1. Introduction

The remarkable diversity of flowers in angiosperms is often attributed to the pollinator diversity (Anderson et al., 2016). Communication between flowers and pollinators heavily relies on floral signals, such as flower color and flower scent, guiding animal pollinators to flowers for pollination (Willmer, 2011). Generally, birds typically favor red flowers with minimal floral scents and nectar containing low sugar concentration, while bees are attracted to brightly colored flowers like yellow and blue, with strong floral scents and nectar rich in sugar (Fenster et al., 2004). Flower color is crucial in attracting pollinators from afar, and changes of flower color can readily lead to shifts in pollinators, potentially causing pollinator mediated reproductive isolation, or even speciation (Sheehan et al., 2016).

Flower color is evolutionarily flexible, and changes of flower color at or above the species level are common in various genera and families. Flower color changes are closely linked to speciation, with the genes responsible for color variations considered as speciation genes. For instance, in two closely related Petunia species (Solanaceae), purple-flowered P. integrifolia and white-flowered P. axillaris, loss of function mutations in the ANTHOCYANIN-2 (AN2) gene resulted in the loss of anthocyanin production in white-flowered P. axillaris, leading to pre-pollination isolation due to a shift in pollinators (Quattrocchio et al., 1999). Obviously, flower color changes could trigger pollinator shifts, leading to reproductive isolation and potentially speciation. Similar patterns have been observed in species pairs of Antirrhinum and Mimulus (Whibley et al., 2006; Yuan et al., 2016). Besides pre-mating reproductive isolation, complete reproductive isolation between species can also occur through post-pollination mechanisms like pollen competition and selection against hybrids (Song et al., 2002; Lee et al., 2008).

In contrast, intra-specific flower color changes (flower color polymorphism) are uncommon but not rare (Zhou et al., 2023). However, ecological and evolutionary consequences of intra-specific flower color changes are largely unknown. In Roscoea cautleoides, an alpine ginger plant, we found two-flower-colored (purple- and yellow-flowered) plants with sympatric distributions in field expeditions (Fig. 1). This prompted us to investigate pollinator associated reproductive isolation, as floral color serves as a key means of communication between plants and pollinators. Additionally, we analyzed the differences in pigments, gene expression, genetic divergence, and genes under selection between the two-flower-colored plants. This study could enhance our understanding of the origins of plant diversity in the mountainous regions of southwest China, one of the biodiversity hotspots.

2. Materials and methods 2.1. Plant materials

Roscoea cautleoides is a perennial herb with bulbs in soils. Like most Roscoea species, R. cautleoides has striking orchid-like flowers with long floral tubes with nectar inside as a reward, indicating a long-insect floral syndrome, but short-tongued bees are successful in pollination (Zhang et al., 2011). Flowers of R. cautleoides are purple, yellow, and occasionally pink or white. Generally, R. cautleoides grows in meadows or under forests at altitudes ranging from 2000 to 3500 m in Yunnan and Sichuan province of China, suggesting a native distribution to the Hengduan Mountains. We conducted field experiments to check the reproductive isolation between purple- and yellow-flowered plants near the Lijiang Alpine Botanical Garden (27°00′N, 100°01′E, 2830 m a.s.l., Yunnan Province, China), which is located on the south-facing slopes of the snow-capped Yulong (Jade Dragon) Mountains. At the study site, both purple- and yellow-flowered plants of the species were abundant, typically mixed completely, although the ratio of purple-to yellow-flowered plants varied across different plots. To perform metabolomic analyses of three types of pigments (anthocyanins, other flavonoids, and carotenoids), flower buds of each color morph were collected from the Lijiang (LJ) and Dali (DL) populations (with a distance of more than 150 km) in Yunnan province, China. For transcriptome sequencing, we collected flower buds from 14 purple-flowered plants and 16 yellow-flowered plants in LJ population, as well as 13 purple-flowered plants and 6 yellow-flowered plants in DL population (Additional file 2: Table S1).

