, 1998; Marroig & Cheverud, 2001), we compared the pooled within-group phenotypic VCV and correlation matrices among genera (for the main and the partial data sets) and species (for only the partial data set).

Covariance patterns were compared using the random skewers method, which is a direct extension of the multivariate evolutionary response equation for natural selection (Lande, 1979). This method compares the responses of a pair of VCV matrices to a given selection vector; if the responses are statistically similar, then the correspondent matrices are also similar (Cheverud & Marroig, 2007). The cosine of the angle formed between any two vectors is a measure of their correlation. The resultant response vectors are then correlated, and the average across a large number of vectors measures the similarity between the matrices compared. In our case, we analysed the responses of each pair of VCV matrices to 1000 random selection vectors, extracted from a uniform distribution and normalized to a length of one (Cheverud & Marroig, 2007). Because each random selection vector applied on both matrices is equal, any differences in the orientation of the responses would be due to differences in the matrices compared. Therefore, the correlation between the response vectors is a measure of matrix similarity (Marroig & Cheverud, 2001; Cheverud & Marroig, 2007). Values of similarity > 0.72 or < −0.72 are significant at P < 0.001, and values > 0.8 or < −0.8 are significant at P < 0.0001 for 16-element vectors.

We evaluated the overall phenotypic integration among floral traits from the correlation matrices for each genus. These correlation matrices were constructed by calculating Pearson's correlation coefficient between traits. We then compared the correlation matrices using matrix correlation, that is, a measure of the strength of the association between matrices that ranges from −1 to +1, with zero indicating no similarity in the patterns of correlation between genera. The statistical significance of this association was evaluated by the Mantel test (Manly, 1986), which compares the original correlation with the distribution of correlations among resampled matrices. This distribution was created from 10 000 permuted matrices, as derived through random permutation of columns and associated rows of the matrices. Matrices were considered to be similar whenever the original correlation was greater than 99% of the correlations distribution of the permuted matrices (Cheverud et al., 1989).

Matrix repeatability

Given the errors that can affect the estimation of VCV and correlation matrices, we carried out corrections based on matrix repeatability (Cheverud, 1996). The adjusted matrix correlation was then estimated by radj = robs/rmax, where robs is the observed matrix correlation and radj is the adjusted matrix correlation (Marroig & Cheverud, 2001). For correlation matrices, this correction considers the maximum possible correlations between two matrices (rmax) given a certain number of samples in each matrix, defined as rmax = (t1 + t2)½, where t1 and t2 represent the repeatability of matrices 1 and 2 (Marroig & Cheverud, 2001). VCV matrices are not suitable to tests of matrix similarity based on randomization procedures because of scale-dependency of variance values. Therefore, we used the procedure described by Marroig & Cheverud (2001), which is based on a Monte Carlo approach. VCV matrix repeatability is the average vector correlation between simulated responses of the original and resampled matrices. Observed vector correlations were thus adjusted for repeatability as described (Marroig & Cheverud, 2001).

Magnitudes of correlation among traits

We evaluated the overall magnitudes of correlation between Bignonieae floral traits through r2, which is the average of the squared correlation coefficients of the off-diagonal elements of the correlation matrix (Cheverud et al., 1989). It is important to note that the overall correlation index (r2) does not account for subsets of within-flower associations among traits. Hence, similar values of overall integration can be achieved through different trait associations (Steppan, 2004). Confidence intervals and significant differences in r2 among genera were assessed by (i) bootstrapping the original data for each genus, (ii) generating the corresponding correlation matrices and (iii) estimating the r2 and respective standard deviations. Pairwise differences in r2 were used to construct dissimilarity matrices among genera and species. Dissimilarity matrices were compared through the Mantel test to the matrix of similarity among the previously obtained correlation matrices (i.e. those obtained to assess the pattern of correlations among traits). These analyses were conducted to assess whether the magnitude of the mean trait correlation was associated with the pattern of trait correlation.

