The three subfamilies of Campanulaceae sampled in this study have distinctly different floral symmetry modifications with radially symmetrical flowers in Campanuloideae, non-resupinate bilaterally symmetric flowers in Cyphioideae, and bilaterally symmetric flowers that are predominately 180° resupinate in Lobelioideae [1, 2, 7]. In these three groups, we uncovered broad gene duplications and losses that correlate with these morphological shifts. We detected all three core eudicot CYC-like genes from the CYC1, CYC2, and CYC3 clades [41]. CamCYC1 was thoroughly sampled from all three subfamilies, while CamCYC2 was likely lost in Cyphioideae and CamCYC3 was likely lost from all except the Impares (F) subclade of Lobelioideae (which along with the Genistoid (E) subclade is sister to the rest of the subfamily; Knox 2014). Additionally, we found evidence for subfamily duplications—CamCYC1 duplicated in Campanuloideae, CamCYC2 duplicated in Lobelioideae, and CamCYC3 duplicated in Cyphioideae (Fig. 7).
CamCYC1 duplicated in the radially symmetric Campanuloideae
CamCYC2—the Campanulaceae member of CYC2, which is a clade that has shown functional conservation in patterning floral bilaterally symmetry [28, 30, 41, 50, 51]—was present across the radially symmetrical Campanuloideae, from four species that span the major clades of the group (Fig. 3). There was no evidence for duplications in CamCYC2, which is consistent with other radially symmetrical groups [39, 41]. Additionally, Campanuloideae CamCYC2 copies had high sequence diversity, being on very long branches, and were therefore difficult to align with Lobelioideae species (Fig. 4). In other lineages with both radially symmetrical and bilaterally symmetrical flowers, such as Fabales, Malpighiales, and Dipsacales, species with radially symmetrical flowers have CYC2-like genes expressed uniformly across the whole corolla or have lost floral expression entirely [36, 38, 40, 52,53,54,55,56,57,58].
CamCYC3 was also found in Campanuloideae, but only in the C2 clade [3] and also on a long branch compared to Cyphioideae and Lobelioideae sequences (Fig. 5). CYC3 has been shown to be involved in axillary bud outgrowth [44] and in floral symmetry [40], but with variable function in different plant groups.
The Campanuloideae show the most diversification in CamCYC1 genes, with a duplication possibly shared across the Campanuloideae clade (Figs. 3, 7). With only minor differences, these duplicate gene trees agree with the estimated Campanuloideae species phylogeny [2, 3]. Studies in plant groups across core eudicots suggest that CYC1 genes are functionally conserved, regulating the number and position of axillary bud development [42,43,44], as well as inflorescence architecture and development [37]. Loss-of-function mutants in Arabidopsis and Populus lead to a marked increase in bud outgrowth and plant branching [42,43,44]. It is possible that the duplication of CamCYC1 set the stage for the variation in plant and inflorescence architecture in Campanuloideae. Flowers vary from solitary to complex inflorescences such as capitulate heads [1]. Broad duplications in CYC1 are less common than in the other CYC clades, although they are consistently duplicated in lineages known for capitulate heads such as in the Asteraceae [31, 45], Dipsacaceae [59], and Actinodium [35].
Cyphioideae have lost CamCYC2 and duplicated CamCYC3
Cyphioideae typically have non-resupinate bilaterally symmetrical flowers with a 3 + 2 form, with one dorsal lobe, two lateral lobes, and two ventral lobes. In all other core eudicot bilateral symmetrical lineages studied to date, CYC2 is differentially expressed across the dorsoventral axis and functions to pattern that bilateral symmetry [39]. Occasionally, CYC2 genes appear to lose floral expression or be lost from the genome of certain species, however, these are always marked by shifts to radial symmetry [53, 54, 56, 57]. Additionally, in almost all cases, CYC2 genes are duplicated in bilaterally symmetrical lineages [39, 41]. Here we report the first case of an apparent loss of CYC2 in a bilaterally symmetrical core eudicot group, Cyphioideae (Fig. 7). Sampling nine species with multiple primer sets, no CamCYC2 sequences were found, despite easily recovering them from Campanuloideae and Lobelioideae.
