Introduction

Among living organisms, angiosperms (flowering plants) display the highest variability in reproductive organs and mating systems1. Distyly is one of the best studied plant mating systems and consists of a floral dimorphism in which male and female sexual organs are reciprocally positioned. Distylous species possess two floral morphs: L-morph (pin) flowers have long style and low anthers, while S-morph (thrum) flowers have short style and high anthers2. This main morphological feature is often accompanied by a heteromorphic self-incompatibility mechanism that prevents fertilization between flowers of the same morph2,3. Additional ancillary features may further differentiate the two floral morphs4, for example number and size of pollen grains2,5,6, length and shape of stigma papillae5,7, number and shape of cells in the upper corolla tube (i.e. above the anthers’ attachment point), and width of the corolla tube mouth8.

Having evolved independently in angiosperms at least 13 times9, distyly represents an ideal case to study convergent evolution10. Research on distyly has mainly focused on Primula (Primulaceae)11,12,13,14,15,16,17, Fagopyrum (Polygonaceae)18,19, and Turnera (Passifloraceae)20,21,22,23, reviewed below, and, to a lesser extent, Linum (Linaceae)24 and Lithospermum (Boraginaceae)25,26. Phenotypic convergence in floral morphology appears to be mirrored by convergence in the genetic architecture of the locus controlling distyly. Specifically, in all studied species, distyly is controlled by a set of genes clustered together in the same genomic region, forming the so-called S-locus supergene, known to be hemizygous in S-morphs and absent from L-morphs in Primula, Fagopyrum Turnera, and Linum12,17,19,20,24.

One of the most popular ornamental plants in Europe27, Primula (primrose) has served as the canonical model to study distyly since Darwin2. Extensive genomic resources are available for this genus, including a chromosome-scale genome assembly for Primula veris, in which the S-locus is a ~ 260 kb region containing five genes (CCMT, GLOT, CYPT, PUMT, KFBT)12,17. Two S-locus genes have been functionally characterized in Primula: GLOT, homologous to the highly-conserved B-class floral homeotic gene GLOBOSA, determines high anther position in S-morphs15 and CYPT, a member of the cytochrome P450 CYP734A family that degrades brassinosteroids28, determines short styles13 and female incompatibility in S-morphs16. The functions of the other three S-locus genes (CCMT, PUMT, KFBT) remain unknown. Four S-locus genes originated via gene duplication, and their closest paralogs have been identified (CCM1, GLO1, CYP734A51, KFB1)12,13,17. A key open question on the evolution of distyly is how the S-locus genes acquired their role in controlling distyly. This might have occurred through a change in protein function, temporal and/or spatial gene expression, or a combination thereof, of the S-locus genes compared to their respective paralogs15. The lack of comparative expression profiles of the S-locus genes and their paralogs has so far precluded our understanding of how S-locus genes acquired their new functions.

With more than 1.8 million tons produced per year, buckwheat (Fagopyrum esculentum) is the most agriculturally important distylous species (www.fao.org/faostat/en/#data/QCL). Thus, knowledge on the genetic control of distyly for this species is of general interest, as it could help improve breeding strategies and artificial selection29. Despite its economic importance, little is known on the genetic underpinnings of distyly in F. esculentum, except that the S-locus is approximately 5.4 Mb long and contains 32 genes19, among which is S-ELF3, likely to control style length18.

Turnera is another genus whose distylous species have been studied for more than a century30. In recent years, three S-locus genes (hemizygous in S-morphs and absent in L-morphs) have been identified: TsSPH1, likely involved in filament elongation; TsYUC6 (a member of the YUCCA gene family), involved in auxin biosynthesis and likely controlling pollen development; and TsBAHD, inactivating brassinosteroids and controlling both female incompatibility and style length, similarly to CYPT in Primula20,21,31. Differentially-expressed genes (DEGs) between L- and S-morphs of Turnera subulata flowers have been characterized in a recent study22. Among these DEGs were several genes related to the phytochrome interacting factor (PIF) signaling network, a large and highly interconnected network that mediates several plant morphogenetics processes, whose key regulators are the PIF transcription factors32. This observation, together with the fact that phytochrome-associated pathways can modulate the morphology of sexual organs in Brassica rapa when exposed to different red:far-red ratios33, led the authors to propose that the recruitment of genes from the PIF network might represent a commonality among S-loci controlling distyly in different species22. However, this hypothesis has never been tested.

Our study is aimed at better linking the genotypic underpinnings of distyly to its phenotypic expression by investigating the transcriptomes of three distylous species: P. veris, F. esculentum and T. subulata. First, we generated the P. veris transcriptome atlas, which allowed us to identify a) the expression patterns of S-locus genes and their paralogs and b) genes that show constitutive expression across samples, tissues, and developmental stages. Second, we identified genes that are differentially expressed between L- and S-morph flowers, thus likely regulated by the S-locus, in P. veris and F. esculentum. Third, a comparative transcriptomic analysis among P. veris, F. esculentum and T. subulata identified, for the first time, homologous genes that are differentially expressed between L- and S-morphs, thus potentially involved in controlling distyly, in all three species. Last, we tested the previously proposed hypothesis that S-loci evolved by recruiting genes from the phytochrome-interacting factor (PIF) signaling network22. The results presented here address fundamental questions on the function and evolution of distyly by identifying genes involved in the expression of distyly in Primula and F. esculentum and revealing commonalities that might explain the convergent evolution of distyly in these two species and T. subulata.