2.2. Reproductive isolation between purple- and yellow-flowered plants

We conducted field experiments from May to August in 2015 and 2016. Reproductive isolation between the two-flower-colored plants was investigated in terms of micro-habitat, flowering phenology, pollinator fidelity, seed number and seed mass (details also see the Supplemental note S1 of Addition file 1). We categorized micro-habitat, flowering phenology, and pollinator fidelity as factors contributing to pre-zygotic isolation, whereas seed number and seed mass were indicators of post-zygotic isolation. The strength of reproductive isolation between the two-flower-colored plants was quantified according to the published method (Sobel and Chen, 2014).

2.3. Metabolomics

To analyze the pigments (anthocyanins, other flavonoids, and carotenoids) in purple and yellow flowers, we collected three independent biological replicates, each consisting of a single fresh flower bud (stage before flowers opening), from each color morph in the LJ and DL populations. Each flower bud was used to measure the three types of pigments and was preserved in liquid nitrogen. After freeze-dried, ground into powder, and dissolved in solvents (50% MeOH solution for anthocyanin and other flavonoid measurements; n-hexane: acetone: ethanol solution (1:1:1) for carotenoid measurement), samples were vortexed, sonicated, centrifuged, and re-extracted by repeating steps again under the same conditions. Metabolomic analyses were conducted using an UPLC-ESI-MS/MS system for anthocyanins and flavonoids, and an UPLC-APCI-MS/MS system for carotenoids (Metware Biotechnology Co., Ltd., Wuhan, Hubei, China). All metabolites were quantified using Analyst v.1.6.3 (Sciex) and Multiquant v.3.0.3 software (Sciex), with each period monitored using a specific set of multiple reaction monitoring (MRM) transitions (details also see the Supplemental note S2 of Addition file 1).

2.4. Transcriptome sequencing

We randomly selected 14 purple-flowered plants and 16 yellow-flowered plants in LJ population, as well as 13 purple-flowered plants and 6 yellow-flowered plants in DL population (Additional file 2: Table S1). We collected one flower bud from each plant in both LJ and DL populations. We set Roscoea schneideriana as outgroup, and collected flower buds from two plants. We kept these flower buds individually in liquid nitrogen and extracted RNA of each flower bud with Eastep® super total RNA extraction kit (Promega, CHINA). cDNA libraries were built according to manufacturer's recommendations, and then paired-end reads were generated on DNBSEQ-T7 platform (BGI, Wuhan Frasergen Bioinformatics Co., Ltd, China). At least 6 GB data were produced for each sample.

2.5. Differential expression analysis

To analyze gene expression, we selected individuals from the LJ and DL populations that had similar microenvironments. The LJ population sample consisted of 7 purple-flowered plants and 7 yellow-flowered plants. Similarly, the DL population sample included 7 purple-flowered plants and 5 yellow-flowered plants (Additional file 2: Table S1). Differential expression analysis was performed for the LJ and DL population against the genome sequence of Roscoea cautleoides (unpublished, completed by Dr. Jian-li Zhao from Yunnan University) with HISAT2 v.2.2.1, Stringtie v.2.2.1 (Pertea et al., 2016), and edgeR v.3.2.4 (Robinson et al., 2010) using default parameters. For each transcript, the fold change value (log₂FC), counts per million (log₂CPM), P value and false discovery rate (FDR) were measured. Genes were considered differentially expressed between the two flower color morphs in each population if they met the following criteria: |log₂FC| > 1 and FDR < 0.05. As flower color is the most distinctive trait between purple flowers and yellow flowers, we specifically examined differentially expressed genes associated with anthocyanin, other flavonoid, and carotenoid biosynthetic pathways using the KEGG automatic annotation server (KAAS) (Moriya et al., 2007). qRT-PCR were conducted to validate the findings from the differential expression analysis (details also see the Supplemental note S3 of Addition file 1) as previously described (Zhou et al., 2019).

