Phenotypic plasticity triggers rapid morphological convergence

Abstract

Phenotypic convergence, the independent evolution of similar traits, is ubiquitous in nature, happening at all levels of biological organizations and in most kinds of living beings. Uncovering its mechanisms remains a fundamental goal in biology. Evolutionary theory considers that convergence emerges through independent genetic changes selected over long periods of time. We show in this study that convergence can also arise through phenotypic plasticity. We illustrate this idea by investigating how plasticity drives Moricandia arvensis, a mustard species displaying within-individual polyphenism in flowers, across the morphological space of the entire Brassicaceae family. By compiling the multidimensional floral phenotype, the phylogenetic relationships, and the pollination niche of over 3000 Brassicaceae species, we demonstrated that Moricandia arvensis exhibits a plastic-mediated within-individual floral disparity greater than that found not only between species but also between higher taxonomical levels such as genera and tribes. As a consequence of this divergence, M. arvensis moves outside the morphospace region occupied by its ancestors and close relatives, crosses into a new region where it encounters a different pollination niche and converges phenotypically with distant Brassicaceae lineages. Our study suggests that, by inducing phenotypes that explore simultaneously different regions of the morphological space, plasticity triggers rapid phenotypic convergence.

Introduction

Phenotypic convergence, the independent evolution of similar traits in different evolutionary lineages, is ubiquitous in nature, happening at all levels of biological organizations and in most kinds of living beings (1–3). Convergent evolution plays a fundamental role in how evolutionary lineages occupy the morphological space (2, 4). The expansion of lineages across the morphological space is a complex process resulting from the ecological opportunities emerging when species enter into different regions of the ecospace and face new ecological niches (5, 6). When this occurs, divergent selection on some phenotypes makes lineages to diversify phenotypically, boosting morphological disparity, triggering a morphological radiation and eventually filling the morphospace (7, 8). Because the ecological space saturate as lineages diversify (9), unoccupied regions become rare in highly diversified lineages (10). Under these circumstances, entering into a new region usually entails sharing it with other species exploiting the same ecological niche (2, 10, 11). In this situation, independent lineages tend to evolve similar phenotypes through convergent evolution (2, 4). In diversified lineages occupying a saturated morphospace, divergent and convergent evolution are ineludibly connected (10, 12), and both processes contribute significantly to shape the geometry of the morphospace occupation (4, 11).

Uncovering the mechanisms triggering convergence remains a fundamental goal in biology. Evolutionary theory shows that convergent phenotypes emerge from several genetic mechanisms, such as independent mutations or gene reuse in different populations or species, polymorphic alleles, parallel gene duplication, introgression or whole-genome duplications, that are selected over long periods of time (13–15). Under these circumstances, the origin of morphological convergence is mostly slow, occurring over evolutionary time and associated with multiple events of speciation and cladogenesis (11). It is increasingly acknowledged, however, that phenotypic plasticity might elicit the emergence of novel phenotypes with new adaptive possibilities, which may be beneficial in some contexts (16, 17). Under these circumstances, plasticity may behave as a facilitator for evolutionary novelty and diversity, shaping the patterns of morphospace occupation (16, 18–21). In this study, we provide compelling evidence showing that phenotypic plasticity also plays a prominent role in the emergence of convergent phenotypes. By inducing the production of several phenotypes, plasticity may cause the species to explore different regions of the morphospace almost simultaneously (18, 19). This opens the opportunity for plastic species to diverge from their lineages and converge with the species already located in other morphospace regions. We illustrate this idea by investigating how plasticity drives Moricandia arvensis, a species exhibiting extreme polyphenism in flowers (18), across the morphological space of the entire Brassicaceae family. Moricandia arvensis displays within-individual floral plasticity, with flower morphs varying seasonally on the same individual (18). By studying the multidimensional floral phenotypes, the phylogenetic relationships, and the pollination niches of over 3000 Brassicaceae species, we demonstrate that phenotypic plasticity makes the flowers of this mustard species to diverge from its ancestors and close relatives, to cross into a new region of the ecospace, and to converge morphologically with distant Brassicaceae lineages. This finding has great implications, suggesting that plasticity might not only promote the evolution of novelties and morphological divergence (16, 17, 20, 21) but can also provide an alternative explanation to the pervasiveness of convergence in nature.

