Introduction

Conical cells (CCs), trichomes, and hairs are all protrusive epidermal cells of plants. Because they differ in various ways (e.g., shape, size, number, and patterns of distribution and ornamentation) but can be found on a single petal or petaloid organ, they have fascinated scientists for centuries. Functionally, CCs, which characterize petals of about 80% of angiosperms, can increase color intensity, brightness, and surface hydrophobicity, and provide grip for pollinators1,2,3. Trichomes or hairs on petals or petaloid organs, however, can increase the complexity of these organs and endow them with specialized functions (Supplementary Fig. 1)4,5. On the adaxial surface of the labella of Ophrys (Orchidaceae), for instance, different types of unicellular trichomes mimic hairs of hymenopteran females and facilitate deceptive pollination6. Within the perianth tube of most Aristolochia (Aristolochiaceae) species, several types of multicellular trichomes coordinately play roles in trapping, retention, and release of insect pollinators7. On petals of many Malvaceae species, unicellular trichomes with different lengths can regulate flower bud shape8,9. Exploring the development and evolution of these protrusive cells, therefore, is not only the key to understanding their differences and relationships, but also provides insights into how such specialized surface architectures contribute to key ecological functions.

The molecular mechanisms underlying the formation of petal CCs, trichomes, and hairs have been studied in some species. Key regulators shaping CCs include genes encoding enzymes involved in microtubule organization (e.g., KATANIN, PP2A, SPIKE1, and ROP GTPases), cell wall synthesis (e.g., RHAMNOSE BIOSYNTHESIS 1 [RHM1]), and vacuolino formation (e.g., RAB5 GTPases)10,11,12,13. Notably, MIXTA-like genes of the MYB family have apparently attracted considerable attention due to their diverse roles in regulating the development of various protrusive cells on petals or petaloid sepals. For instance, the MIXTA homologs specify the identity or control the shape of CCs in Antirrhinum majus (Plantaginaceae)14, Petunia hybrida (Solanaceae)15, and Thalictrum thalictroides (Ranunculaceae)16,17; promote cuticle development on the CC surface in Arabidopsis thaliana (Brassicaceae)18; or perform both functions in Phalaenopsis aphrodite (Orchidaceae)19. Additionally, the MIXTA-like genes were found to be involved in petal trichome formation, e.g., in Gossypium hirsutum (Malvaceae)8,9. It is also notable that CCs have been classified as trichomes because they are reminiscent of emerging or arrested trichomes20,21,22. Overexpressing MIXTA homologs of A. majus and other species in Nicotiana tabacum (Solanaceae) converted flat cells into conical shape or trichomes15,16,17,23. Consequently, it has been hypothesized that CCs and trichomes may share a common developmental pathway, where the relative timing of MIXTA expression specifies fates of the two cell types23,24.

Notably, apart from MIXTA-like genes, other genes have been identified as critical determinants of trichome or hair formation on leaves, seeds, or fruits in species including A. thaliana, G. hirsutum, G. arboreum, Solanum lycopersicum (Solanaceae), and Cucumis sativus (Cucurbitaceae)25,26. For instance, in A. thaliana, the leaf trichome fate is determined by a MYB-bHLH-WDR (MBW) protein complex-GLABRA2 (GL2, a class IV HD-Zip gene) module, in which the MYB protein is represented by GLABROUS1 (GL1), bHLH by GLABRA3 (GL3) and ENHANCER OF GLABRA3 (EGL3), and WDR by TRANSPARENT TESTA GLABRA1 (TTG1)27,28,29,30,31. Whether genes of these lineages or families contribute to petal trichome specification remains to be investigated. In addition, it has been shown that petal CCs vary in overall size, steepness and height of the cone, base shape, and surface decoration2, and that petal trichomes can be unicellular or multicellular, straight or curly, long or short, glandular or non-glandular (Supplementary Fig. 1). However, due to the lack of comprehensive studies from an evolutionary developmental perspective, how different types of protrusive cells have evolved on petals in a certain lineage remains poorly understood. Clearly, these limitations have hindered our understanding of the differences and relationships of protrusive cells on petals and petaloid organs.