Effect of evolutionary divergence on the patterns and magnitudes of phenotypic integration

We used Mantel tests to evaluate the phylogenetic effect on the covariance structure among floral traits between genera of Bignonieae. In spite of the limited power of the Mantel test when applied for phylogenetic analyses, it remains the only test that allows the comparison between two or more matrices (Harmon & Glor, 2010). More specifically, we compared the phylogenetic distance matrix among groups (i.e. species and genera) with the matrices of (i) similarity in VCV patterns, (ii) similarity in correlation patterns and (iii) dissimilarity in r2 for each pair of genera. Three main patterns could arise from these analyses (i) the similarity of phenotypic matrices might be proportional to the time of divergence between groups, (ii) differences between taxa could not be correlated with the time of divergence among groups, revealing a highly labile and variable pattern of phenotypic integration or (iii) the pattern of phenotypic integration could be highly constrained and constant among groups, not correlating with divergence times. Under neutrality, a positive association between temporal divergence and pattern/magnitude of phenotypic integration among clades is expected, given that neutral theory assumes that phenotypic evolution is mainly driven by genetic drift (Marroig & Cheverud, 2001). Alternatively, a constant P structure at different taxonomic levels and along considerable evolutionary periods might be attributed to strong stabilizing selection acting on G (Marroig & Cheverud, 2001).

Effect of floral morphology on the patterns and magnitudes of phenotypic integration

To assess whether the differences in floral morphology influence the overall phenotypic integration in Bignonieae, we reorganized our whole data set based on floral morphology instead of taxonomic status. That is, we regrouped species according to Gentry's (1974) floral ‘types’ as reviewed in Alcantara & Lohmann (2010), which also identified a ‘mixed’ floral type category, including species with features from more than one floral type (Fig. 1, see Appendix S1 in Alcantara & Lohmann, 2010). Given that representatives of this ‘mixed’ floral type vary in their overall morphology, we also categorized those species to account for their morphological differences. The six ‘mixed-type’ species sampled were grouped following Alcantara & Lohmann (2010), with two species showing morphologies with characteristics of Anemopaegma- and Martinella-type flowers, two species showing characteristics of Anemopaegma-, Martinella- and Tanaecium-type floral morphologies and another two with characteristics of Anemopaegma- and Pithecoctenium-type flowers. The two categories including Anemopaegma/Martinella and Anemopaegma/Pithecoctenium mixed-type flowers were not analysed because of the low sample sizes (n = 10 and 13, respectively). Thus, only the ‘mixed-type’ flowered species with Anemopaegma-, Martinella- and Tanaecium-type flower traits (n = 16), F. triplinervia and S. inaequilaterum, were included in these analyses.

We estimated within-type pooled VCV and correlation matrices in SYSTAT Software (2000), accounting for differences attributable to species variation, using the same methods employed in the taxonomically arranged analyses, as described above. We also evaluated the similarities among within-type pooled VCV and correlation matrices and estimated the morphological integration index (r2). The analyses that included taxa grouped by floral types aimed to evaluate whether shared morphology among unrelated species, as opposed to shared phylogenetic history, as evaluated in the previous section, could lead to higher intertrait correlations. If similarities in VCV and correlation patterns within floral types are higher than those within genera, this would suggest that convergence in floral morphology is more important for the generation of phenotypic integration among floral traits of Bignonieae than species relatedness, leading to lower (nonsignificant) similarities in VCV and correlation comparisons between different floral types.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Trait measurements

The repeatability of measurements of all 16 traits varied from 0.94 to 1.0, with a mean value of 0.98 (SD = 0.023). Given the high repeatability of our data, we used average values in all analyses. For each measured trait within each analysed group, that is, genera and species, raw values generally presented a continuous, slightly skewed distribution, which was normalized by the ln transformation. No genus exhibited more than one trait with non-normal distribution after transformation, and therefore, departure from normality is not expected to have impacted our analyses. Total coefficients of variation of most traits ranged from 0.16 to 0.25; the exceptions were calyx length, calyx median diameter, location of the insertion of the smallest stamens, length of the smallest stamens, anther width and stigma width, all of which varied from 0.31 to 0.37 (Appendix S2).