Along with a likely loss of CYC2 in Cyphioideae, CamCYC3 appears to be duplicated in this lineage (Figs. 5, 7). This is in stark contrast to Lobelioideae, which appear to have lost CamCYC3 in all but the Impares (F) clade, with no evidence of gene duplication. CYC3 is the most understudied paralog across core eudicots and also appears to be the most variable in function. CYC3 genes are duplicated in some groups such as Dipsacales and Asteraceae [45, 59, 60], but have been likely lost in others such as Leguminosae and Gesneriaceae [61, 62]. In Arabidopsis (Brassicaceae) and Populus (Salicaceae), CYC1 (Branched1) and CYC3 (Branched2) orthologs have redundant function in regulating bud outgrowth [44, 63] with an increase in branching in loss-of-function mutants. Interestingly, branched1 had the stronger phenotype in Arabidopsis [44] and branched2 had the stronger phenotype in Populus [63]. Although CYC3 gene function in floral symmetry has not previously been shown, studies in Dipsacales and Asteraceae reported expression patterns that are suggestive of this role, with dorsoventral expression of KmCYC3B in Knautia macedonica [40] and HaCYC3a expression specific to ray florets in Helianthus annuus [37]. This evidence suggests that CYC3 function is highly labile. Additionally, function specific to plant branching appears to be found in rosids while floral expression has been seen in campanulid asterids. CYC3 could play a role in floral symmetry in campanulids such as Cyphia, and is possibly filling the role of the lost CYC2.
Even though Cyphioideae are bilaterally symmetrical, their genetic signature compared to other core eudicot species would actually suggest they are radially symmetrical, with an apparent loss of functional CYC2-like genes. In the latest phylogenetic analyses [2] Cyphioideae are sister to radially symmetrical Campanuloideae, which retain CYC2-like genes, however, they are highly diverged. Additionally, Cyphioideae are not resupinate as are most species of Lobelioideae. However, unlike the lobe arrangement of most bilaterally symmetrical core eudicots, Cyphia flowers (Fig. 1J, K) have 3 dorsal corolla lobes and 2 ventral lobes. Standard orientation of core eudicot bilaterally symmetrical flowers have a single ventral corolla lobe, pointed downward, with two lateral and two dorsal lobes each acting as pairs that can shift along the dorsoventral axis in tandem [17]. Campanuloideae and Lobelioideae have medial ventral petal lobes [64], but the later only after resupination via torsion of the pedicel [65, 66]. The lobe arrangement in Cyphioideae, with a 3 + 2 corolla lobe arrangement, necessitates a shift in that axis at some point early in development, possibly through an independent resupination event or differentiation in the location of primordial initiation. The latter is suggested by Leins and Erbar [67] with initiation of petal lobe primordia in a 3 + 2 arrangement, however, with asymmetric early development across the dorsoventral axis. Therefore, these data support the hypothesis that the ancestral Campanulaceae was radially symmetrical and that the genetic programming of bilateral symmetry likely evolved independently in Cyphioideae and Lobelioideae.
In Lobelioideae, CamCYC1 duplicated within two subclades while CamCYC3 appears to be lost in all but the Impares clade
In Lobelioideae, CamCYC1 is broadly congruent with the hypothesized species phylogeny with no obvious subfamily-wide duplications (Figs. 3, 7) [7,8,9,10, 46, 47]. There are multiple sequences in a few species; however, these are likely alleles or more recent isolated duplications. CamCYC1 has not been implicated in bilateral symmetry in any groups, instead being involved in plant and inflorescence branching in several lineages [44, 63, 68]. CamCYC1, in keeping with the general paucity of CYC1 gene duplications found in other groups, lacks the broad duplication pattern commonly seen in CYC2 and CYC3 genes correlating with a shift to bilateral symmetry [45, 60, 62]. However, there are duplications found in the Impares (F) clade as well as the giant lobelioids (U), likely due to independent ancient genome duplications [48, 49, 69].