Results and discussion

Primula veris transcriptome atlas

Twenty RNA-seq samples were used to analyze the transcriptomes of seven tissues (root, seed, seedling, leaf, inflorescence stem, floral bud, and flower) of P. veris. A total of 2.27 billion paired-end reads (342.77 Gb; Suppl. Table S1) were used to quantify gene expression of the 34,441 P. veris genes (Suppl. Table S2)17. Plotting the normalized RNA-seq counts for the genes in each tissue showed, as expected, a bimodal distribution, in which only the peak at higher expression levels comprises the active, functional transcriptome34 (Suppl. Fig. S1). Using this distribution and a previously-developed method35, we classified a gene as expressed in a tissue if the average number of normalized RNA-seq reads for the replicates of that tissue was ≥ 4. A total of 26,338 genes (76.47%) were expressed in at least one tissue (Fig. 1c). Transcriptional activity did not vary much among tissues, with the number of expressed genes ranging from 21,731 (63.1%) in the seedling to 23,348 (67.8%) in the flower (Fig. 1a). A total of 18,441 genes (53.5%) were expressed across all tissues (Suppl. Fig. S2).

Figure 1
figure 1

Overview of P. veris transcriptomic data. Samples of the same tissue are color coded, following the color scheme in the top-left corner of the figure. (a) Number of transcriptionally active genes (i.e. genes with normalized counts ≥ 4) per sample. (b) Heatmap of hierarchical clustering of the samples based on their pairwise similarity (estimated as Spearman’s rank correlations) for all 20 samples, with color scale indicating the degree of correlation. (c) Number of genes expressed in any given number of samples. (d) PCA plot for the 20 RNA-seq samples. Samples belonging to the same tissue cluster together. The first principal component (x-axis) separates samples of reproductive tissues (floral bud (purple) and flower (salmon)) from all remaining tissues.

Full size image

RNA-seq data have proven to be important for the identification of constitutively expressed genes, i.e. genes whose expression does not considerably vary across tissues, developmental stages, environmental conditions and experimental factors36. Constitutively expressed genes are especially important in real-time qPCR experiments, where they are used as internal controls for normalizing gene expression among samples36. Here, 53 constitutively expressed genes, defined as having a coefficient of variation (CV = standard deviation/mean of normalized counts) ≤ 0.15, were identified (Fig. 2a). The degree of tissue specificity was calculated for each gene using the tau (τ) index37 (Suppl Fig S3): the low (< 0.3) tau values found in the constitutively expressed genes further supported their constitutive expression across tissues (Fig. 2a). Housekeeping genes (HKGs) are often used as normalizing factors in qPCR experiments, but such an approach can introduce bias in the quantification of gene expression, as many HKGs have been shown not to be constitutively expressed across tissues and developmental stages38. We noticed that none of the eight HKGs commonly used as reference in qPCR experiments36 was found to be constitutively expressed in P. veris (all having CV > 0.15), demonstrating once again the importance of identifying species-specific constitutively expressed genes through RNA-seq36 (Fig. 2b). The set of P. veris constitutively expressed genes identified here represents a useful resource for normalizing gene expression in Primula qPCR analyses.

Figure 2
figure 2

Constitutively expressed genes in P. veris. Heatmap showing the variability of expression for the 53 constitutively expressed genes of P. veris identified in the current study (a) and for eight HKGs commonly used as standards for normalizing qPCR experiments (b). Each cell in the heatmap represents the deviation from the mean expression (expressed in normalized read counts) for a gene (row) in a sample (column), normalized by the mean expression for that gene, calculated across all samples. The table on the right reports the P. veris gene IDs, the A. thaliana ortholog IDs and gene name (when available), the coefficient of variation (CV) and the tau tissue-specificity index.

Full size image

The expression patterns of the 20 P. veris samples matched the biological nature of the respective tissues, as shown by their hierarchical clustering (Fig. 1b). The highest Spearman's rank correlation coefficients were found between samples of the same tissue; between tissues, floral bud and flower samples shared the highest Spearman's coefficients (ρ = 0.95), as expected since they represent two developmental stages of the same organ (Suppl. Fig. S4). Samples from the same tissue clustered together also in a PCA plot, which additionally showed a clear distinction between samples of floral (floral buds, flowers) and non-floral tissues (roots, seeds, seedlings, leaves, and inflorescence stems) on the first principal component (Fig. 1d).

Expression profiles of S-locus genes and their paralogs in P. veris

With the exception of CYPT and GLOT, known to be expressed exclusively in style and corolla tube, respectively13,15, the expression profiles of the P. veris S-locus genes were previously unknown. Given this lack of knowledge, elucidating how the S-locus genes acquired their role in controlling distyly after their origin via duplication has so far been impossible. To better characterize the expression profiles of all S-locus genes and their paralogs, we performed differential gene expression analyses between samples of floral and non-floral tissues. This allowed us to test if the duplicate S-locus genes diversified their expression profiles compared to their respective paralogs, which might have resulted in them gaining their functions in controlling distyly.