Patterns of phenotypic covariance and correlation among traits

VCV and correlation matrix similarities were high in most of the pairwise generic comparisons, indicating that variance/covariance and correlation structures are largely shared by Bignonieae genera. When only VCV matrices were regarded, all comparisons resulted in significant and high similarities (Table 1). General patterns revealed by the main and the partial generic data sets were the same; we thus show the results for the whole data set (results for partial data set including 7 traits and 19 genera are shown in Appendix S3, S4 and S5).

Table 1. Similarity matrix among 16 genera of Bignonieae (‘whole data set’, including 16 characters): VCV (above diagonal) and correlation (below diagonal) corrected by matrix repeatability. Boldface numbers indicate significant associations between correlation matrices at 0.01; all paired correlations between VCV matrices were significant at 0.001
 12345678910111213141516
1. Adenocalymma 0.870.850.940.860.840.810.850.810.810.840.780.790.830.920.88
2. Amphilophium 0.60  0.970.970.980.980.980.980.960.990.980.980.980.980.990.98
3. Anemopaegma 0.73 0.58  0.980.980.980.980.970.970.970.970.970.980.960.970.96
4. Bignonia 0.90 0.75 0.53  0.980.990.980.970.970.960.970.980.980.960.970.97
5. Cuspidaria 0.85 0.53 0.41 0.63  0.990.980.980.980.970.980.990.990.970.980.97
6. Dolichandra 0.97 0.350.550.30 0.82  0.990.980.980.970.980.990.990.970.980.97
7. Fridericia 0.94 0.40 0.79 0.60 0.83 1.00  0.980.980.960.980.990.980.970.980.97
8. Lundia 0.93 0.71 0.71 0.54 1.00 1.00 1.00  0.980.960.980.980.990.970.980.96
9. Mansoa 0.98 0.93 0.92 0.97 0.64 0.83 0.71 0.71  0.960.980.980.980.980.980.97
10. Martinella 0.56 0.260.480.120.37 1.00 0.54 0.39 0.67  0.960.960.970.950.960.95
11. Neojobertia 0.97 0.62 0.59 0.71 0.70 1.00 1.00 0.85 1.00 0.65 0.980.980.980.980.98
12. Pyrostegia 1.00 0.41 0.59 0.54 1.00 0.85 0.97 1.00 0.65 0.480.63 0.990.970.990.97
13. Stizophyllum 0.92 0.40 0.57 0.75 1.00 0.65 0.79 0.93 0.99 0.23 1.00 0.86  0.970.980.97
14. Tanaecium 0.93 0.80 0.47 0.85 0.95 0.72 0.81 1.00 0.82 0.34 0.69 0.89 0.82  0.970.97
15. Tynanthus 0.74 0.29 0.86 0.580.570.66 1.00 0.650.540.500.400.620.250.56 0.97
16. Xylophragma 1.00 0.83 0.96 1.00 0.530.53 0.70 0.58 1.00 0.400.610.710.90 1.00 0.53 

Concerning correlation matrix comparisons, most paired correlation matrices between genera rejected the null hypothesis of no association between matrices: 67.5% (81 of 120) against 32.5% of the comparisons that did not yield significant similarities (Table 1). Almost all correlation matrix comparisons resulted in statistical similarity in Adenocalymma, Fridericia, Tanaecium and Mansoa. In other genera (e.g. Anemopaegma, Amphilophium, Cuspidaria, Dolichandra, Neojobertia and Pyrostegia), at least 60% of the comparisons resulted in significant similarity. The lower similarities were concentrated in the comparisons involving Xylophragma, Martinella and Tynanthus. The number of nonsignificant similarities dropped to 14.1% (11 of 78) when these genera were excluded from the analyses. The main pattern emerging from these data is that VCV and correlation patterns are very similar among all Bignonieae genera.