2.6. SNP calling and population structure analysis

For each sequenced individual (Additional file 2: Table S1), clean transcriptome reads were aligned to the genome sequence of Roscoea cautleoides with BWA v.0.7.17 (Li and Durbin, 2009). Sorted mapped reads were generated and marked duplicates with Samtools v.1.11 (Li et al., 2009; Li, 2011) and Picard Tool (Broad institute, Cambridge, USA, http://broadinstitute.github.io/picard/). SNPs detection and filtration were performed with GATK v.4.1.2.0 (McKenna et al., 2010). Variant sites were further removed using vcftools (Danecek et al., 2011), including SNP with missing rates > 0.05, SNPs with possibility out of Hardy–Weinberg equilibrium (HWE) > 0.001, and minor allele frequencies (MAF) < 0.05. The genetic structure was analyzed using the filtered SNP sites, including principal component analysis (PCA), structure analysis, and maximum likelihood (ML) phylogenetic tree construction. PCA was conducted with Genome-wide Complex Trait Analysis (GCTA) v.1.91.6 (Yang et al., 2011). The genetic ancestry of each individual was estimated through ADMIXTURE v.1.3.0 (Alexander and Lange, 2011), with the postulated ancestral populations of one to six. Additionally, maximum likelihood (ML) phylogenetic trees were constructed with IQtree v.2.1.2 (Nguyen et al., 2015), with R. schneideriana as the outgroup. Phylogenetic relationships among the purple- and yellow-flowered plants were further explored using TWISST (Martin and Van Belleghem, 2017) with R. schneideriana as the outgroup. TWISST quantified the frequency of alternative phylogenetic topologies across the genome using a sliding window approach. The non-overlapping windows with a fixed size of 10 kb and a fixed count of 1000 SNPs were used, respectively, to infer local genealogies (details see the Supplemental note S4 of Addition file 1).

2.7. Genomic divergence and selection

To assess the genetic differentiation between two-flower-colored plants of Roscoea cautleoides, we calculated F_ST and D_XY values with pixy software. These calculations were performed over sliding windows of 10 kb across the genomes of both the LJ and DL populations. We identified highly divergent genomic regions using the permutation-empirical combined method (Mu et al., 2023). This approach considered genomic windows within the top 5% F_ST values and with FDR vales lower than 0.01. Based on a threshold of 100 kb (Mu et al., 2023), genomic regions were classified into large and small "genomic islands". The overlaps of outlier F_ST and D_XY windows (top 5% F_ST values and top 5% D_XY values) were considered divergent regions of reduced gene flow (Mu et al., 2023). The intersection of F_ST and π diff (top 10% of F_ST values and top 10% of π diff values) (Wang et al., 2022) were used to detect genes under selection between two-flower-colored plants in two populations. To further investigate the functions of the highly divergent genes and genes under selection, we performed gene ontology (GO) analysis and searched genes against the non-redundant (Nr) protein database.

3. Results 3.1. Fitness advantages of yellow flowers

We examined various floral traits in purple and yellow flowers of Roscoea cautleoides (Additional file 2: Table S2), and found that the corolla tube length of purple flowers was significantly longer than that of yellow ones (Fig. 2A; Additional file 2: Table S2). Pollen number per flower was similar in purple and yellow flowers, but purple flowers produced more ovules than yellow flowers (Additional file 2: Table S2), resulting in a significant decrease in pollen/ovule ratio of purple flowers (Fig. 2B). The flowering time of the two-flower-colored plants was similar (Supplemental note S1 of Additional file 1: Fig. S4).

Both moths (Macroglossum nycteris, Fig. 1B) and bumblebees (Bombus secures, Fig. 1C) were identified as potential pollinators for R. cautleoides. Bumblebees spent more time on individual flowers compared to moths in both color morphs, and deposited significantly more pollen grains on the virgin stigma than moths (Supplemental note S1 of Additional file 1: Fig. S1). The visitation rate of pollinators was significantly influenced by the interaction between pollinator and flower color (Fig. 2C and D; Additional file 2: Table S3), indicating a shift in pollinator preference due to flower color. Moths exhibited a preference for purple flowers (Fig. 3A and C) while bumblebees tended to favor yellow flowers (Fig. 3B and D).