Of the five S-locus genes, GLOT and CYPT were significantly more expressed in floral than non-floral tissues, confirming previous results13,15 (adjusted p value < 0.01); PUMT and CCMT were expressed at roughly the same level in floral and non-floral tissues; KFBT appeared not to be expressed in any tissue (normalized counts < 4) (Fig. 3a). Since KFBT was reported to be expressed in P. vulgaris flower (see ref.12 and below), we believe that we did not detect KFBT expression in P. veris because it was not expressed at the time when the floral tissues were harvested, rather than not being expressed at all. This observation may also be indicative of KFBT being expressed only for a short temporal window. Furthermore, the lack of flower-specific expression for CCMT and PUMT, together with the fact that their functions are still unknown, leaves open the question of whether these genes play a role in distyly. Functional studies on CCMT and PUMT will be necessary to address this question.

Figure 3
figure 3

Expression profile of S-locus genes and their paralogs in P. veris. Box-plots showing the expression (as normalized counts) of S-locus genes (a) and their paralogs (b) in samples of floral (floral bud and flower; n = 5) and non-floral (root, seed, seedling, leaf, inflorescence stem; n = 15) tissues. For each gene, the log2fold change of expression between floral and non-floral tissues is reported together with the adjusted p value, calculated by DESeq2. NA, not applicable; NS, not significant.

Full size image

Among the four S-locus gene paralogs, only GLO1 was significantly more expressed in floral than non-floral tissues (as expected, as it is a B-class floral homeotic gene15), while no significant differences between tissue types were found for CCM1 and CYP734A51, and KFB1 was significantly more expressed in non-floral than floral tissues (Fig. 3b). Thus, the flower-confined expression of CYPT is not shared with its closest paralog (CYP734A51). This suggests that CYPT acquired its role in controlling distyly via a change in the expression profile compared to its paralog; whether this was accompanied by a change in its protein function remains to be tested. This observation marks a difference on how the two S-locus genes with a known function (GLOT and CYPT) acquired their involvement in controlling distyly, i.e. GLOT through mutations that changed the activity of its encoded protein15 but no changes in its expression profile, while CYPT through changes in its expression profile that limited its expression to the style but potentially no changes in its protein function.

Differential expression analysis between L- and S-morph flowers

To better link the genotypic underpinnings of heterostyly to its phenotypic expression, we identified differentially expressed genes (DEGs) between L- and S-morph flowers using RNA-seq data from different samples of P. veris (style and corolla tube), P. vulgaris (whole flower) (Suppl. Table S3), and F. esculentum (stamen filament and corolla tube; Suppl. Table S4). As the S-locus is the only region consistently differing between L- and S-morph genomes, we hereinafter refer to any gene up-regulated in S- compared to L-morphs as ‘up-regulated’ and any gene down-regulated in S- compared to L-morphs as ‘down-regulated’, implying that these genes are up- or down-regulated by the S-locus.

Differential expression analysis in Primula

In the P. veris style, 245 DEGs (78 up- and 167 down-regulated) were identified (Suppl. Table S5). Of the S-locus genes, GLOT, CYPT and PUMT (but not CCMT and KFBT) were found in the up-regulated gene set. Eight DEGs were involved in cell wall modifications, among which was the up-regulated pveT_jg12120, homologous to A. thaliana IBL1, a IBH1-like transcription factor known to negatively regulate cell elongation in response to brassinosteroid signaling39. A GO analysis on the down-regulated genes revealed an enrichment of terms related to two main categories: DNA replication and sugar transport (Fig. 4a,b; Suppl. Table S6). Among the genes associated to “DNA replication” (GO:0006260), “DNA replication initiation” (GO:0006270) and “DNA-dependent DNA replication” (GO:0006261) were key regulators of the cell cycle, such as: pveT_jg12216, homologous to A. thaliana Cell Division Cycle 6 (CDC6; AT1G07270), which confers cells the ability to initiate DNA replication40; pveT_jg13060, homologous to an A. thaliana transcription factor (AT3G02820), member of the zinc knuckle (CCHC-type) protein family, which is required for cell cycle progression and DNA replication upon regulation by E2F transcription factors41; pveT_jg18344, pveT_jg26277, pveT_jg29014, pveT_jg30321, and pveT_jg16023, all putative members of the Minichromosome Maintenance (MCM) protein family, which is pivotal in the initiation of DNA replication42. GO terms related to sugar transport were “carbohydrate transport” (GO:0008643), “sugar transmembrane transporter activity” (GO:0051119), and “monosaccharide transport” (GO:0015749) and comprised, among others, pveT_jg31090, homologous to A. thaliana Sugar Transporter 1 (STP1; AT1G11260).

Figure 4
figure 4

Differential expression analysis in Primula. (a) Bar-plot showing the enriched GO terms among down-regulated DEGs in the P. veris style, ordered bottom-up by increasing adjusted p values. The bar length is proportional to the number of genes belonging to each GO category (x-axis). (b) Heat-plot showing which genes (x-axis) belong to which GO category. (c), (d) Same as a and b, but for the 20 most significantly enriched GO terms among up-regulated DEGs in the P. vulgaris flower (full list in Suppl. Table S10).