In most cases, the similarity between VCV matrices was higher than the similarity between correlation matrix counterparts (Table 1). Matrix repeatability was also higher in VCV than in correlation matrices (Table 2). In contrast to VCV matrix repeatability, correlation matrix repeatability was positively associated with sample size, indicating the effect of small sample sizes for the correlation matrix estimations (Table 2).

Table 2. VCV and correlation matrix repeatability at the genus level in Bignonieae for the whole data set (including 16 genera and all 16 traits). R vs. N (P) indicates the correlation between matrix repeatability and sample size, followed by the respective probabilities
GenusNRVCVRcor
1. Adenocalymma530.890.67
2. Amphilophium490.710.65
3. Anemopaegma240.80.55
4. Bignonia520.660.51
5. Cuspidaria400.750.53
6. Dolichandra210.810.31
7. Fridericia1290.840.79
8. Lundia450.760.75
9. Mansoa280.80.4
10. Martinella170.890.49
11. Neojobertia170.760.28
12. Pyrostegia250.910.52
13. Stizophyllum160.90.6
14. Tanaecium540.760.69
15. Tynanthus210.660.29
16. Xylophragma160.70.18
Correlation R vs. N (P) 0.02 (0.94)0.7 (0.002)

Patterns of similarity of VCV between paired species also resulted in high similarities and high repeatability (data not shown). Similarities in correlation patterns were more variable and generally lower between species than between genera, with 10.9% of the correlations significant at 0.01. At this level of comparison, the repeatability of both VCV and correlation matrices was influenced by low sample sizes (respectively: r2 = 0.57, P = < 0.0001 and 0.26, P = 0.022).

Magnitudes of correlation among traits

Overall integration index (r2) ranged from 0.065 to 0.22 (Fig. 3). Pairwise differences in r2 were unrelated to the patterns of similarity in VCV and correlation matrices (Table 3). The highest r2 values were obtained for genera with (i) all representatives showing the same floral morphology (e.g. Anemopaegma, Pyrostegia and Xylophragma), (ii) only one species, in Lundia, showing a discrepant morphology and (iii) all species classified as ‘mixed floral type’ (Stizophyllum; Fig. 3). Amphilophium, Bignonia, Mansoa and Tynanthus yielded the lowest r2 values.

Table 3. Pairwise correlations between similarity matrices in the patterns of covariation (VCV), correlation (CORR), phylogenetic distance (PD) and differences in the magnitude of integration (r2) among 16 genera of Bignonieae. All correlations are nonsignificant at 0.05
 VCVCORRr2PD
VCV1   
CORR1  
r2−0.066−0.0771 
PD−0.3070.022−0.0191
image

Figure 3. Generic level phylogeny of Bignonieae modified from Lohmann (2006) to represent the 16 genera analysed for 16 traits representing the four floral whorls. Terminal labels indicate the number of species sampled for each genus and their respective floral morphologies. Black bars represent r2 values estimated for each genus.

Download figure to PowerPoint

Effect of evolutionary divergence and floral morphology on patterns and magnitudes of phenotypic integration

Mantel tests indicated that phylogeny does not influence the structure of covariance and correlation among floral traits in Bignonieae. Similarly, the magnitude of correlation was also unaffected by the phylogenetic distance between taxa (Table 3).

Patterns of phenotypic integration between floral morphologies, that is, species grouped according to morphological ‘type’, instead of phylogenetic relatedness, were similar to those reported for the patterns of phenotypic integration among genera. Despite the low values involving similarities of VCV matrices of Amphilophium-type flowers, almost all values of similarity in the patterns of phenotypic integration were significant (Table 4). The magnitudes of trait association within floral types did not differ from magnitudes estimated for individual genera (t22 = 0.42, P = 0.68, Fig. 4). Overall, no correlation was observed between the pattern and magnitude of phenotypic integration estimated for floral types (Mantel test: r = −0.12, P = 0.69). The lowest r2 values were associated with the Tynanthus-type and the Cydista-type flowers. In contrast, Amphilophium-, Martinella- and Tanaecium-type flowers exhibited high values of r2. The ‘mixed floral type’, that is, flowers presenting traits of Anemopaegma-, Martinella- and Tanaecium-type flowers, was associated with the highest magnitude of integration among floral traits.