The average seed production under natural conditions was generally higher for yellow flowers than for purple ones (Additional file 2: Table S3), with a clear correlation to the ratios of purple to yellow flower-colored plants (Supplemental note S1 of Additional file 1: Fig. S2). Both seed production and fruit set of yellow flowers showed a significant increase with a higher frequency of yellow flowers in the five plots, but not for purple flowers. Additionally, the seed size of yellow flowers was typically larger than that of purple flowers (Additional file 2: Table S3). Consequently, yellow flowers demonstrated a higher fitness over purple flowers.

3.2. Reproductive isolation between two-flower-colored plants

We examined both pre-zygotic and post-zygotic reproductive isolation between two-flower-colored plants (details also see Supplemental note S1 of Additional file 1). Pollinator-mediated isolation was 0.349 for purple flower-colored plants and 0.665 for yellow flower-colored plants (Table 1), accounting for 73.1% and 89.3% of the total reproductive isolation, respectively (Table 1). In contrast, reproductive isolation mediated by micro-habitat (Supplemental note S1 of Additional file 1: Fig. S3) and flowering time (Supplemental note S1 of Additional file 1: Fig. S4) was much lower than pollinator-mediated isolation (Table 1). Inter-color pollination reduced seed size and seed production for both flower-colored plants compared to intra-color pollination (Fig. 3E and F; Additional file 2: Table S3), and post-zygotic reproductive isolation was 0.067 (8.4%) and 0.154 (5.8%) for purple- and yellow-flowered plants, respectively. Overall, the total reproductive isolation in the sympatric population was 0.439 for purple-flowered plants and 0.718 for yellow-flowered plants, with a major contribution from pollinator specialization (Table 1).

3.3. Loss of anthocyanins in yellow flowers

We conducted metabolomics on purple and yellow flowers from two populations in Lijiang (LJ) and Dali (DL) to analyze pigments, including anthocyanins, other flavonoids, and carotenoids (details also see Supplemental note S2 of Additional file 1). PCA analyses of all metabolites (Supplemental note S2 of Additional file 1: Fig. S5A) and each type of pigments (Supplemental note S2 of Additional file 1: Fig. S5B-D) revealed four distinct clusters, showing significant differences in metabolites between the two-colored flowers and the two populations. Further analysis revealed that metabolites clusters were found in abundance in plants of specific colors from one of the populations. For example, purple flowers from the LJ population (20%, Supplemental note S2 of Additional file 1: Fig. S6A) and DL population (17%, Supplemental note S2 of Additional file 1: Fig. S6B) showed higher metabolite levels in Cluster 1 and 2, respectively, and yellow flowers from the LJ population (15%, Supplemental note S2 of Additional file 1: Fig. S6G) and DL population (14%, Supplemental note S2 of Additional file 1: Fig. S6H) exhibited higher metabolite levels in Cluster 7 and 8, respectively. In contrast, the shared increased metabolites in yellow-flowered plants from the two populations had lower proportions. Specifically, the metabolites grouped into Cluster 9 only accounted for 9% of the total metabolites (Supplemental note S2 of Additional file 1: Fig. S6I). These results suggested that the metabolites in yellow flowers may have originated independently from LJ and DL populations.

To investigate the pigment changes underlying flower color variation, we analyzed the types and amount of metabolites within and between populations. Our findings revealed that, among the three pigment types, anthocyanins exhibited the highest abundance in purple flowers (Fig. 4A, Supplemental note S2 of Additional file 1: Fig. S7). Furthermore, anthocyanin metabolites were observed to be significantly more abundant in purple flowers than yellow flowers for both populations (Fig. 4A). Among a total of 93 anthocyanin, 43 anthocyanin metabolites showed lower level in yellow flowers compared with purple flowers from the LJ population (T-test, P < 0.05, Additional file 2: Table S4). Similarly, 33 anthocyanin metabolites exhibited lower abundance in yellow flowers relative to purple flowers from the DL population (T-test, P < 0.05, Additional file 2: Table S5). However, only 19 of these differentially abundant anthocyanin metabolites were shared between the two populations (Fig. 4B). Although anthocyanin levels reduced in both populations, the anthocyanin metabolite changes differed between populations, showing different mechanisms underlying anthocyanin loss.