Full size image

Of the 143 DEGs identified in the P. veris corolla tube, 74 were up-regulated, which included all S-locus genes except KFBT, and 69 down-regulated (Suppl. Table S7). A GO enrichment analysis on this gene set did not provide useful information, as the enriched GO terms were not related to developmental processes that could be associated to the floral dimorphism (Suppl. Table S8). Among the up-regulated genes was pveT_jg5930, homologous to a WALLS ARE THIN1 (WAT1)-related protein. WAT1 encodes for an auxin transporter which regulates secondary cell wall thickness and auxin transport in A. thaliana43.

In the P. vulgaris whole flower, 268 DEGs were identified, 198 being up- and 70 down-regulated (Suppl. Table S9). The up-regulated gene set, which contained all S-locus genes except CCMT, showed an enrichment in GO terms related to cell-wall modifications, such as “plant-type secondary cell wall biogenesis” (GO:0009834), “polysaccharide metabolic process” (GO:0044264), and “cell wall biogenesis” (GO:0042546) (Fig. 4c,d; Suppl. Table S10). The down-regulated gene set showed an enrichment in genes associated to “carbohydrates transport” (GO:0008643), driven by three genes annotated as members of the TC 2.A.1.1 sugar transporter family (pveT_jg2967, pveT_jg2968 ,pveT_jg2971) and by a UDP-galactose UDP-glucose transporter (pveT_jg9592).

Research on the mechanisms underlying style length dimorphism in Primula has shown a predominant role of cell elongation. The S-locus gene CYPT, which is expressed exclusively in the style, degrades brassinosteroids, thus limiting cell elongation and ultimately resulting in a shorter style in S-morphs than in L-morphs5,13. However, it has previously been proposed that differential cell division might also affect differential style length between morphs8, following the observation that L-morph styles are usually twice the length of S-morph styles but L-morph style cells are not twice as long as S-morph style cells5. The down-regulation of genes involved in DNA replication observed in the S-morph style suggests that this tissue is undergoing reduced cell division, thus supporting the hypothesis that cell division also contributes to style-length dimorphism. Furthermore, the decrease in sugar transport may also be responsible for the style-length dimorphism, as carbohydrates play a key role in both cell expansion and division44.

Differential expression analysis in F. esculentum

Two S-locus genes have so far been identified in F. esculentum, namely S-ELF3 and SSG218,19, while a third gene (PG1) has been shown to be expressed exclusively in S-morph styles, despite not being physically linked to the S-locus45. We identified both S-ELF3 and PG1 (but not SSG2) in the F. esculentum gene set46, as tr_15748 and tr_3984, respectively (see Methods). PG1 was expressed only in S-morph carpel samples, while S-ELF3 was expressed also in the filament, albeit at lower level (Fig. 5g).

Figure 5
figure 5

Differential expression analysis in F. esculentum. (a) Bar-plot showing the 20 most significantly enriched GO terms among down-regulated DEGs in the stamen filament, ordered bottom-up by increasing adjusted p values. The bar length is proportional to the number of genes belonging to each GO category. (b) Heat-plot showing which genes (x-axis) belong to which GO category. To ease visualization, genes associated to only one GO term were removed (full list in Suppl. Table S12). Genes belonging to the OG0000060 orthogroup (i.e. putatively coding for SAUR63 proteins) are boldfaced and highlighted by a black horizontal bar. (c), (d) Same as a and b, but for all significantly enriched GO terms among down-regulated genes in the mature carpel. (e) Phylogeny of the OG0000060 orthogroup. Bootstrap values < 75 are shown. To ease visualization, genes from A. majus, S. lycopersicum, P. veris, F. tataricum, F. esculentum, V. vinifera, A. thaliana, M. truncatula, P. persica, P. edulis, and T. subulata are indicated with the prefixes Ama, Sly, Pve, Fta, Fes, Vvi, Ath, Mtr, Ppe, Ped, Tsu, respectively (details on the nomenclature in Suppl. Table S18). (f) Cladogram of the twelve angiosperm species (modified from Janssens et al. 2020) used in our OrthoFinder analysis with the number of putative SAUR63 subfamily members identified in each species (i.e. the number of genes in the OG0000060 orthogroup). (g) Heatmap showing the normalized counts in three tissues (filament, mature carpel and young carpel) of two floral morph (S and L) for the S-locus genes S-ELF3 and PG1. For plotting, the normalization was performed for each gene independently. Maximum values correspond to 311.9 and 2513.5 normalized counts for S-ELF3 and PG1, respectively; minimum value is zero for both genes.

Full size image

In the filament, 1316 genes were differentially expressed between the two morphs (553 up- and 763 down-regulated; Suppl. Table S11). A GO enrichment analysis revealed a set of 12 down-regulated genes (tr_4884, tr_7943, tr_8778, tr_8786, tr_8787, tr_8800, tr_8807, tr_8810, tr_8811, tr_8812, tr_8814, tr_8815) associated to GO terms that indicate a negative role of these genes in stamen filament development, mediated by auxin: “negative regulation of organ growth” (GO:0046621), “auxin transport” (GO:0060918), and “stamen filament development” (GO:0080086) (Fig. 5a,b; Suppl. Table S12).