The genus Nigella (Ranunculaceae) appears to be an excellent system for addressing the above issues, for four reasons. First, petals of all Nigella species, excluding the three basal species, bear three types of unicellular protrusive cells, i.e., CCs, short trichomes (STs), and long hairs (LHs), each type displaying a unique spatial distribution32. For example, on petals of N. damascena, CCs are distributed on the adaxial epidermis of lower lip lobes, STs are densely packed on the contact surfaces between upper and lower lips, and LHs are sparsely distributed on pseudonectaries and the adaxial side of lower lip lobes (Fig. 1a). The coexistence of CCs, STs, and LHs on petals provide an opportunity for a direct comparison of these protrusive cell types in one system. Second, it has been shown that noticeably differentiated LHs and STs evolved after N. integrifolia split from all other extant species of Nigella, and that CCs originated later in the genus32. Therefore, a comparison between species with CCs, STs, and LHs (e.g., N. damascena) and the basalmost species with only one type of hair (i.e., N. integrifolia) allows us to explore the molecular mechanisms underlying the evolution of these protrusive cells in Nigella. Third, previous studies speculated that STs may serve at least three functions: filtering pollen grains adhering to the insect proboscis, preventing nectar from drying out, and protecting nectar from being stolen by unfavorable visitors32,33, implying that STs represent an adaptive character. And fourth, several species in this genus have been developed into model systems, to which virus-induced gene silencing (VIGS), other molecular techniques, and phenotype-controlled pollination experiments are applicable34,35,36,37. These advantages enable detailed gene expression and functional studies, even the precise assessment of ecological functions for a specific protrusive cell type on petals.

Fig. 1: Cellular features of conical cells, short trichomes, and long hairs on Nigella damascena petals.
Fig. 1: Cellular features of conical cells, short trichomes, and long hairs on Nigella damascena petals.
Full size image

a The distribution of different cell types on a mature petal under stereomicroscope and scanning electron microscope. Numbered regions (1–4) are magnified to show the micromorphology of long hairs (LHs), short trichomes (STs), conical cells (CCs), and pavement cells (PCs), respectively. b–e Semi-thin sections of regions containing LHs, STs, CCs, and PCs. f–i Propidium iodide (PI) staining of the four cell types. The white arrowhead points to the nucleus. Scale bars: (a left) 100 μm; (a right and b–i) 50 μm. j Total numbers of STs and LHs on each petal. k Lengths of STs and LHs. l The largest PI fluorescence areas of nuclei in the four cell types. m Volumes of reconstructed three-dimensional nuclei in the four cell types. In (j–m), violin outline width shows the density of the data; the thick bar represents the interquartile range (IQR) between the first and third quartiles, with the median marked by a white dot and labeled with the corresponding value; whiskers extend up to 1.5 times the IQR. **P < 0.01, ***P < 0.001 (two-sided Mann-Whitney U test). Source data for (j–m) are provided in Supplementary Data 1. See also Supplementary Fig. 2.

In a previous study, a class I HD-Zip gene, i.e., LATE MERISTEM IDENTITY1 (LMI1) from N. damascena, was found to be indispensable for the development of STs on petals; knockdown of it led to the disappearance of STs, but did not affect the development of LHs and CCs36, implying that different regulatory mechanisms may underlie the formation of STs and the other two protrusive cells. Here, we use petals of N. damascena and N. integrifolia as research systems and conduct extensive micromorphological, anatomical, cytological, comparative transcriptomic, expression, functional, pollination, and phylogenetic studies. We attempt to explore the cellular features and underlying developmental bases of CCs, STs, and LHs; to understand the ecological functions and evolutionary mechanisms of STs and LHs in Nigella; and eventually to disentangle the differences and relationships among CCs, STs, and LHs.