Table 4. Similarity in VCV and correlation matrices, corrected by matrix repeatability, between paired floral ‘types’ of Bignonieae: VCV (above diagonal) and correlation (below diagonal). Boldface values in the diagonal represent the repeatability of the correlation values and VCV values, respectively. Species effects were removed from the original data set whenever necessary, and matrix similarities were estimated from the pooled matrices. Significant associations between correlation matrices are indicated by * (P < 0.05) and ** (P < 0.001)
 12345678
1. Amphilophium- type flowers 0.54/0.76 0.6470.6440.7580.6760.7380.7050.801
2. Anemopaegma-type flowers0.513* 0.88/0.83 0.9600.9430.9600.9770.9580.971
3. Cydista-typeflowers0.878*0.755** 0.32/0.85 0.9600.9750.9490.9780.978
4. Martinella-type flowers0.300*0.838**0.319 0.73/0.70 0.9520.9480.9650.954
5. Mixed-type flowers0.105*0.941**0.817*0.747** 0.34/0.83 0.9580.9690.969
6. Pithecoctenium-type flowers0.855**0.795**0.710*0.466*0.683* 0.49/0.76 0.9510.960
7. Tanaecium-type flowers0.581*0.882**1.0**0.802**1.000**0.917** 0.43/0.70 0.976
8. Tynanthus-type flowers0.433*0.956**0.877**0.760**0.589*0.876**0.677** 0.50/0.54
image

Figure 4. Ordered values of r2 for the eight floral type morphologies of Bignonieae analysed. Type morphologies are illustrated in Fig. 1.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Our findings indicate that in spite of the floral diversification in Bignonieae, there is considerable stasis in the pattern of phenotypic integration in flowers of the whole tribe. To the best of our knowledge, this is the first study to assess the covariation among floral traits in a broad phylogenetic context, using analyses that were explicitly developed to explore the evolutionary patterns of phenotypic integration.

Patterns of phenotypic covariance and correlation among traits

The similarities in the patterns of VCV among floral traits between genera and species of Bignonieae are surprising. Phenotypic covariance among traits results from the interplay between genetic and environmental covariances. Hence, constancy in the phenotypic matrix of VCV (P) is unlikely without constancy of the genetic or environmental components (Lande, 1979). If the patterns of phenotypic covariation among traits are stable across several species or higher taxa, the most likely explanation is that genetic variance/covariance matrices (G) are also stable (Lande, 1979). An alternative explanation would be that environmental effects have been reacting to changes in genetic patterns, compensating exactly so that the phenotype remained unchanged, which is highly improbable (Marroig & Cheverud, 2001). When G and P are similar in some structural aspects, the within-group pooled phenotypic matrices (W) could still be used to assess G in many simulated scenarios (Prôa et al., 2012). Furthermore, a meta-analysis of plant studies encountered higher similarities between genetic and phenotypic VCV patterns than those encountered for animals (Waitt & Levin, 1998). These findings, coupled with our results, led us to conclude that W represents a suitable surrogate for G among floral traits in Bignonieae. This study is the first to suggest that P and W are indeed a good substitute for G in plant studies at taxonomic levels higher than species. However, this suggestion should be taken with caution given that nearly no information is available on the structure of G in Bignonieae.