Results

Plasticity-mediated floral disparity and divergence

Changes in temperature, radiation and water availability induce the production of different types of flowers by the same M. arvensis individuals; large, cross-shaped lilac flowers in spring but small, rounded, white flowers in summer (18). To quantify the magnitude of floral disparity between these two phenotypes of M. arvensis, we first assessed floral disparity for the entire mustard family. Brassicaceae is one of the largest angiosperm families, with almost 4000 species grouped in 351 genera and 51 tribes (7, 22–24). We determined the magnitude and extent of floral disparity among 3140 plant species (approx. 80% of the accepted species) belonging to 330 genera (94% of the genera) from the 51 tribes. Because we were interested in floral characters mediating the interaction with pollinators, we recorded for each studied species a total of 31 traits associated with pollination in Brassicaceae (Supplementary Data 1, Methods). We used the resulting phenotypic matrix to generate a family-wide floral morphospace. We first run a principal coordinate analysis (PCoA) to obtain a low-dimensional Euclidean representation of the multidimensional phenotypic similarity existing among the Brassicaceae species (25). Because the raw matrix was composed of quantitative, semi-quantitative and discrete variables, PCoA was based on Gower dissimilarities (25). We optimized this initial Euclidean configuration by running a non-metric multidimensional scaling (NMDS) algorithm with 5000 random starts (25). The resulting morphospace (Figure 1a) was significantly correlated with the initial PCoA configuration (r = 0.40, P < 0.0001, Mantel test) and was a good representation of the original relationship among the species (R² > 0.95, Stress = 0.2, Figure 1b). The distribution of the species across the morphospace was significantly associated with different pollination traits (Figure S1; Table S1). Species in the central region were mostly medium-sized plants bearing a moderate to high number of small, polysymmetric white flowers with short corolla tubes, exposed nectaries and visible sepals (Figure 1a, Figure S1). Species in the bottom right corner were small or prostrate, bearing minute flowers, many time apetalous and with just 2 or 4 stamens, whereas species located in the bottom left corner were medium-sized plants with asymmetric flowers arranged in corymbous inflorescences. Plants with yellow flowers were located in the right region of the morphospace. In contrast, large plants with strongly tetradynamous androceum and large, veined, dissymmetrical to asymmetrical, pink to blue flowers with concealed nectaries, long corolla tubes and bullseyes were located in the upper left region (Figure 1a, Figure S1). Moricandia arvensis, when blooming in spring (Figure 1c), occupies this later peripheral region of the morphospace, close to other Moricandia species (purple dots in Figure 1a). However, during summertime, the individuals of M. arvensis are shorter and produce fewer, much smaller flowers with white, unveined and rounded corollas with overlapped petals and green sepals that are mostly arranged alone the floral stems (Figure 1d) (18). Due to this radical phenotypic change, the summer phenotype of M. arvensis was located in a different, more central position of the floral morphospace (Figure 1a), far away from the region occupied by the Moricandia species. As a consequence of this jump, the morphological disparity between the spring and summer phenotypes of M. arvensis, calculated as their distance in the morphospace (26), was very high (0.264). In fact, it was much higher than the average pairwise disparities among all studied Brassicaceae species (0.155 ± 0.090, mean ± s.e.m., 4,912,545 pairwise disparities) and almost 50% of the largest observed disparity (0.55) (Table S4). This outcome suggests that phenotypic plasticity prompts M. arvensis to explore two distant regions of the Brassicaceae floral morphospace simultaneously.

Figure 1. Plasticity-mediated floral disparity.

(A) Floral morphospace of the Brassicaceae, showed as the projection of 31 traits recorded in 3140 species onto two NMDS axes. The position of the spring and summer phenotypes of Moricandia arvensis is linked by a thick lilac dashed line. We have also indicated the movements across this morphospace of several species changing their phenotypes due to floral colour polymorphism (black lines), single mutations in floral colour (blue lines), changes in breeding systems (orange lines), changes in gender expression (green lines), homeotic mutations (brown lines), and plasticity (lilac lines). Numbers matching species are as follow: 1-Lobularia maritima; 2-Raphanus raphanistrum; 3-Matthiola incana; 4-Mathiola fruticulosa; 5-Erysimum cheiri; 6-Cakile maritima; 7-Matthiola lunata; 8-Marcus-kochia littorea; 9-Hesperis matronalis; 10-Hesperis laciniata; 11-Parrya nudicalis; 12-Streptanthus glandulosus; 13-Eruca vesicaria; 14-Capsella bursa-pastoris; 15-Hormathophylla spinosa; 16-Brassica napus; 17-Cardamine hirsuta; 18-Lepidium sisymbrioides; 19-Lepidium solandri; 20-Arabidopsis thaliana; 21-Boechera stricta; 22-Leavenworthia stylosa; 23-Leavenworthia crassa; 24-Pachycladon stellatum; 25-Pachycladon wallii; 26-Cardamine kokairensis; 27-Brassica rapa. (B) Shepard plot showing the goodness of fit of the NMDS ordination. (C) Moricandia arvensis in spring. (D) Moricandia arvensis in summer. (E) Magnitude of floral disparity between different taxonomic levels of Brassicaceae species. The number above each boxplot shows the number of disparities per level. We have compared this value with the disparity between spring and summer phenotypes of M. arvensis (this comparison with boxplots in red is statistically significant at P < 0.05, in orange is marginally significant at P < 0.1, and in grey is non-significant).

To know how intense is the plasticity-mediated M. arvensis disparity, we compared its value with the disparity values observed at different taxonomic levels within Brassicaceae. At the lowest level, discrete changes in pollination traits have been reported between individuals of the same species. In some species, this intraspecific phenotypic change is stable, like the gender polymorphism (27, 28) or the adaptive floral colour polymorphism exhibited as a response to the selective pressures exerted by certain pollinators (29, 30). In other species, discrete phenotypic changes, although affecting pollination traits, seem to be just the consequence of some singular and often unstable mutations affecting floral colour (31), the production of cleistogamous flowers (32) or changes in the expression of homeotic genes that modify the formation of the floral organs (33, 34). We compiled information on the phenotypes of the different morphs in 34 polymorphic species and calculated their values of intraspecific disparities (Figure 1a, Supplementary Data 2). Although several polymorphic species showed considerable values of between-morph disparity, they were significantly smaller than the disparity between spring and summer floral phenotypes of M. arvensis (Z-score = 5.06, P < 0.0001, Figure 1e, Table S2). We subsequently tested at what taxonomic level of Brassicaceae the disparity was equivalent to the plasticity-mediated disparity observed in M. arvensis. For this, we calculated the floral disparity between pair of species belonging to the genus Moricandia, the same genus, the same tribe, and different tribes (Methods). The plasticity-mediated disparity of M. arvensis was significantly higher than the disparity existing between the Moricandia species (0.057 ± 0.033, mean ± 1 s.e.m., Z-score = 6.27, P < 0.0001) and between the species belonging to the same genus (0.069 ± 0.055, Z-score = 3.51, P < 0.0002). It was marginally different from the disparity existing between species of different genera but the same tribes (0.150 ± 0.085, Z-score = 1.34, P = 0.089) and it was statistically similar to the disparity occurring between species belonging to different tribes (0.167 ± 0.087, Z-score = 1.11, P = 0.133, Figure 1e). These findings suggest that phenotypic plasticity allows M. arvensis individuals to jump in the morphospace longer distances than those granted by some macroevolutionary processes.