Results

Cellular features of CCs, STs, and LHs on N. damascena petals

To gain more insights into the cellular features of CCs, STs, and LHs on mature petals of N. damascena, we conducted micromorphological, anatomical, and cytological observations (Fig. 1, Supplementary Fig. 2 and Supplementary Data 1). We found that CCs are covered with grainy cuticular striations, whereas STs possess smooth surfaces and LHs show slightly textured cuticular surfaces (Fig. 1a and Supplementary Fig. 2a). We also found that, on each petal, CCs are obviously the highest in number and the shortest in length among the three cell types (Fig. 1a–d). The median number of STs per petal (432) is almost eight times that of LHs (55), whereas the median length of STs (101.55 μm) is only one-tenth that of LHs (1180.48 μm) (Fig. 1j, k). Propidium iodide (PI) and 4′,6-diamidino-2-phenylindole (DAPI) staining indicated that nuclei of CCs showed sizes comparable to those of flat pavement cells (PCs) on petals and guard cells (GCs) on leaves, which are generally believed to have a 2C DNA content38 (Fig. 1f–i, Supplementary Fig. 2b–i and Supplementary Data 1). In contrast, the nuclear sizes of STs (median volume: 2572.37 μm3) are approximately twice those of CCs (median volume: 1373.71 μm3) and PCs (median volume: 1213.08 μm3), while the nuclear sizes of LHs (median volume: 4622.63 μm3) are about twice those of STs (Fig. 1l, m and Supplementary Data 1). Flow cytometry analysis further revealed that while the majority of cells were 2C in samples containing STs and LHs, clear peaks at 4C and 8C levels were detectable (Supplementary Fig. 2j–l), suggesting that endoreduplication occurs in STs and LHs. Taken together, these findings suggest that the CCs, STs, and LHs on N. damascena petals significantly differ in shape, length, number, distribution pattern, surface ornamentation, relative nuclear size, and possibly ploidy level.

Identification of candidate genes for CC, ST, and LH formation

To understand how CCs, STs, and LHs are formed, we first investigated their morphological changes during the development of N. damascena petals. We found that LHs start to emerge at stage 6 (S6), then gradually elongate, and stop growing at S12, whereas STs arise at S10 and reach final lengths at S12, and CCs become distinguishable at S11 (Fig. 2a, Supplementary Fig. 3a and Supplementary Data 1). We then examined digital gene expression (DGE) profiles across eight petal stages36 and in four parts of petals at S10 (Fig. 2c) to identify potential genes involved in the formation of CCs, STs, and LHs. We particularly focused on genes homologous to those involved in CC and trichome/hair development in model plants (Supplementary Data 2). By correlating gene expression dynamics with developmental processes and distribution patterns of CCs, STs, and LHs, we identified five genes that may specify these cells (Fig. 2b, d). NidaGL1-1 (a R2R3 MYB gene) was expressed at very early stages (S4 and S5), likely only regulating the initiation of LHs. NidaGL3 (a bHLH gene) was highly expressed at S6 and specifically expressed in the ST-containing Part II of S10 petals, whereas NidaGL2 (a class IV HD-Zip gene) was expressed from S6 to S12 and in regions containing both STs and LHs of S10 petals, indicative of their roles in regulating the initiation or development of both STs and LHs. NidaMIXTA (a R2R3 MYB gene) showed an expression peak at S10 and was preferentially expressed in the CC-containing Part IV, matching the initiation of STs and the differentiation of CCs, respectively. NidaTT8 (a bHLH gene) started to be highly expressed from S10, hinting at a specific role in promoting the initiation of STs. In addition, we found seven genes that likely regulate the progression and termination of endoreduplication in STs and LHs, including five potential activators, i.e., NidaCONSTITUTIVE EXPRESSION OF PR GENES 5-1 (NidaCPR5-1), NidaCPR5-2, NidaROOT HAIRLESS 2 (NidaRHL2), NidaHYPOCOTYL6 (NidaHYP6), and NidaCELL CYCLE SWITCH PROTEIN 52 A1 (NidaCCS52A1), which showed relatively higher expression during elongation of LHs and STs; and two potential inhibitors, i.e., NidaGT-2-LIKE1-1 (NidaGTL1-1) and NidaGTL1-2, which started to be highly expressed at late stages (Fig. 2b, d and Supplementary Fig. 3b, c).