The similar patterns of VCV in Bignonieae taxa imply that even very different morphologies, such as Tanaecium- and Tynanthus-type flowers (Fig. 1), share a similar VCV structure. The VCV patterns remain similar for all data sets analysed, at generic and specific levels. The stasis in VCV structure is even more surprising given the relatively old age of Bignonieae (approximately 50 Myr; Lohmann et al., 2013), suggesting that variance/covariance among traits remained relatively unchanged throughout the history of diversification of this group. VCV matrix comparisons yielded higher similarities than their correlation counterparts in Bignonieae, a pattern already reported in other studies (Cheverud et al., 1989; Marroig & Cheverud, 2001; Marroig et al., 2009; Oliveira et al., 2009). This pattern is attributed to less precise estimations of correlation matrices when compared to VCV matrices, especially in taxa with lower sample sizes (Cheverud & Marroig, 2007). This could explain the correlation between repeatability and sample sizes at species-level analyses in Bignonieae. If low sample sizes were to affect the estimation of correlation matrices, the stasis in the patterns of phenotypic integration of Bignonieae flowers would be even more remarkable.

Importantly, the broad similarity of trait correlation/covariation structure in Bignonieae implies that taxa that share similar patterns of G should respond similarly to selection (Marroig & Cheverud, 2001). Genetic correlations among morphological traits can limit the number of axes of the phenotypic space along which populations respond to selection (Kirkpatrick, 2010). The unequal distribution of genetic variation available for selection may cause the presumed ‘prohibitive’ morphologies in Bignonieae (Alcantara & Lohmann, 2010). In such case, the correlated changes reported for discrete floral traits (Alcantara & Lohmann, 2010) would be the evolutionary result of genetic lines of least resistance existent in this lineage. Genetic lines of least resistance represent the multivariate direction of greatest additive genetic variance, estimated from the major axis or the dominant eigenvector of the genetic covariance matrix, and the first stages of adaptive differentiation seem to occur mainly along this axis (Stebbins, 1974; Schluter, 1996).

Magnitudes of correlation among traits

In contrast to the relative stasis in VCV and correlation patterns, the overall magnitudes of trait associations (r2) were more variable across Bignonieae taxa, ranging from 0.06 to 0.22. Such amplitude of variation indicates that the magnitude of intertrait correlations is more labile than their pattern. The highest r2 values were detected in genera including the same floral morphology or with only one discrepant species (Fig. 3). On the other hand, some genera that showed the lowest r2 values, that is, Amphilophium, Bignonia and Tanaecium, also include species with remarkable variation in flower morphology. These observations suggest that lineages with lower magnitudes of trait correlation would be more prone to evolve variable floral morphologies. However, Mansoa and Tynanthus, which include species with the same floral morphology (Anemopaegma-type and Tynanthus-type, respectively), also showed low values of r2. Other studies at lower taxonomic scales have, in general, reported low magnitudes of phenotypic integration in flowers (Herrera et al., 2002; Pérez et al., 2007; Ordano et al., 2008; Rosas-Guerrero et al., 2011). Empirical and theoretical studies have highlighted that the evolutionary response of organisms that share VCV structure, but differ in the magnitude of associations among traits, varies greatly (Lande, 1979; Wagner & Altenberg, 1996; Merilä & Björklund, 2004; Marroig et al., 2009; Oliveira et al., 2009). Higher magnitudes of correlation allow faster responses if selective pressures act in the same direction as existent correlation patterns. On the other hand, if selective regimes act in directions different from the existent covariation, the constraints imposed by G are stronger with higher magnitudes of trait associations.

Effect of evolutionary divergence on patterns and magnitudes of phenotypic integration

The lack of association between covariance/correlation patterns and phylogenetic distance illustrates that the VCV is independent of the divergence time between taxa. Yet, the high similarity in the pattern of VCV and the independence of phylogeny remain at specific and generic level (Appendix S5). Similar results have also been documented in mammals (Marroig & Cheverud, 2001; Estes & Arnold, 2007; Arnold et al., 2008). These results suggest that stochastic processes alone, that is, genetic drift, should not lead to the observed similarity in covariance patterns. Several theoretical studies have attributed the stasis in VCV along different taxonomic levels to stabilizing selection (Lande, 1979; Cheverud, 1996; Estes & Arnold, 2007; Revell, 2007). Correlational selection and pleiotropic mutations can sustain the stability of G through time (Revell, 2007). Our data set comprised floral traits that share a large proportion of developmental pathways; therefore, it is reasonable to assume that mutations in some of the underlying genes might affect several of those traits. Moreover, shared development and/or function among floral traits might also facilitate evolution by correlated selection (Cheverud, 1996). Considering that floral traits interact to perform several functions, one can expect that those traits would be under similar stabilizing selection regimes (Wagner, 2010). Thus, the remarkable stability of W, and putatively of G, evidenced in Bignonieae may have resulted from stabilizing selection. Unfortunately, we still lack evidence concerning model parameters to explicitly test the effect of selection versus drift in the shared VCV structure of Bignonieae, as proposed by Ackermann & Cheverud (2002; Prôa et al., 2012).