In the young carpel, 825 DEGs (299 up- and 526 down-regulated) were identified (Suppl. Table S13). The GO categories “pectinesterase activity” (GO:0030599) and “cell wall organization” (GO:0071555) were significantly enriched among down-regulated genes (Suppl. Table S14). Of the 955 DEGs identified in the mature carpel (402 up- and 553 down-regulated; Suppl. Table S15), a group of eleven down-regulated genes (tr_14602, tr_14604, tr_22182, tr_23371, tr_28613, tr_29408, tr_4884, tr_8800, tr_8805, tr_8809, tr_8812) was associated to GO terms suggesting their role in auxin-mediated organ growth, such as “regulation of organ growth” (GO:0046620) and “auxin transport” (GO:0060918) (Fig. 5c,d; Suppl. Table S16).

Noting the overlap between filament and carpel down-regulated genes, we further investigated these genes by identifying their homologous genes in eleven additional angiosperms via an OrthoFinder analysis. All the auxin- and growth-related genes that were down-regulated in the filament (except tr_8787) and in the mature carpel (except tr_22182 and tr_29408) were in the OG0000060 orthogroup, which also contained the A. thaliana SMALL AUXIN UP RNA (SAUR)63 subfamily (i.e. SAUR61-68 and SAUR75) members AT1G29420, AT1G29430, AT1G29440, AT1G29450, AT1G29460, AT1G29500, AT1G29510, AT1G29490, and AT5G27780. Members of the SAUR63 subfamily are known to play a role in hypocotyl and stamen filament elongation by activating plasma membrane H+-ATPases in response to auxin, thus promoting cell elongation47,48. We found that the two Fagopyrum species (F. esculentum and F. tataricum) displayed the highest number of SAUR63 members (49 and 30, respectively; Fig. 5f) among all the angiosperms included in our gene orthology analysis. To better characterize the relationships among SAURs, we generated a phylogeny for the OG0000060 orthogroup, which showed that the SAUR63 expansion observed in Fagopyrum is independent from the one observed in Arabidopsis (Fig. 5e).

The results above imply that auxin is likely the main hormone controlling the differential elongation of stamen and pistil in F. esculentum, potentially favored by a Fagopyrum-specific expansion of the SAUR63 subfamily, whereas brassinosteroids were shown to be the main mediator of differential style elongation in Primula13.

Comparative transcriptomics among distylous species

Distyly evolved independently at least 13 times9, representing a classic example of evolutionary convergence10. In the present study we identified DEGs between L- and S-morph pistils and stamens of P. veris and F. esculentum. The availability of DEGs between L- and S-morph flowers of Turnera subulata22 (Table 1) allowed us to perform a three-way comparative transcriptomics analysis aimed at discovering whether some genes are potentially involved in the control of distyly in all three species. Since it is difficult to identify one-to-one orthologs among distantly related species, we performed an OrthoFinder analysis including the proteomes of 12 angiosperms (see Methods for details) and searched for any orthogroup containing genes that were differentially expressed in all three species (Suppl. Table S17). This search was performed separately for up-regulated genes in female organs (Fig. 6a), down-regulated genes in female organs (Fig. 6b), up-regulated genes in male organs (Fig. 6c), and down-regulated genes in male organs (Fig. 6d). Of these, only the set of genes down-regulated in male organs did not have any shared orthogroup (Fig. 6d).

Table 1 Summary of differentially expressed genes (DEGs) in distylous species.
Full size table
Figure 6
figure 6

Orthogroups containing genes differentially expressed in P. veris, F. esculentum and T. subulata. (a)–(d) Venn diagrams showing the number of orthogroups containing differentially expressed genes between L- and S-morphs of P. veris (purple), F. esculentum (yellow) and T. subulata (green). (a) Orthogroups containing genes up-regulated in female organs. One orthogroup (OG0000318) contains genes up-regulated in female organs in the three species (pveT_jg29738, tr_14668, Tsub_00016967-RA), annotated as HSL3 receptor kinases. (b) Orthogroups containing genes down-regulated in female organs. One orthogroup (OG0000198) contains genes down-regulated in female organs in the three species (pveT_jg28601, tr_16287, Tsub_00008119-RA, and Tsub_00000118-RA), annotated as PIP aquaporins. (c) Orthogroups containing genes up-regulated in male organs. One orthogroup (OG0000315) contains genes up-regulated in male organs in the three species (pveT_jg17457, tr_6752, Tsub_00026904-RA, and Tsub_00006450-RA). (d) Orthogroups containing genes down-regulated in male organs. No orthogroup contains genes down-regulated in male organs in the three species. (e) Cladogram of the OG0000315 orthogroup genes, which contains genes annotated as FMOs, plus A. thaliana and H. vulgare FMOs. Bootstrap values < 75 are shown. The P. veris, F. esculentum and T. subulata DEGs are boldfaced. To ease visualization, genes from the different species are indicated with the prefixes as in Fig. 5f, plus H. vulgare, and D. carota, indicated as Hvu and Dca, respectively (details on the nomenclature in Suppl. Table S19. Genes that were differentially expressed between L- and S-morphs are boldfaced and marked by black circles.