3.4. Distinct gene expression underlying the flower color change

The quality assessment of our transcriptome data, with an average Q30 value of 93.81% and an average coverage of 99.21% across all sequenced samples (Additional file 2: Table S6 and S7), confirmed that the data were highly qualified for further analysis. By using transcriptome sequencing, we compared gene expression between purple and yellow flowers from LJ and DL population (Additional file 2: Table S1), with the genome sequence of R. cautleoides as reference (unpublished, completed by Dr. Jian-Li Zhao from Yunnan University, details also see Supplemental note S3 of Additional file 1). Among a total of 40 and 113 differentially expressed genes detected in LJ and DL population, none of them shared between the two populations (Additional file 2: Table S8). Instead, 24 genes (60%) and 44 genes (38.94%) were highly expressed in purple flowers compared to yellow flowers in LJ population and DL population, respectively (Fig. 4C).

We further focused on differentially expressed genes associated with pigment biosynthesis. In the LJ population, UGT77B2 gene (myricetin 3-O-rhamnosyltransferase) and SQD2 gene (sulfoquinovosyltransferase 2), which were involved in the flavonol glycosylation (Irmisch et al., 2018; Zhan et al., 2019), showed significantly higher expression level (log₂FC > 1, FDR < 0.05) in yellow flowers than purple flowers (Fig. 4D). Nevertheless, F3′5′H gene (flavonoid 3′, 5′-hydroxylase 1), which determined the delphinidin-based anthocyanin biosynthesis (Tanaka and Brugliera, 2013), was more highly expressed in purple flowers than in yellow flowers (log₂FC > 1, FDR < 0.05) in DL population (Fig. 4D). The differentially expressed genes identified in the two populations were subsequently validated through our quantitative real-time PCR (qRT-PCR) analysis (Supplemental Note S3 in Additional file 1: Fig. S10). Our expressional analyses therefore showed the independent trajectories of gene regulations associated with flower color change in two populations. Thus, the parallel loss of anthocyanins in the LJ and DL population, driven by changes in anthocyanin metabolites and gene regulations associated with flower color variation, could contribute to maintaining the flower color polymorphism in R. cautleoides.

3.5. Signatures of genomic divergence between two-flower-colored plants

We genotyped purple- and yellow-flowered plants from LJ and DL population using transcriptome sequencing data (Additional file 2: Table S1), resulting in 393, 493 genome-mapped polymorphic loci (details also see Supplemental note S4 of Additional file 1). By performing genomic admixture analyses (Fig. 5A, Supplemental note S4 of Additional file 1: Fig. S11) and PCA analyses (Additional file 1: Fig. S13), we found that all individuals could be divided into LJ and DL lineages, which corresponded to the geographical locations of populations but not the flower colors. Additionally, we found several individuals exhibiting clear introgression between purple- and yellow-flowered plants within both the LJ and DL lineages. These individuals were excluded to get the 'non introgression' group, leaving two yellow-flowered plants and two purple-flowered plants in each lineage. The ML phylogenetic trees constructed from both 'non introgression' individuals (Fig. 5B) and all individuals (Supplemental note S4 of Additional file 1: Fig. S12) also showed that individuals could be clearly grouped into the LJ and DL lineages. We further tested past introgressions based on fixed window size and fixed number of SNPs. Both approaches revealed that the most common topology (topo3) clearly divided all samples into the distinct LJ and DL lineages (Supplemental note S4 of Additional file 1: Fig. S14). This finding further reinforced that the underlying genetic structure was not primarily differentiated by flower colors, but rather by population-level divergence.