Additionally, the dissociation between the magnitudes of association among traits and the phylogenetic distance among taxa suggests that the evolution of magnitudes is not directly correlated with lineage diversification. This finding indicates that floral evolution is either driven by different dynamics or responds differently to evolutionary forces that determine cladogenesis. Discrepant evolutionary rates and contrasting phylogenetic signals to the same floral traits evaluated here have been documented elsewhere (Alcantara & Lohmann, 2011), also suggesting the action of complex evolutionary dynamics. These complex evolutionary dynamics might have resulted in great lability of floral morphology, even in the presence of a shared pattern of phenotypic integration. In sum, the magnitude of trait associations, rather than the pattern, has driven the diversification of floral form in Bignonieae.

Effect of floral morphology on patterns and magnitudes of phenotypic integration

VCV and correlation patterns of species grouped according to floral morphology were highly similar in all cases and did not differ between taxonomic-based and morphology-based analyses. This indicates that grouping different floral morphologies in our taxonomic-based analyses does not affect the structure of the correlations encountered. The different floral morphologies represented in Bignonieae have likely evolved repeatedly from the same floral morphology (Alcantara & Lohmann, 2010), and the parallel evolution of homoplastic floral types in different genera might have been facilitated by a shared, highly correlated G structure. However, such results also indicate that the evolution of homoplastic floral types in different genera of Bignonieae does not imply convergence or higher similarity in the patterns of correlation among homoplastic morphologies. Levels, that is, magnitude, of phenotypic integration in flowers have been hypothesized to be under pollinator and floral antagonistic selection (Berg, 1960; Strauss & Whittall, 2006; Ordano et al., 2008). Overall, specialization might increase the morphological complexity of a given organ and, consequently, the emergence of modular sets with highly integrated traits (Wagner & Altenberg, 1996; Wagner, 2010). The associations among particular sets of floral traits, highly intercorrelated within the whole set of floral traits, have emerged from several studies (Pérez et al., 2007; Ordano et al., 2008; Rosas-Guerrero et al., 2011). It would be important to search for functional associations among subsets of floral traits of species of Bignonieae with different selective pressures in future studies to better characterize the functional floral modules in this group, as well as the genetic lines of least resistance that have allowed the adaptive changes to occur in this tribe.

Conclusions and perspectives

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Overall, the magnitude of association among traits varied across Bignonieae taxa, indicating that their differences have evolved independently of phylogenetic history, in spite of the stasis in the patterns. Whenever different floral morphologies were compared, the patterns of VCV and correlation among traits remained significantly similar. Broad-scale studies revealed a similar pattern of overall stasis in covariance structure and evolution of magnitude associated with marked diversity in skull morphology in mammals (Marroig & Cheverud, 2001; Oliveira et al., 2009; Porto et al., 2009). On the other hand, a study on floral traits of four Rosaceae species concluded the evolution of covariance patterns in a narrow phylogenetic scale (Ordano et al., 2008), a result also suggested in the genus Ipomoea (Rosas-Guerrero et al., 2011). These results are in agreement with other plant studies at the population and species levels (Herrera et al., 2002; Pérez et al., 2007). It is important to note, however, that those studies have analysed phenotypic integration from a different perspective and did not employ the random skewers method. Furthermore, most of these analyses have tested for the equality of matrices instead of searching for a significant pattern of similarity (see 'Discussion' in Steppan, 2004; Cheverud & Marroig, 2007). Indeed, some of these studies refer to ‘pattern’ of phenotypic integration as the CVI or r2 index–average measures of the covariance or correlation among floral traits, which are in fact measures of overall intertrait correlation. Thus, our results do not contradict the cumulative evidence of a labile evolution of intertrait correlations at lower phylogenetic scales; instead, we suggest that the observed lability of traits association at species level may evolve even when VCV structure is stable.