The Impares clade, appearing to have duplicated CamCYC1 early in its diversification (Fig. 3), is notable for having a diversity of chromosome numbers, varying among 8, 9, 10, and 11 [69], while most of Lobeliaceae have multiples of 7 chromosomes. This suggests that a genome duplication occurred early in the diversification of the Impares clade, followed by subsequent frequent chromosome losses. The duplication in CYC1 likely correlates with that genome duplication; however, we have no hypothesis for why these genes were maintained in this lineage. The Impares clade also appears to be the only group to have retained CamCYC3 genes (Figs. 5, 7). This means that this lineage maintains both an extra CYC1 and an extra CYC3 gene compared to most other Lobelioideae clades. The Impares corolla shape does differ from other groups in having large, broad ventral and lateral corolla lobes and greatly reduced, nearly scale-like dorsal lobes [70]. However, there are no data that tie this morphology with extra CYC1 and CYC3 gene copies to date.
The giant lobelias (U clade) primarily grow in tropical montane habitats around the globe and have synapomorphies of a tree-like habit, often with lignification, and are tetraploid with a chromosome number of n = 14 [7, 10, 71]. In the U clade, there are 3 subclades of CamCYC1, with the U1 clade grouping with other Neotropical, Australia, and South American Lobelioideae species sister to a clade including U2A and U2B. The current topology suggests separate duplications in Pacific Basin species (Fig. 3, green) and non-Pacific Basin species (Fig. 3, yellow); however, there was no bootstrap support for the relationships of these clades, so a single duplication could be shared across all the giant lobelias. These groups were difficult to tease apart because sequence divergence is minimal and they were amplified and cloned together, which resulted in some mixing of sequences among copies. Nevertheless, CYC1 duplicates are maintained in the giant lobelias, and better sampling could shed light on the precise ancestor(s) of this clade. For instance, in the U1 clade, Lobelia doniana is sister to the rest, supporting the East Asian origin hypothesis of giant lobelias [46], although they are nested within a grade of North American species.
Duplication of CamCYC2 in Campanulaceae is highly associated with bilateral symmetry in Lobelioideae
CamCYC2 genes are the orthologs of CYCLOIDEA, a gene which has been shown repeatedly to exhibit dorsally restricted expression in bilaterally symmetrical groups (see Hileman [39]). Additionally, the evolution of bilateral symmetry has been correlated with duplications in CYC2 genes [39, 60]. These genes are of interest in bilaterally symmetrical species of Campanulaceae, where we expect gene expression to be restricted to one side of the flower and that duplications will likely be frequent. CamCYC2 in Lobelioideae was well-sampled and, as expected, had a clear duplication across the entire clade (Figs. 4, 7). The CamCYC2 duplication very likely occurred in the Lobelioideae ancestral lineage after it diverged from Campanulaceae sensu stricto. Both Lobelioideae CamCYC2 gene clades share a similar pattern and are broadly congruent both with previous research [8, 10, 46, 47] and with the Lobelioideae CamCYC1 gene clade. As in CamCYC1, we also detected duplications in the U subclade in both CamCYC2 paralogs, likely due to tetraploidy. Flowers of Lobelioideae are resupinate, twisting their pedicel (Fig. 2A, C). However, since mature flowers, after turning, end up having a flower that looks right side up (i.e., a standard core eudicot 2 + 3 lobe arrangement); this suggests there is a developmentally earlier change in orientation to create an initial 3 up, 2 down lobe arrangement. Taxa such as species in Monopsis (G) do not twist their pedicel and end up with mature 3 + 2 flowers, reverting to the hypothesized ancestral Lobelioideae flower orientation. That said, Monopsis species did not lose their CamCYC2 copies like Cyphioideae, which similarly does not undergo resupination. There are currently no known genes involved in twisting of plant tissues, for instance, to present the flower upside down, allowing us to potentially uncover novel gene functions of CYC-like genes with further studies of these groups.
Within both of the CamCYC2A and CamCYC2B clades, the U subclade (giant lobelias) occur in two duplicate locations in the phylogeny, likely due to their tetraploid ancestry [48, 49]. This means that there are four separate clades of CYC2 in giant lobelias. One of the U clades in each of CamCYC2A and CamCYC2B have no clear sister group; however, the other clade in each is most-closely related to Lobelia urens. This relationship to Lo. urens is not well-supported in either clade, and Lo. urens is clearly part of a Mediterranean clade likely derived by amphitropical dispersal from what is now the Western Cape of South Africa [46]. Two species in the Tupa group of South America have evolved woody growth independently from the giant lobelias [5, 10, 46]. The hexaploid Tupa group [19] appears to have independently duplicated in CamCYC2B, similar to the giant lobelia (U) group.