To investigate the genetic differentiation between two flower colors, we estimated the average genome-wide genetic divergences (F_ST) and identified "genomic islands" in LJ and DL populations. In the LJ population, the average F_ST was 0.03, and 498 "genomic islands" exhibited genetic divergence between purple- and yellow-flowered plants, which could be merged into 469 non-overlapping windows. Based on the delimitation of large "genomic islands" as above 100 kb (Mu et al., 2023), no large "genomic islands" was detected and all small "genomic islands" were scattered across the chromosomes in the LJ population, with the largest "genomic islands" as 30 kb (Fig. 5C; Additional file 2: Table S9). However, the average F_ST in the DL population was 0.07, and 700 non-overlapping windows from 730 "genomic islands" showed genetic differentiation between purple- and yellow-flowered plants. No large "genomic islands" was distributed in DL population and all small "genomic islands" were scattered across the positions, with the largest "genomic islands" as 50 kb (Fig. 5C; Additional file 2: Table S9).

We then explored highly divergent genomic regions that may be resistant to gene flow based on the intersection regions of top 5% F_ST and top 5% D_XY (details also see Supplemental note S5 of Additional file 1). In the LJ population, 51 genes were obtained (F_ST > 0.08 and D_XY > 0.32, Fig. 5D). GO annotation of these genes detected several biological processes associated with reproduction, responses to biotic stress, etc. (Additional file 2: Table S10). For example, among the reproduction associated genes, POK (poky pollen tube) affected polar growth beyond the pollen tube elongation process (Lobstein et al., 2004) in LJ population. On the other hand, 100 genes were identified in DL population (F_ST > 0.19 and D_XY > 0.35, Fig. 5E). These genes were found to be functionally associated with reproduction, responses to light and drought, responses to biotic stress, etc. (Additional file 2: Table S10). For instance, IPUT1 (inositol phosphorylceramide glucuronosyltransferase 1), associated with reproduction, controlled pollen tube behaviors beyond pollen tube guidance process (Tartaglio et al., 2017) in DL population. Although these highly divergent genes in the LJ and DL population were similarly associated with reproduction and response to biotic stress, their location (Fig. 5F) and the detailed function remained largely divergent (Additional file 2: Table S10). These highly differentiated genes, associated with reproduction and stress responses, might further contribute to the reproductive isolation between the two-flower-colored plants.

We then used F_ST and π diff (F_ST top 10%, π diff top 10%) to detect genes under selection via flower color change. A total of 327 genes were detected in the LJ population, which were primarily enriched in biological processes such as the nucleobase-containing compound catabolic process (Supplemental note S5 of Additional file 1: Fig. S15A). 361 genes were identified in the DL population, which were found to be functionally associated with biological process of response to radiation, response to light stimulus, etc. (Supplemental note S5 of Additional file 1: Fig. S15B). The fact these genes were under selection in parallel pathways may reflect the local adaptation to their respective habitat.

4. Discussion 4.1. Reproductive isolation between the two-color morphs

Changes of flower color can cause a shift in pollinators (Ramsey et al., 2003). The ginger family (Zingiberaceae) is predominantly found in the tropics and subtropics (Wu and Larsen, 2000), and the diversification of this family is thought to have evolved alongside long-tongued flies as pollinators (Morita, 2008). The alpine gingers (Roscoea) are restricted to the Himalaya-Hengduan Mountains and typically have blue to purple flower colors (Wu and Larsen, 2000; Zhao et al., 2017). In R. cautleoides, our study found both purple- and yellow-flowered plants coexisted at the same location. Our study found that both bumblebees and moths acted as pollinators for R. cautleoides, but the purple flowers were preferred by long-tongue moths while the yellow-flowered plants attracted bumblebees as their primary pollinator (Fig. 2C and D; Fig. 3A–D). The pollination efficiency of moths was significantly lower than that of flies, due to the high risk of pollen limitation and reduced seed production in the purple flowers (Paudel et al., 2015). In contrast, bumblebees are considered the dominant and efficient pollinators in alpine ecosystems (Bingham and Ort, 1998; Liu et al., 2014). As a result, the yellow-flowered plants had a selective advantage over the purple ones, benefiting from higher fitness in terms of seed production and seed size (Supplemental note S1 of Additional file 1: Fig. S2).