Shared similarity in G patterns provides empirical support for the existence of a conserved underlying structure that permits variability in morphological traits (Jamniczky, 2008). The covariation structure in skulls of several groups of mammals conserved across large morphological and phylogenetic distances has been suggested as the result of developmental processes allowing wide variability in phenotypic expression within a conserved framework (Marroig & Cheverud, 2001; Steppan et al., 2002; Jamniczky & Hallgrímsson, 2009; Porto et al., 2009). Whereas this shared pattern of phenotypic integration maintains coherence, identity and functionality of complex organs, it also allows for their evolution (Jamniczky, 2008). Our data suggest a conserved, but changeable, morphological structure allowing flower diversification in Bignonieae. The characterization of VCV patterns and morphological lability over time in other plant groups is central for a better understanding of the evolvability of flowers and other complex organs in angiosperms as a whole. If a conserved VCV structure were to be shown in flowers of distantly related taxa, it would imply the important role of developmental processes for the generation of morphological variation in plant reproductive organs (Diggle, 1992). Yet, the establishment of a taxonomic range in which VCV remains unaltered may have a direct impact for systematic botanists as it may suggest that the long-recognized changes in the basic floral patterns of various plant families might be associated with changes in G. This observation makes us wonder about what would determine G evolution and, consequently, the origin of plant families. Our data indicate that morphological diversification occurred without changes in the Bignonieae flower's bauplan, as bignon flowers are still easily recognizable as flowers of Bignoniaceae. We are very curious to study the patterns of VCV in other large taxonomic groups of angiosperms, which might bring important insights for a better understanding of the origin of body patterns that make plant families recognizable. Consequently, the study of correlation and VCV patterns might open a new perspective for the investigation of the origin of plant families.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

The authors thank Diana Carneiro for the illustrations in Figs 1 and 2; R. Ree, A. Hipp and B. Igic for excellent discussions; and A. Nogueira, L.B. Klaczko, L.P. de Queiroz, M. Pace and M. Souza-Baena for comments on earlier version of this manuscript. The two anonymous reviewers have also improved the final version of this paper. We are also grateful to curators of the herbaria who allowed access to herbarium specimens, specifically: A. Reis (HBR), B.M. Thiers (NY), G. Hatschbach (MBM), I. Cordeiro (SP), J. Solomon (MO), J.R. Pirani (SPF), R. Forzza (R) and W. Oliveira (UEC). This article is part of the Ph.D. thesis of S.A., which was supported by FAPESP (Grant 06/59916-0) and MBG (Elizabeth E. Bascom Fellowship).

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions and perspectives
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jeb12228-sup-0001-AppendixS1-S5.docWord document2928K

Appendix S1 Morphological data used in this study. Measurements are shown in mm and untransformed.

Appendix S2 Penalized-likelihood tree with 102 species used for the analyses carried out in this paper; branch lengths are proportional to time.

Appendix S3 Matrix of similarity among 19 genera of Bignonieae (partial dataset, including 7 characters): VCV (above diagonal) and correlation (below diagonal), corrected by matrix repeatability.

Appendix S4 VCV and correlation matrix repeatabilities at the genus level in Bignonieae for the partial dataset (including 19 genera and 7 of the 16 traits analyzed).

Appendix S5 Pairwise correlation between matrices of similarity in the patterns of covariation (VCV), similarity in the patterns of correlation (CORR), phylogenetic distance (PD), and differences in the magnitude of integration (r2) among 19 genera (below diagonal) and 88 species of Bignonieae (above diagonal).

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.