We explored whether plasticity-mediated disparity may cause evolutionary divergence by calculating the disparity of M. arvensis spring and summer phenotypes to their phylogenetic ancestors. We retrieved 80 partial phylogenies from the literature and online repositories (Methods), and assembled them into a supertree comprising 1876 taxa with information on their floral phenotype. We then projected this supertree onto the morphospace to get a family-wide phylomorphospace. We did not find evidence of phylogenetic constraints on morphospace occupation since there was not significant phylogenetic signal for the position occupied by each species (Multivariate Mantel test=0.005, P = 0.34). The family-wide phylomorphospace was very tangled (Figure 2a), with 492,751 intersections among lineages, suggesting the presence of many events of floral divergence and convergence in the evolution of Brassicaceae pollination traits (11). To calculate the disparity of the M. arvensis floral phenotypes to their ancestor, because these analyses are sensitive to the tree topology and the inferred branch lengths (26), we used four independent, time-calibrated phylogenies that included this species (Methods). The results were consistent across phylogenies (Figure 2b,c; Tables S3). The spring phenotype did not significantly diverge neither from the most recent common ancestor (MRCA) of Moricandia (Z-score = 0.36, P = 0.36) nor from its direct ancestor (Z-score = −1.24, P = 0.108). In contrast, the summer phenotype of M. arvensis diverged significantly both from Moricandia MRCA (Z-score = 2.48, P = 0.007) and from its direct ancestor (Z-score = 1.77, P = 0.038). Hence, the summer phenotype explores a region of the floral morphospace located out of its phylogenetic clade range (Figure 2b). The ancestral disparity of the summer phenotype was even significantly higher than the ancestral disparity of most other Brassicaceae species (Figure 2c). These findings suggest that phenotypic plasticity causes the appearance of a novel phenotype that diverges radically from its ancestors.

Figure 2. Phylogenetic-mediated floral divergence.

(A) Floral phylomorphospace using the supertree that includes 1876 Brassicaceae species. (B) Phylomorphospace considering only the eight Moricandia species, using the Perfectti et al.’s phylogeny (phylogeny # 1 in Table S8). (C) Floral disparity to the nearest ancestor, according to the supertree and 18 time-calibrated phylogenies (phylogeny codes in Table S8). We show the disparity between the two M. arvensis phenotypes and their direct ancestor (spring: lilac dots; summer: white dots) in those phylogenies that include Moricandia. We also show the disparities to their direct ancestors of those Brassicaceae species included in time-calibrated phylogenies of more than 45 species.

Plastic shifts in pollination niches

Evolutionary divergence is mostly associated with the occupation of new ecological niches (2, 5). Shifts between pollination niches are an important factor driving diversification in angiosperms (35), including Brassicaceae (36, 37). We investigated whether the plasticity-mediated jump of M. arvensis across the floral morphospace implicated the exploration of new pollination niches. We compiled a comprehensive database comprising 456,031 visits done by over 800 animal species from 19 taxonomical orders, 276 families and 43 functional groups to 554 Brassicaceae species of 39 tribes (Methods, Supplementary Data 3). Afterwards, we identified the pollination niches of these Brassicaceae plants and determined the niche of each M. arvensis floral phenotype by means of bipartite modularity, a complex network tool that identifies the set of plants interacting with similar groups of pollinators (18). This analysis showed that the network was significantly modular (Modularity = 0.385, P < 0.0001) and identified eight different pollination niches associated with different groups of pollinators (Figure 3a) located in different regions of the morphospace (Figure 3b, F = 44.4, P < 0.001, R² = 0.39, Adonis test; Table S4).

Figure 3. Plasticity-mediated changes in pollination niches.

(A) Outcome of the modularity analysis showing the number of pollination niches inferred, the among-niche differences in relative frequency of each pollinator functional group, and the pollinator functional groups defining the niches (n = 511 Brassicaceae species). (B) Morphospatial distribution of the eight pollination niches detected in Brassicaceae. Insect silhouettes were drawn by Divulgare (www.divulgare.net) under a Creative Commons license (http://creativecommons.org/licenses/by-nc-sa/3.0). (C) Estimate of the ancestral pollination niche of the Moricandia lineage using a stochastic character mapping inference analysis. The numbers underneath each ancestral node indicate the posterior Bayesian probability of belonging to pollination niche 5.