Full size image

One orthogroup (OG0000315) was found to contain four genes that were up-regulated in the male organs of the three species (pveT_jg17457, tr_6752, Tsub_00026904-RA, and Tsub_00006450-RA; Fig. 6c). All OG0000315 genes were functionally annotated as flavin-containing monooxygenases (FMOs). FMOs belong to a large and highly-conserved protein family whose function is to incorporate an oxygen atom from molecular oxygen into small nucleophilic or electrophilic molecules49. This finding is of particular interest for two reasons. First, FMOs overlap in function with cytochrome P450 monooxygenases, a superfamily of enzymes that also includes the protein encoded by CYPT, the S-locus gene that controls style length and female incompatibility in P. veris13,16. Second, the T. subulata S-locus gene YUC6, which controls male mating type and pollen size23, is a FMO, specifically a member of the YUCCA gene family which catalyzes the second step in auxin synthesis from L- tryptophan20,50. To better characterize the putative function of these differentially expressed FMOs, we generated a phylogeny that included the sequences of OG0000315 genes as well as those of A. thaliana and barley (Hordeum vulgare) FMOs51 (Fig. 6e). The OG0000315 genes fall in the FMO Clade I, whose genes are supposedly involved in pathogen defense52, and to our knowledge are not involved in floral development in any species. Whether pveT_jg17457, tr_6752, Tsub_00026904-RA, and Tsub_00006450-RA play a role in controlling distyly thus remains unclear, and functional investigations would be required to clarify this point.

Another orthogroup (OG0000318) included three genes up-regulated in the female organs of the three species (pveT_jg29738, tr_14668, Tsub_00016967-RA; Fig. 6a), all annotated as members of the Serine/Threonine protein kinases family. Specifically, these genes are homologous to A. thaliana HAESA-LIKE 3 (HSL3; AT5G25930), a highly-conserved leucine-rich repeat receptor kinase (LRR-RK) which forms a complex with BRASSINOSTEROID INSENSITIVE 1-ASSOCIATED KINASE 1 (BAK1) upon induction by a class of small signaling peptides named CTNIPs53. The CTNIP-HSL3 signaling pathway seems to be involved in stress response but also affects plant growth, although it remains unknown whether it plays a role in floral development 53. This observation, together with the fact that BAK1 plays a pivotal role in regulating brassinosteroid-dependent growth54,55, suggests the involvement of brassinosteroids in causing the style length dimorphism not only in Primula and T. subulata13,21 but also in F. esculentum.

Finally, a third orthogroup (OG0000198) contained genes down-regulated in the female organs of the three species (pveT_jg28601, tr_16287, Tsub_00008119-RA, and Tsub_00000118-RA; Fig. 6b). All OG0000198 genes were annotated as aquaporins and, more specifically, the eight A. thaliana genes in this orthogroup were annotated as plasma membrane intrinsic proteins (PIPs). Aquaporins are integral membrane proteins that function as water channels but also mediate the transport of other important substrates and play a key role in plant growth and development by regulating cell turgor, thus cell expansion56,57,58. Indeed, the expression of PIP aquaporins has been shown to strongly correlate with cell expansion and tissue growth in several plant species59,60,61. Thus, the decreased expression of PIP aquaporins in S-morph pistils is compatible with the shortening of the style in this floral morph, due to reduced cell expansion.

Our results show that, even though the three species tested here are distantly related and evolved distyly independently9, some genes are differentially expressed in the same organs between L- and S-morph flowers, hence some shared biological pathways might underlie the expression of distyly in these three species. Thus, the convergent evolution of distyly at the phenotypic level is mirrored by some convergence also at the molecular level, representing one of the first studied cases of convergent evolution of complex traits62,63.

Enrichment analysis of PIF-regulated genes among DEGs

We tested the previously proposed hypothesis that the recruitment of genes intersecting with the PIF signaling network is a common motif in the evolution of distyly supergenes22 by verifying whether the DEGs identified between L- and S-morphs of Primula, F. esculentum and T. subulata were enriched in genes known to be PIF-regulated. If S-locus genes are indeed related to PIF network hubs, we expect an enrichment of PIF-regulated genes among DEGs.

Of the ten floral tissues analyzed, six showed a significant enrichment in PIF-regulated genes among DEGs compared to the genome background (Fisher’s exact test; p value < 0.05; Table 2). In T. subulata we found an enrichment of PIF-regulated genes in female organs (young and mature pistil), but not in male organs (young and mature stamen). In Primula, both style and corolla tube (but not whole flower) DEGs were enriched in PIF-regulated genes. In F. esculentum, DEGs identified in the mature carpel and in the filament (but not those identified in the young carpel) showed an enrichment in PIF-regulated genes. Of particular interest in this regard was the identification of the SAUR63 subfamily as a putative modulator of stamen filament and pistil elongation in F. esculentum (see above), as SAURs can induce organ growth by promoting cell elongation64 and can be up-regulated by auxin, brassinosteroids (the two key hormones in the floral dimorphism), and PIFs64.

Table 2 Enrichment analysis of PIF-regulated genes among DEGs in ten samples.
Full size table

These results demonstrate that genes known to be part of the PIF signaling network are enriched among DEGs between S- and L-morph flowers of three species that evolved distyly independently. Thus, S-loci might indeed evolve via the recruitment of PIF-related genes, as previously proposed22.

Conclusions

We generated a transcriptome atlas for the distylous P. veris, which allowed us to identify a set of 53 genes that are constitutively expressed across tissues, representing a useful resource for normalizing gene expression in qPCR experiments (Fig. 2). Thanks to extensive transcriptomic data from multiple floral and non-floral tissues of P. veris, we could also determine that the S-locus gene CYPT likely acquired its role in distyly via a change in expression profile, compared to its closest paralog (Fig. 3).