Pollinators tend to visit flowers with similar morphology to avoid the additional energy costs associated with learning new flower structures (Gegear and Thomson, 2004). Consequently, the fine discrimination and visiting constancy of pollinators could reduce pollen flow between plants displaying different floral traits, potentially leading to reproductive isolation (Ramsey et al., 2003; Husband and Sabara, 2004; Anderson et al., 2016). In our study, we observed adaptations for bumblebee pollination in both the flower color and corolla tubes of yellow flowers in R. cautleoides. Specifically, the corolla tube of bumblebee-preferred yellow flowers was shorter than those of moth-preferred purple flowers in R. cautleoides (Fig. 2A; Additional file 2: Table S2), which was in accordance with the general observation that moth-pollinated flowers typically have longer corolla tubes than those pollinated by bumblebees (Chapurlat et al., 2015). Although pre-zygotic reproductive isolation factors such as micro-habitat and flowering phenology contributed minimally to the total reproductive isolation between the two-color morphs in R. cautleoides, the consistent movements of pollinator within each color morph provided substantial reproductive isolation between them (Table 1).

Post-zygotic reproductive isolation could also play a significant role in preventing gene exchange among sympatric plant species, but it was generally less effective than pre-zygotic isolation in maintaining species boundaries since post-zygotic isolation would lead to considerable gamete wastage (Dell'Olivo et al., 2011; Carrio and Gueemes, 2014; Wang et al., 2021). In our study, we observed modest post-zygotic isolation in terms of seed production and seed size in R. cautleoides (Fig. 3E and F). Overall, the total reproductive isolation was 0.718 for yellow-flowered plants and 0.439 for purple-flowered plants, and both of which were induced by pollinator specialization. It's noteworthy that the reproductive isolation was more pronounced in the yellow morph than in the purple one, suggesting potential selective advantages for yellow-flowered plants in R. cautleoides. Moreover, the total reproductive isolation observed between the two-color morph in the sympatric population (Table 1) was lower than other reported inter-specific ones (Ramsey et al., 2003; Dell'Olivo et al., 2011; Carrio and Gueemes, 2014; Paudel et al., 2018), indicating that the two-color morphs might be on the way of forming new species.

4.2. Parallel changes of metabolites and gene expression in yellow flowers

The yellow color flowers in R. cautleoides could be derived from purple color, as purple is the predominant flower color in the Roscoea genus. Although yellow flowers occurred in both LJ and DL populations, our cluster analysis based on genotyped data identified two distinct subgroups of yellow-flowered plants, corresponding to LJ and DL population (Fig. 5A and B). Metabolomics could provide clear clues to interpret the diverse evolution between species and within species (Deng et al., 2020; Yang et al., 2022). Among secondary metabolites, anthocyanins played a crucial role in flower color change (Tanaka et al., 2008). In our study, we found that nearly half of the differentially abundant anthocyanin metabolites between purple and yellow flowers were different between the two populations (Fig. 4B), suggesting independent loss of anthocyanins in yellow flowers.

The convergent loss of floral anthocyanins at species level could be influenced by similar patterns in the expression of pathway genes (Larter et al., 2018; Kellenberger et al., 2019). However, our results showed that expression levels of UGT77B2 and SQD2 were elevated in yellow flowers in LJ population (Fig. 4D), potentially enhancing the glycosylation of flavonols (Irmisch et al., 2018; Zhan et al., 2019). The increased glycosylation of flavonols could lead to a competition for sugar donors between flavonols and anthocyanins, consequently reducing the transformation of unstable anthocyanidins into stable anthocyanins (Zhang et al., 2014), and thus contributing to the loss of anthocyanins. Conversely, in the DL population, down-regulations of F3′5′H in yellow flowers may inhibit the biosynthesis of delphinidin-based anthocyanins (Tanaka and Brugliera, 2013), which also results in anthocyanin loss. Obviously, the loss of anthocyanins is occurring in parallel across different populations. Genes underlying changes of flower color are considered to be "speciation genes" (Rieseberg and Blackman, 2010), as flower color change can lead to shifts in pollinator preferences and enhance reproductive isolation (Yuan et al., 2016). Therefore, flower color change and pollinator shift in R. cautleoides might drive the divergence between the purple- and yellow-flowered plants.