Because different insects visited M. arvensis in spring and summer (Table S5), this plant species shifted between pollination niches seasonally (Figure 3b). During spring, M. arvensis belonged to a niche where most frequent pollinators were long-tongued bees, beeflies, and hawkmoths (pollination niche 5 in Figure 3a) (18). This pollination niche was also shared by the other Moricandia species (Figure 3c). In contrast, during summer M. arvensis belonged to a niche dominated by short-tongued bees (pollination niche 3 in Figure 3a). This niche shift was substantial. In fact, the overlap between the spring and summer pollinator niches of M. arvensis (Czekanowski overlap index = 0.35) was significantly lower than the overlap between congeneric species of Brassicaceae (0.57 ± 0.42, Z-score = −0.51, P = 0.003). This shift even entailed the divergence from the ancestral niche of the Moricandia lineage (pollination niche 5 according to a stochastic character mapping inference, Figure 3c). The within-individual floral plasticity allows M. arvensis to exploit a pollination niche that differs markedly from that exploited by its closest relatives and that have largely diverged from the ancestral niche.

Plasticity-mediated floral convergence

A common consequence of adaptation to the same niche is convergent evolution (1, 2, 4). We explored the possibility of convergent evolution of M. arvensis with other Brassicaceae sharing either the spring niche (pollination niche 5) or the summer niche (pollination niche 3). We first checked for the occurrence of convergence among species belonging to these pollination niches. Because these analyses are extremely sensitive to the inferred branch lengths, we explored morphological convergence using three time-calibrated large (> 150 spp) phylogenies that included M. arvensis (Methods). We tested for the occurrence of floral convergence between the species belonging to each of those two pollination niches using three methods: the angle formed by the phenotypic vectors connecting the position in the floral morphospace of each pair of species with that of their most recent common ancestor (38), the difference in phenotypic distances between convergent species and the maximum distances between all other lineages (39), and the phenotypic similarity of the allegedly convergent species penalized by their phylogenetic distance (Wheatsheaf index) (40). The three methods gave similar results, indicating that floral convergence was frequent among the species belonging to any of the two studied niches, irrespective of the method and the time-calibrated tree used (Table S6). These results show that, despite the rampant generalization observed in the pollination system of Brassicaceae, species interacting with similar pollinators converge phenotypically.

Once we determined the occurrence of convergence in these two pollination niches, we assessed whether plasticity caused the evolution of morphological convergence in M. arvensis. To do so, we first assessed the convergence region of Moricandia, the region that includes the lineages converging morphologically to the Moricandia lineage. We found that this region included most species of Moricandia, the spring phenotype of M. arvensis, and several clades belonging to disparate tribes that interact with pollination niche 5, but excluded the summer phenotype of M. arvensis (Figure 4, Table S7). Afterwards, we checked whether any of the two M. arvensis floral phenotypes entered the region of the phylomorphospace defined by their pollination niches. We used the C5 index, defined as the number of lineages that cross into the morphospace region of interest from outside³⁹. This index detected between two and six convergent events towards pollination niche 5 depending on the phylogeny used (blue arrows in Figure 4a-c), but none was associated with the spring phenotype of M. arvensis. In contrast, the C5 index consistently detected that the summer phenotype of M. arvensis has converged with the species belonging to the pollination niche 3 (red arrow in Figure 4d-f). Altogether, these analyses suggest that, whereas the spring phenotype did not show any evidence of convergence, the summer phenotype of M. arvensis has converged with other distant Brassicaceae exploiting the same pollination niche.

Figure 4. Plasticity-mediated floral convergence.

Convergent lineages crossing into the region of the morphospace delimited by the pollination niche of the M. arvensis during spring (the shade convex hull) according to (A) Smith & Brown’s phylogeny, (B) Gaynor et al.’s phylogeny, and (C) Huang et al.’s phylogeny (phylogenies 2-4, respectively, in Table S8). Convergent lineages crossing into the region of the morphospace delimited by the pollination niche of the M. arvensis during summer (the shade convex hull) according to (D) Smith & Brown’s phylogeny, (E) Gaynor et al.’s phylogeny, and (F) Huang et al.’s phylogeny. Red arrows indicate the plasticity-mediated convergence, blue arrows the convergence events of the other lineages. The small purple area in all panels is the region of the floral morphospace that includes the lineages that have converged with the entire Moricandia clade according to each time-calibrated phylogeny.

Conclusions

Convergent selection exerted by efficient pollinators causes the evolution of similar suites of floral traits in different plant species (41–44). Our study shows that plasticity can promote the rapid convergent evolution of floral traits, providing an additional explanation about how pollination syndromes may evolve. Under this idea, changes in floral traits precede shifts in pollinators, as frequently observed in generalist systems (37, 45). This may explain why many pollination systems are evolutionarily labile, undergoing frequent shifts and evolve multiple times within the same lineages by diverse evolutionary pathways (35, 46).

Morphological convergence is universally acknowledged to be the result of several genetic mechanisms, such as independent mutations in different populations or species, polymorphic genes or introgression (13). We provide in this study compelling evidence suggesting that morphological convergence may also arise as a consequence of phenotypic plasticity. The role of plasticity as a mechanism favouring quick responses of organisms to novel and rapidly changing environments is already beyond doubt (17, 21, 47, 48). Its evolutionary consequences are more debated though (20, 21, 49, 50). The ‘plasticity-led evolution’ hypothesis states that selection acting on a plastic lineage may either boost its environmental sensitivity and trigger the origin of polyphenisms or alternatively may promote the loss of plasticity and the canalization of the new phenotype through genetic assimilation (21, 49). The related ‘flexible stem’ hypothesis of adaptive radiation suggests that when a plastic lineage repeatedly colonizes similar niches, the multiple phenotypes fixed by genetic assimilation could converge among them giving rise to a collection of phylogenetically related convergent morphs (16, 50, 51). Our comprehensive study complements these hypotheses by suggesting that plasticity-mediated convergence may even evolve without the existence of basal flexible lineages. Rather, it can occur when plasticity evolving in otherwise non-plastic lineages promotes the colonization of a niche previously occupied by unrelated species. Under these circumstances, contrary to what it is predicted by the previous hypotheses, plasticity-mediated convergence is not circumscribed to phylogenetic-related species arising from a common stem lineage. This overlooked role of phenotypic plasticity may contribute to explain the ubiquity of morphological convergence in nature.