Gene expression of CamCYC2 in Lobelioideae species is conserved following resupination and paralog dominance is correlated with dorsal petal size
In Lobelioideae species, we isolated two copies of CamCYC2 genes and utilized qRT-PCR to examine their temporal and spatial expression patterns. As previous researchers have shown, CYC2-like genes are dorsally restricted, limited to the adaxial region of flower tissues. In most examined bilaterally symmetrical species, CYC2-like paralogs are diverged in their expression, with one copy being more restricted dorsally than the other [36, 39]. Lobelioideae have resupinate flowers and we hypothesized that CamCYC2 expression would, like other bilaterally symmetrical flowers, be adaxial, corresponding to finally positioned ventral in these resupinate flowers. Using four Lobelioideae species, Lobelia erinus, Lo. siphilitica, Lo. polyphylla, and Lithotoma axillaris, we found that (1) the paralogs varied in how restricted they were on the dorsoventral axis, (2) that resupinate flowers led to the highest expression in finally positioned ventral regions, and (3) the overall patterns of expression among lobes was similar across species; however, which paralog exhibited greater expression varied (Figs. 6, 8).
The temporal expression patterns of CamCYC2 genes were relatively uniform through development, which is similar to that observed in other groups [21]. The spatial expression patterns of CamCYC2A and CamCYC2B are relatively concordant among the four species (Figs. 6A-II, B-II, C-II, D-II, 8). CamCYC2A is expressed similarly in lateral and ventral lobes, or the whole ventral region of the flower (the adaxial region) and has very low expression in the dorsal lobes (abaxial initiation). CamCYC2B is always highly expressed in the ventral lobe (adaxial initiation), has an intermediate expression level in lateral lobes, and is barely expressed in dorsal lobes (abaxial initiation). Both CamCYC2A and CamCYC2B have little to no expression in the dorsal lobes, similar to leaf expression. A similar phenomenon is seen in Malpighiaceae, with a shift in the axis of the early flower primordia resulting in a rearrangement of floral petals in the New World Malpighiaceae species, which have 1 dorsal petal, 2 lateral petals, and 2 ventral petals. The expression CYC2-like genes in Malpighiaceae, however, remains in the dorsal regions of the corolla [56,57,58].
Previous work in Dipsacales has shown that even subtle differences in the dorsoventral gradient of CYC2 expression is correlated with significantly different growth patterns of the lobes [40]. In Lobelia erinus, flowers with small dorsal lobes, CamCYC2A is significantly more expressed than CamCYC2B (Fig. 6A-II). In Lo. siphilitica, flowers with relatively larger dorsal lobes than Lo. erinus, the pattern is the same, but the distinction between the level of expression is not as great, especially in the ventral lobe where the CamCYC2B gene expression level is almost 50% of the CamCYC2A gene expression level (Fig. 6B-II). In Lithotoma axillaris and Lo. polyphylla, there is no distinct difference in shape or size between dorsal, lateral, and ventral lobes. In an opposite pattern, the CamCYC2B gene is more highly expressed than the CamCYC2A gene (Fig. 6C-II, and D-II). This is effectively an increase in expression of the gene with the broader zone of expression, which has been shown to be correlated with a more radialized flower [36]. In flower primordia, CYC genes repress cell growth and control organ number, and in later stages, A. majus CYC2 paralogs can also upregulate cell division [26, 28]. In this case, Lobelioideae species have relatively larger lateral and ventral lobes (the genetically adaxial region), likely due to the high CamCYC2 gene expression. However, it does not easily explain how Li. axillaris and Lo. polyphylla flowers have lobes with almost the same size and shape. Nonetheless, this change in the expression ratio among paralogs sets up an intriguing system to study not just CYC function and evolution, but also how morphology can be substantially altered by shifts in expression dominance among gene paralogs.