4.3. Genetic differentiation between the two-flower-colored plants

The Hengduan Mountains in southwest China restrict intra-specific gene flows due to the vicariant isolation caused by the north-south trending parallel mountains and valleys (Liu et al., 2013), but intra-specific genetic differentiation in sympatric populations is rarely reported. In the early and late stages of speciation, "genomic islands" could reduce the gene flow (Wang et al., 2022). Large genomic islands have been identified as barriers to gene flow in species that are not geographically isolated (Sun et al., 2022; Mu et al., 2023). In R. cautleoides, "genomic islands" have been found between two-flower-colored plants within both LJ and DL populations. However, low levels of genetic differentiation between these two-flower-colored plants may be attributed to the limited number and size of these "genomic islands" (Fig. 5C–E), suggesting an early stage of speciation (Huang et al., 2020; Bock et al., 2023). Additionally, the differences of genes with high divergence (Additional file 2: Table S10) and under selection between two-flower-colored plants (Supplemental note S5 of Additional file 1: Fig. S15) further supported the parallel genetic differentiation in LJ and DL population.

Our results suggest that changes of flower color led to shifts in pollinator preferences, which primarily contribute to pre-zygotic reproductive isolation between two-flower-colored plants of R. cautleoides. Flower color changes independently in different populations due to the parallel loss of anthocyanins, but the low genetic divergence observed between the two-flower-colored plants suggests an early stage of speciation. Our comprehensive studies on the parallel loss of anthocyanins and its role in driving reproductive isolation of R. cautleoides allow us to witness incipient sympatric speciation of plant species native to Hengduan Mountains.

Acknowledgements

We are grateful to Donanld M. Waller for his suggestions and efforts in manuscript preparation, Mr. Zhuoheng Jiang in Shanghai Normal University and Mr. Ze-Qing Niu in Institute of Zoology (Chinese Academy of Sciences) for their help in pollinator identification, Mr. Kun Xu in Lijiang Alpine Botanical Garden for his logistical supports in the fields in the past years. The work was financially supported by National Natural Science Foundation of China (32102429, 32371586), the Second Tibetan Plateau Scientific Expedition and Research (STEP) program (2019QZKK0502), the Yunnan Young Talent Project (YNWR-QNBJ-2019-214), and the CAS-TWAS Presidents Fellowship to Tial Cung Ling.

CRediT authorship contribution statement

Zhi-Li Zhou: Writing – review & editing, Writing – original draft, Visualization, Validation, Formal analysis, Data curation. Tial C. Ling: Writing – original draft, Visualization, Validation, Investigation, Data curation. Jian-Li Zhao: Formal analysis, Data curation. Xin-Zhi Wang: Formal analysis, Data curation. Lin-Lin Wang: Investigation, Data curation. Li Li: Data curation. Wen-Jing Wang: Data curation. Dong-Rui Jia: Investigation, Data curation. Zhi-Kun Wu: Investigation, Data curation. Xu-Dong Sun: Writing – review & editing, Writing – original draft, Formal analysis, Data curation, Conceptualization. Yong-Ping Yang: Writing – review & editing, Writing – original draft, Investigation, Data curation, Conceptualization. Yuan-Wen Duan: Writing – review & editing, Writing – original draft, Investigation, Formal analysis, Data curation, Conceptualization.

Data availability

The clean sequence data reported in this paper have been deposited in the Genome Sequence Archive (Chen et al., 2021) at National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession numbers CRA001642 that are publicly accessible at https://ngdc.cncb.ac.cn/gsa.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

The author Yuan-Wen Duan is an Editor for Plant Diversity and was not involved in the editorial review or the decision to publish this article.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2025.03.004.