Materials and Methods

Floral traits

We recorded from the literature 31 floral traits in 3140 Brassicaceae plant species belonging to 330 genera and 51 tribes (Supplementary Data 1). All these traits have been proven to be important for the interaction with pollinators (Table S8). These traits were: (1) Plant height; (2) Flower display size; (3) Inflorescence architecture; (4) Presence of apetalous flowers; (5) Number of symmetry axes of the corolla; (6) Orientation of dominant symmetry axis of the corolla; (7) Corolla with overlapped petals; (8) Corolla with multilobed petals; (9) Corolla with visible sepals; (10) Petal length; (11) Sepal length; (12) Asymmetric petals; (13) Petal limb length; (14) Length of long stamens; (15) Length of short stamens; (16) Stamen dimorphism; (17) Tetradynamous condition; (18) Visible anthers; (19) Exserted stamens; (20) Number of stamens; (21) Concealed nectaries; (22) Petal carotenoids; (23) Petal anthocyanins; (24) Presence of bullseyes; (25) Presence of veins in the petals; (26) Coloured sepals; (27) Relative attractiveness of petals versus sepals; (28) Petal hue; (29) Petal colour as b CIELAB; (30) Sepal hue; (31) Sepal colour as b CIELAB. A detailed definition and description of these traits and their states is provided in Key Resource Table 1, whereas the original references used to determine the states of each trait per plant species is provided in Supplementary Data 1.

Family-wide floral morphospace

Using the original multidimensional trait-species matrix, we built a floral morphospace. For this, we reduced the high-dimensional matrix of floral traits to a two-dimensional space using an ordination technique (25). Because the set of floral traits included in this study were quantitative, semi-quantitative and qualitative, we used ordination techniques based on dissimilarity values. For this, we first constructed a pairwise square distance matrix of length equal to the number of Brassicaceae species included in the analysis (n = 3140). We used the Gower distance, the number of mismatched traits over the number of shared traits. This dissimilarity index is preferable to the raw Euclidean distance when there are discrete and continuous traits co-occurring in the same dataset (52).

We reduced the dimensionality of this phenotypic matrix by projecting it in a two-dimensional space. For this, to ensure an accurate description of the distribution of the species in the morphospace, we first run a principal coordinate analysis (PCoA), a technique providing a Euclidean representation of a set of objects whose relationship is measured by any dissimilarity index. We corrected for negative eigenvalues using the Cailliez procedure (25). Afterwards, we used this metric configuration as the initial configuration to run a non-metric multidimensional scaling (NMDS) algorithm (25), a method that will further optimise the sample distribution so as more variation in species composition is represented by fewer ordination axes. Unlike methods that attempt to maximise the variance or correspondence between objects in an ordination, NMDS attempts to represent, as closely as possible, the pairwise dissimilarity between objects in a low-dimensional space. NMDS is a rank-based approach, where the original distance data is substituted with ranks, preserving the ordering relationships among species (25). Objects that are ordinated closer to one another are likely to be more similar than those further apart (53). This method is more robust than distance-based methods when the original matrix includes variables of contrasting nature. However, NMDS is an iterative algorithm that can fail to find the optimal solution. We decreased the potential effect of falling in local optima by running the analysis with 5000 random starts and iterating each run 1 x 10⁶ times (54). The NMDS was run using a monotone regression minimizing the Kruskal’s stress-1 (55, 56), and compared each solution using Procrustes analysis, retaining that with the lowest residual. Because many species did not share trait states, a condition complicating ordination, we used stepacross dissimilarities, a function that replaces dissimilarities with shortest paths stepping across intermediate sites while regarding dissimilarities above a threshold as missing data (57). Furthermore, we used weak tie treatment, allowing equal observed dissimilarities to have different fitted values. The scores of the species in the final ordination configuration were obtained using weighted averaging. We checked if the reduction in dimensionality maintained the between-species relationship by checking the stress of the resulting ordination and finding goodness of fit measure for points in nonmetric multidimensional scaling (54). Both PCoA and NMDS ordinations were done using the R package vegan (58) and ecodist (59). It is important to note that, although the transfer function from observed dissimilarities to ordination distances is non-metric, the resulting NMDS configuration is Euclidean and rotation-invariant (60).

Morphological Disparity

Because we were interested in describing the position of the species in the floral morphospace, we calculated the morphological disparity using indices related to the distance between elements (26, 61). We first determined the absolute position of each of the Brassicaceae species in the morphospace by calculated their Euclidean distance with the overall centroid of the morphospace (61). The disparity between the spring and summer phenotype of M. arvensis was also calculated as their Euclidean distance in the floral morphospace. We then calculated the pairwise disparities between all species included in our analysis, between the different morphs of the polymorphic species considered here (Supplementary Data 2), between the species of the genus Moricandia, between species of the same genus, between species of different genera but same tribe and between species of different tribes. These disparity values were calculated using the function dispRity of the R package dispRity using the command centroid (62). We checked whether the disparity between spring and summer M. arvensis phenotypes was significantly different from the disparities of each of these sets of species using Z-score tests.