A differential gene expression analysis between L- and S-morph flowers confirmed that in Primula the differential style elongation between the two morphs is caused by a difference in style cell expansion (up-regulation of genes involved in cell-wall modification in the P. vulgaris flower; Fig. 4c,d), but also revealed a potential role of cell division, as implied by the down-regulation of genes associated to DNA replication and sugar transport in the S-morph style (Fig. 4a,b). In F. esculentum, a set of 17 SAURs was linked to the differential elongation of both stamen filament and pistil, indicating auxin, rather than brassinosteroids, as the main growth-inducing hormone determining the floral dimorphism in this species (Fig. 5).

In conclusion, this is the first study that identifies the main differences and commonalities in the genetic underpinnings of distyly among distantly related taxa (here, P. veris, F. esculentum and T. subulata; Table 1, Fig. 6). The main difference concerns the hormones involved in the control of style elongation, which appear to be mediated mainly by brassinosteroids in Primula and Turnera, and auxin in Fagopyrum. On the other hand, two main commonalities emerged. First, we identified three groups of homologous genes that were differentially expressed between L- and S-morphs in the three species studied here, all potentially involved in the phenotypic expression of distyly. Second, DEGs identified in the three species mentioned above are enriched in genes intersecting with the PIF signaling network, thus supporting the hypothesis that distyly supergenes evolved via the recruitment of PIF-related genes (Table 2).

This is the first time that specific genes have been identified as shared players in the expression of distyly in distantly related taxa; the increasing availability of genomic resources for distylous species will clarify whether the shared patterns observed in the three species studied here are also shared among all distylous species or not.

Methods

Data retrieval

All transcriptomic data used in this study was downloaded from NCBI GenBank. All Primula RNA-seq samples comprised paired-end Illumina reads. The RNA-seq data used for generating the P. veris transcriptome atlas consisted in 20 samples belonging to seven tissues, originally published in ref.17 and available under the BioProject PRJEB44353 (33.5–84.2 M reads per sample). These 20 P. veris samples consisted of three replicates per each tissue (except floral buds, which had only two replicates) and floral tissues contained both L- and S-morph individuals pooled together. For Primula, differential expression analyses between L- and S-morphs were performed on three tissues: P. veris style (three L- and three S-morph samples), P. veris corolla tube with attached anthers (three L- and three S-morph samples), and P. vulgaris whole flower (four L- and four S-morph samples). Primula veris samples are available under BioProject PRJNA317964 (22.2–39.0 M reads per sample) and were prepared as follows13: styles and corolla tubes with attached anthers were harvested from 25 plants per sample when petals were 4–10 mm long, i.e. when visible differences in style and anther position first arise8,13. Primula vulgaris samples are available under BioProject PRJEB9683 (15.6–25.5 M reads per sample) and were prepared from 15 to 20 mm buds, each sample representing a single individual12. The F. esculentum RNA-seq data consisted of single-end Illumina reads, were originally published in ref.46 and available under BioProject PRJNA487842, and generated from three tissues: stamen filament of mature (10-weeks old) flower, carpel of mature (10-weeks old) flower, carpel of young (8-weeks old) flower; two L- and two S-morph samples were available for each tissue. Accession numbers of each sample used in this study can be found in Supplementary Tables S1, S3 and S4.

Quantification of gene expression

Reads of each RNA-seq sample files were trimmed using Trimmomatic65 v0.38, with the following parameters: ILLUMINACLIP:2:30:10 SLIDINGWINDOW:4:5 LEADING:5 TRAILING:5 MINLEN:25. Trimmed reads were then used to run Salmon66 v1.4.0 ‘quant’ (mapping-based mode) to quantify gene expression (–gcBias –validateMappings). P. veris and P. vulgaris RNA-seq reads were mapped against the P. veris coding sequences17, while F. esculentum RNA-seq reads were mapped against the F. esculentum coding sequences from ref. 46.

To build the P. veris transcriptome atlas, first the Salmon output (i.e. read counts in the 20 P. veris RNA-seq samples for each of the 34,441 gene) was imported into R v3.6.3 (https://www.R-project.org/) using tximport67,68 and a DESeqDataSet was created with the DESeqDataSetFromTximport function of DESeq269 (R/Bioconductor70 package). In importing transcript quantifications into DESeq2, we summarized expression at the gene level. Genes showing zero counts in all samples were removed, leaving with a total of 31,112 genes whose counts were then normalized using the default median of ratios method69. An “expression per tissue” matrix was created by calculating the average among the normalized per-gene counts of the two/three samples for each tissue. The normalized counts per sample were log2 transformed and their distribution plotted (Suppl. Fig. S1); the resulting plot showed a main distribution centered at a value of ~ 10 log2 counts (~ 1024 counts) with a shoulder at the left of this distribution. Such a bimodal distribution of gene expression is often observed35 and can be used to discriminate transcriptionally active genes (main distribution) from low-expression genes (left shoulder of the distribution)35. Based on this distribution we selected a threshold of 2 log2 counts (~ 4 counts) to classify a gene as transcriptionally active (≥ 4 counts) or as not expressed (< 4 counts).