Family-wide phylogeny

We retrieved 80 phylogenetic trees from the literature and from the online repositories TreeBase (Table S9). All trees were downloaded in nexus format. The taxonomy of the species included in each tree was checked and updated using the species checklist with accepted names provided by Brassibase (https://brassibase.cos.uni-heidelberg.de/) (7, 23, 63). All trees were converted to TreeMan format (64) and concatenated into a single TreeMen file that was then converted into a multiPhylo class. Afterward, we estimated a supertree from this set of trees. Because trees did not share the same taxa, we used the Matrix representation parsimony method (65). To make this supertree more accurate, it was re-constructed using as backbone phylogeny the tree provided by Walden et al. (7). We removed from the supertree those species without information on floral phenotype, resulting in a tree with 1876 taxa. Because the original trees used to assemble this supertree where very heterogeneous, this supertree was not dated. We finally rooted the supertree using several species belonging to the sister families Capparaceae and Cleomaceae (66). All phylogenetic manipulations were performed using the R libraries treebase (67), ape (68), treeman (64), phangorn (69) and phytools (70).

We tested whether the position of the Brassicaceae species in the morphospace was associated with the phylogenetic relationship by assessing the phylogenetic signal of the morphospace position. This analysis was performed by means of a multivariate Mantel test, using the pairwise disparity (the Euclidean distance between species in the family-wide morphospace) as a morphological distance and the patristic distances between pairs of tips of the supertree as the phylogenetic distance (71). The correlation method used was Pearson and the statistical significance was found after bootstrapping 999 times the analysis (25). The test was done using the R libraries vegan (58) and ecodist (59).

Family-wide phylomorphospace

We reconstructed a family-wide phylomorphospace by projecting the phylogenetic relationships provided by the supertree into the floral morphospace. The ancestral character estimation of morphospace coordinate values for each internal tree node was done using maximum likelihood. For this, we used the function fastAnc in phytools. This function performs fast estimations of the ML ancestral states for continuous traits by re-rooting the tree at all internal nodes and computing the contrasts state at the root each time (70).

We counted the number of intersections between lineages as a measurement of the disorder of the phylomorphospace and evidence of the mode of evolution of the phenotypes (11). For this, we used R codes provided in Ref 11. We compared the observed number of crossings with those expected under several modes of evolution. For this, we counted the number of intersections in 10 simulated sets of species with floral phenotypes following Brownian Motion, Ornstein Uhlenbeck and Early Burst modes of evolution. All simulations were done using as backbone tree the family-wide supertree and considering 1875 species, and by means of the command mvSIM in mvMORPH (72).

Morphological divergence of the plastic phenotypes

Divergence in floral phenotype was estimated by calculating the disparity of Moricandia arvensis and the rest of Brassicaceae species from their ancestors. We first determined the floral phenotype of the Most Recent Common Ancestor (MRCA) using the projection of a recent time-calibrated phylogeny made for the genus Moricandia (73) into the above-described phylomorphospace. We used this phylogeny because it is the only one including all the species of the genus. Once we inferred the coordinates of the MRCA in the morphospace, we calculated the disparity of all the Moricandia species and the two plastic phenotypes of M. arvensis to it. Afterwards, we calculated the divergence of the two plastic phenotypes from the direct ancestor of M. arvensis. This analysis was done for the family-wide supertree and for any of the four time-calibrated phylogenies included in our dataset that had Moricandia species (73–76). In addition, we calculated the divergence from the direct ancestors of the rest of Brassicaceae species included in these four phylogenies and in the rest of the time-calibrated trees included in our dataset (Table S9). All floral divergences were calculated using the command ancestral.dist of the function dispRity in the R package dispRity (62).

Pollinator Database

We have compiled a massive database including 21,212 records comprising 455,014 visits done by over 800 animal species from 19 taxonomical orders, 276 families and 43 functional groups to 554 Brassicaceae species belonging to 39 tribes (Supplementary Data 3). Information is coming from literature, personal observation, online repositories and personal communication of several colleagues. The source of information is indicated in the database (Supplementary Data 3, Table S10). In those species studied by us (coded as UNIGEN data origin in the Supplementary Data 3), we conducted flower visitor counts in 1-16 populations per plant species. We visited the populations during the blooming peak, always at the same phenological stage and between 11:00 am and 5:00 pm. In these visits, we recorded the insects visiting the flowers for two hours without differentiating between individual plants. Insects were identified in the field, and some specimens were captured for further identification in the laboratory. We only recorded those insects contacting anthers or stigma and doing legitimate visits at least during part of their foraging at flowers. We did not record those insects only eating petals or thieving nectar without doing any legitimate visit. The information obtained from the literature and online repositories (coded as LITERATURE data origin in the Supplementary Data 3) includes records done during ecological studies, taxonomical studies and naturalistic studies. The reference of every record is included in the dataset. The plant species included in our network do not coexist, implying that this is a clade-oriented network rather than an ecological network (77).

Spatial distribution of pollinator groups

We tested the autocorrelation across the morphospace in the abundance of the functional groups using a multivariate Mantel test. The correlation method used was Pearson, and the statistical significance was found after bootstrapping 999 times the analysis (25). The test was done using the R libraries vegan (58).

Pollination niches

In plant species interacting with a diverse assemblage of pollinators, like those included in this study, many pollinator species interact with the flowers in a similar manner, have similar effectiveness and exert similar selective pressures and are thus indistinguishable for the plant (46, 78). These pollinators are thus grouped into functional groups, which are the relevant interaction units in generalised systems (46, 78, 79). We thereby grouped all pollinators visiting the Brassicaceae species using criteria of similarity in body length, proboscis length, morphological match with the flower, foraging behaviour, and feeding habits (46, 78, 79). Table S11 describes the 43 functional groups used in this study. Supplementary Data 4 shows the species with an autogamous pollination system.

We determined the occurrence of different pollination niches in our studied populations and seasons using bipartite modularity, a complex-network metric. Modularity has proven to be a good proxy of interaction niches both in ecological networks, those included coexisting species or population, as well as in clade-oriented network, those including species with information coming from disparate and contrasting sources (77). We constructed a weighted bipartite network, including pollinator data of four populations during the spring and summer flowering. In this network, we pooled the data from the different individuals in a population and did not consider the time difference involved in sampling across different species. We removed all plant species with less than 20 visits. We subsequently determined the modularity level in this weighted bipartite network by using the QuanBiMo algorithm (80). This method uses a Simulated Annealing Monte-Carlo approach to find the best division of populations into modules. A maximum of 10¹⁰ MCMC steps with a tolerance level = 10^-10 was used in 100 iterations, retaining the iterations with the highest likelihood value as the optimal modular configuration. We tested whether our network was significantly more modular than random networks by running the same algorithm in 100 random networks, with the same linkage density as the empirical one (81). Modularity significance was tested for each iteration by comparing the empirical versus the random modularity indices using a Z-score test (80). After testing the modularity of our network, we determined the number of modules (82). We subsequently identified the pollinator functional groups defining each module and the plant species ascribed to each module. Modularity analysis was performed using the R package bipartite 2.0 (83). We quantified the niche overlap between all pair of Brassicaceae species using the Czekanowski index of resource utilization, an index that measures the area of intersection of the resource utilization histograms of each species pair (84). This index was calculated using the function niche.overlap in the R package spaa (85).

Estimation of ancestral values of pollination niches

The ancestral states of the pollination niche was inferred for the Moricandia lineage by simulate stochastic character mapping of discrete traits with Bayesian posterior probability distribution (86, 87). Three models of character evolution (“ER” - Equal Rates; “SYM” – symmetric; and “ARD” - All Rates Different) were first evaluated using the fitDiscrete function of the R package Geiger (88). The best model was selected using the Akaike Information Criterion (AIC) and used for stochastic character mapping. The posterior distribution of the transition rate matrix was determined using a Markov chain Monte Carlo (MCMC) simulation, and the stochastic mapping was simulated 100 times. Stochastic character mapping was performed using the make.simmap function and a plot of posterior probabilities were mapped using the describe.simmap function in R package ‘phytools (70).

Morphological convergence

To explore morphological convergence, we reconstructed the ancestral states of the species belonging to these two pollination niches and tested for each niche whether the species were morphologically more similar to each other than expected by their phylogenetic relationship (39, 40). We used three different approaches to detect morphological convergence, one based on comparing phenotypic and phylogenetic distances (39) and the other based on comparing the angles formed by two tested clades from their most recent common ancestor with the expected angle according to null evolutionary models (38). Because all these analyses are sensitive to the number of tips in the phylogeny and the inferred branch lengths, we tested for the occurrence of morphological convergence using three independent, time-calibrated phylogenies including more than 45 species (74–76).

Under the first approach, we calculated both distance- and frequency-based measures of convergence (39). Distance-based measures (C1–C4) are calculated between two lineages relative to their distance at the point in evolutionary history where the two lineages were maximally dissimilar. C1 specifically measures the proportion of phenotypic distance closed by evolution, ranging from 0 to 1 (where 1 indicates complete convergence). To calculate C1, ancestral states are reconstructed (via a Brownian motion model of evolution) for two or more putatively convergent lineages, back to their most recent common ancestor. The maximum phenotypic distance between any pair of ancestors (Dmax) is calculated, and compared with the phenotypic distance between the current putatively convergent taxa (Dtip). The greater the difference between Dmax and Dtip, the higher the index. C2 is the raw value of the difference between the maximum and extant distance between the two lineages. C3 is C2 scaled by the total evolution (sum of squared ancestor-to-descendant changes) between the two lineages. C4 is C2 scaled by the total evolution in the whole clade. These four measures quantify incomplete convergence in multidimensional space. Finally, C5, the frequency-based measure, quantifies and reports the number of convergent events where lineages evolve into a specific region of morphospace (crossing it from outside). C5 sums the number of times through the evolution of a clade that lineages evolve into a given region of phenotypic space. C5 is the number of focal taxa that reside within a limited but convergent region of a phylomorphospace (the phylogenetic connections between taxa represented graphically in a plot of morphological space). The significance of C1–C5 was found by running 1000 simulations for each comparison using Brownian-Motion on a variance–covariance matrix based on data-derived parameters, with convergence measures for each simulation calculated to determine if the observed C value is greater than expected by chance. A priori focal groups forming the basis of convergence tests were the same niche categorizations used in OUwie analyses. These analyses were performed using the R package convevol (89).