Differential gene expression analysis between floral and non-floral tissues

The 20 P. veris RNA-seq samples from BioProject PRJEB44353 were used also to compare the expression of S-locus genes and their paralogs between floral tissues (floral buds, flower) and non-floral tissues (root, seed, seedling, leaf, inflorescence stem). In brief, each RNA-seq sample was defined as “floral” or “non-floral” depending on its tissue of origin and imported into DESeq2 as described above. Then the DESeq function of DESeq2 was run with default parameters (false discovery rate controlled using the Benjamini–Hochberg method) to identify DEGs between floral and non-floral tissues.

Differential gene expression analysis between L- and S-morphs

Differential gene expression analysis between L- and S-morphs was performed for two P. veris samples (style and corolla tube), one P. vulgaris samples (whole flower) and three F. esculentum samples (stamen filament, young carpel, mature carpel) using DESeq2. For all RNA-seq samples, reads were trimmed, gene expression quantified and counts imported into DESeq2 as described above. The identification of DEGs was then carried out using the DESeq function of DESeq2 with default parameters: false discovery rate was controlled using the Benjamini–Hochberg method and a filter was then applied to exclude genes showing log2-fold between -1 and 1, and adjusted p-value (padj) > 0.05.

Identification of F. esculentum S-locus-related genes

Sequences for S-ELF3 (GenBank accession: AB642167), SSG2 (AB668598), and PG1 (Buckwheat Genome Data Base19 gene ID: Fes_sc0006922.1. g000006.aua.1) were translated and searched against the translated CDS of F. esculentum (from ref. 46) using Proteinortho71 v6.0.31 (-p = blastp -e = 1e-5 -sim = 1).

Functional annotation of P. veris and F. esculentum genes

The functional annotation of P. veris and F. esculentum genes was performed in two ways. First, 8,437 and 8,380 gene ontology (GO) terms were assigned to 23,673 and 23,516 genes of P. veris and F. esculentum, respectively, using TRAPID72 v2.0 (http://bioinformatics.psb.ugent.be/trapid_02). Second, one-to-one A. thaliana73 (TAIR10) orthologs were identified for 10,872 and 9,659 P. veris and F. esculentum genes, respectively using Proteinortho71 v6.0.31 (-p = blastp -e = 1e-5 -sim = 1), to aid functional description of the genes. To perform GO enrichment analyses on the DEGs identified in P. veris and F. esculentum, we used the ‘enricher’ function of the clusterProfiler v3.14.3 package74 (pvalueCutoff = 0.05, pAdjustMethod = BH, qvalueCutoff = 0.2).

Gene orthology and comparative transcriptomics analyses

DEGs were identified in the present study between L- and S-morphs in Primula and F. esculentum (see above). A list of DEGs between L- and S-morphs identified in four floral tissues of T. subulata (young pistil, mature pistil, young stamen, mature stamen) had already been generated in a previous study22. We performed an analysis to identify orthologous and paralogous genes using OrthoFinder75,76 v2.3.11 (-I 1.7) on the proteomes of the three distylous species studied here (P. veris, F. esculentum, T. subulata) and nine other angiosperm species selected to be closely-related to the distylous species mentioned above (Fagopyrum tataricum, Passiflora edulis) or to have high-quality and well-annotated proteomes (Antirrhinum majus, Arabidopsis thaliana, Daucus carota, Medicago truncatula, Prunus persica, Solanum lycopersicum, Vitis vinifera). In summary, 359,400 genes (out of 408,413; 88%) were assigned to 32,436 orthogroups, (mean number of genes per orthogroup: 11.1). We then investigated whether any orthogroup contained DEGs identified in all the three species, analyzing separately male and female organs and S-locus up- and down- regulated genes (four analyses in total).

Phylogenetic reconstruction was conducted for OG0000060 orthogroup and for OG0000315 orthogroup plus other FMO sequences from A. thaliana and H. vulgare51. Sequences were aligned with MAFFT77 using global pairwise alignment (–globalpair) and a maximum of 1000 iterations (–maxiterate 1000). For the analysis of the OG0000060 orthogroup, seven sequences (DCAR_003205, FtPinG0505021700.01.T01, FtPinG0505025300.01.T01, FtPinG0505414500.01.T01, Solyc10g052570.1.1, tr_4883, tr_8798) were removed after being identified as having ambiguous or insufficient phylogenetic signal (complete phylogeny in Suppl. Fig. S5). Maximum Likelihood (ML) phylogenetic trees were constructed using IQ-TREE78 v2.1.2. For each alignment, the best protein model was selected by ModelFinder79 and compared by Bayesian Information Criterion (BIC). Branch support was assessed by 1,000 ultrafast bootstrap replicates80.

PIF enrichment analysis

A total of 9,122 one-to-one A. thaliana orthologs were identified for T. subulata in the same way described for P. veris and F. esculentum. Each gene of the three distylous species with an A. thaliana ortholog was marked as “PIF-regulated” or “non-PIF-regulated” based on a list of 1,070 A. thaliana genes annotated as PIF-regulated obtained from a previous study32. For each sample of each species a contingency table containing the number of PIF-regulated and non-PIF-regulated for both DEGs and non-DEGs was built and a Fisher’s exact test was applied.