The second approach to measure convergence was based on comparing the angles formed by two tested clades from their most recent common ancestor with the expected angle according to null evolutionary models (38). Under the “state case”, search.conv computes the mean angle over all possible combinations of species pairs using one species per state. Each individual angle is divided by the patristic distance between the species. Significance is assessed by contrasting this value with a family of 1,000 random angles obtained by shuffling the state across the species (38). These analyses were performed using the R package RRphylo (90).

The third approach to measure convergence used the Wheatleaf metric (40). This index generates phenotypic (Euclidean) distances from any number of traits across species and penalizes them by phylogenetic distance before investigating similarity (in order to weight close phenotypic similarity higher for distantly related species). It uses an a priori designation of convergent species, which are defined as species belonging to a niche for which the traits are hypothesized to converge. The method then calculates a ratio of the mean (penalized) distances between all species to the mean (penalized) distances between allegedly convergent species. The index detects if convergent species diverge more in phenotypic space from the non-convergent species and show a tighter clustering to each other (40). The significance of this index was found by comparing the empirical values of the index with a distribution of simulated indices obtained running 5000 bootstrap simulations. These analyses were performed using the R package windex (91).

Competing interests

The authors declare no competing interests.

SI FIGURES

Figure S1.

Association among the 31 pollination traits of 3140 Brassicaceae species. Trait vectors represent the Spearman correlations, with the length and direction indicating the relationship with composite NMDS axes.

Table S1.

Fitting of the floral traits onto the NMDS vectors.

Table S2.

Disparity, calculated as the Euclidean distance in the family-wide floral morphospace, between each of the 38 morphs included in our dataset (see Supplementary Data 2 for details and references) and their respective wild types.

Table S3.

Floral disparity of each species of Moricandia from the most recent common ancestor (MRCA) of the genus and from the direct ancestor of each species.

Table S4.

Significance of the Mantel tests checking for spatial autocorrelation across the morphospace of the pollinator functional groups. Due to the small abundance of some pollinators, the original 43 functional groups have been pooled in 26 main functional groups.

Table S5.

Diferences between the two Moricandia arvensis phenotypes in the visitation frequency (both in absolute number of insects and in proportion of visits) of every pollinator functional group. Fifteen censuses of 1 hr and two researchers per phenotype.

Table S6.

Outcome of the analyses to test the occurrence of floral convergence among plants from niches 3 and 5. Angle is the mean theta angle between all species belonging to the same niche. Angle/time is the angle divided by time distance. The significance of these angles has been found by comparing with a null model consisting in shuffling each niche 1,000 times across the tree tips and calculating a distribution of random angle. C1 measures the proportion of phenotypic distance closed by evolution, ranging from 0 to 1 (where 1 indicates complete convergence). C2 is the raw value of the difference between the maximum and extant distance between the lineages. C3 is C2 scaled by the total evolution (sum of squared ancestor-to-descendant changes) between the two lineages. C4 is C2 scaled by the total evolution in the whole clade. The significance of C1-C2, was evaluated by running 1000 simulations for each comparison using Brownian-Motion models. Wheatleaf is the ratio of the mean (penalized) distances between all species to the mean (penalized) distances between allegedly convergent species. Significance found by running 2000 bootstrapping simulations. In bold, significant values.

Table S7.

Outcome of the analyses testing for morphological convergence between the Moricandia clade and the rest of clades included in each time-calibrated phylogeny. Clade size is the number of species within the Moricandia clade. θ_real is the mean angle over all possible combinations of pairs of species taking one species per clade. θ_ace is the mean angle between ancestral states between each pairs of clades. dist_mrca is the patristic distance (sum of brach length) between the most recent common ancestors of each pair of clade. We indicate the congervent clades and the pollination niches of each species included in the convergent clades. In red Moricandia clades including Moricandia arvensis spring phenotype. Tribes (E= Erysimeae, A= Anchonieae, C=Cardamineae, M=Malcolmieae, An=Anastaticeae).

Table S8. Description of floral traits related to pollinator attraction used to generate the floral morphospace in Brassicaceae.

Pollinators respond to the variability of numerous phenotypic traits of plants, and the magnitude of their response shapes the reproductive success of the plants. We estimated for each plant included in our data set the values of several important floral traits.

Table S9.

List of the phylogenies retrieved from the online repositories and from the literature to built up the Brassicaceae supertree. Within brackets appears the number of species included in the analysis of disparity

Table S10.

List of ecologists kindly sharing unpublished information on Brassicaceae pollinators. The host institutions are those at the time of the contact with our team.

Table S11.

Brief description of the functional groups of the insects visiting the flowers of the studied species.

Acknowledgments

Authors thank Raquel Sánchez, Angel Caravantes, Isabel Sánchez Almazo, María José Jorquera, and Iván Rodríguez Arós for helping us during several phases of the study. We also thank all contributors to the pollinator database (Table S10) for kindly sending us unpublished information on Brassicaceae floral visitors. This research is supported by grants from the Spanish Ministry of Science, Innovation and Universities (CGL2015-63827-P, CGL2017-86626-C2-1-P, CGL2017-86626-C2-2-P, UNGR15-CE-3315), Junta de Andalucía (P18-FR-3641, IE19_238 EEZA CSIC), LIFE18 GIE/IT/000755, and Xunta de Galicia (CITACA), including EU FEDER funds. This is a contribution to the Research Unit Modeling Nature, funded by the Consejería de Economía, Conocimiento, Empresas y Universidad, and European Regional Development Fund (ERDF), reference SOMM17/6109/UGR.