Fig. 2: Development processes of conical cells, short trichomes, and long hairs as well as expression profiles of key candidate genes in Nigella damascena petals.
Fig. 2: Development processes of conical cells, short trichomes, and long hairs as well as expression profiles of key candidate genes in Nigella damascena petals.
Full size image

a The development of long hairs (LHs), short trichomes (STs), and conical cells (CCs). Images below petals show magnified views of the boxed regions containing STs (sky blue) and CCs (yellow). Red, sky blue, and yellow arrowheads point to the emerging LH, ST, and CC, respectively. For each stage, at least 5 petals were observed. Scale bars: (Rows 1 and 2) 100 μm; (Row 3) 50 μm. b Expression profiles of key candidate genes across eight developmental stages of petals. Red, sky blue, and yellow bars indicate the stages when LHs, STs, and CCs arise, respectively. c Sampling strategy for the digital gene profiling of the S10 petal. I, II, III, and IV represent four sampled parts. d Expression profiles of key candidate genes in the four parts of the S10 petals. Note that in (b) and (d), except NidaCPR5-2, which is a nuclear replication activator candidate, all other genes are candidates for CC or trichome specification. For expression profiles of other candidates, see Supplementary Fig. 3. Source data for (b) and (d) are provided in Supplementary Data 2. e A hypothetical model showing genes involved in the formation of LHs, STs, and CCs. The interaction and regulatory relationships among proteins of candidate genes are illustrated in reference to Arabidopsis thaliana. Note that the broadly expressed TTG1 is shown in the model as a potential partner of GL1/MYB5-1/MYB5-2 and GL3/TT8.

To identify additional candidate genes, we checked the previously identified stage-specific coexpression modules containing NidaGL1-1, NidaGL3, NidaGL2, NidaMIXTA, NidaTT8, and NidaLMI136. By examining the top 10 transcription factors in each module (Supplementary Data 2), we noticed that NidaGL2 and NidaLMI1 were coexpressed with a R2R3 MYB family member, NidaMYB5-1. Additionally, NidaMYB5-1 was specifically expressed in the ST-containing Part II of S10 petals and spatially coexpressed with NidaGL3 (Fig. 2b, d), strongly implying the involvement of NidaMYB5-1 in the formation of both STs and LHs. Interestingly, the paralogous gene of NidaMYB5-1, NidaMYB5-2, was expressed later than NidaMYB5-1 and showed an expression peak at S10, indicating its role in regulating the development of STs. By taking all the findings into account, we hypothesized that CCs are specified by NidaMIXTA, STs by NidaLMI1, NidaMIXTA, NidaMYB5-1, NidaMYB5-2, NidaGL3, NidaTT8, and NidaGL2, and LHs by NidaGL1-1, NidaMYB5-1, NidaGL3, and NidaGL2 (Fig. 2e).

Expression patterns and functions of NidaMIXTA, NidaGL1-1, NidaGL2, and NidaGL3

To test our hypothesis, and more importantly, to understand the relationships among CCs, STs, and LHs, we conducted detailed expression and functional studies for the aforementioned genes in N. damascena using mRNA in situ hybridization and VIGS techniques. We started with the well-known MIXTA-, GL1-, GL2-, and GL3-like genes. According to our hypothesis, NidaMIXTA likely regulates the differentiation of both CCs and STs, and NidaGL1-1 may be involved in the formation of LHs. However, we found that NidaMIXTA was mainly expressed on the adaxial surface of lower lip lobes at S10 and S11, but not in the region producing STs during petal development (Fig. 3a and Supplementary Fig. 4a). Compared with mock (Fig. 3d), knockdown of NidaMIXTA did not affect the shapes of CCs and STs, but rather led to a relatively smooth surface of CCs due to the defect in cuticle development (Fig. 3e, Supplementary Fig. 4b–f and Supplementary Data 3). For NidaGL1-1 and its paralog NidaGL1-2, no in situ hybridization signal was observed in protrusive cells on petals at different stages (Supplementary Fig. 5); and no observable phenotypic change was detected on petals of TRV2-NidaGL1-1-, TRV2-NidaGL1-2-, and TRV2-NidaGL1-1-NidaGL1-2-treated plants (Supplementary Fig. 6 and Supplementary Data 3). These results strongly suggest that NidaMIXTA is only involved in the differentiation of CCs by promoting cuticular modifications, whereas NidaGL1-1 is not involved in the formation of LHs.

Fig. 3: Expression patterns and functions of NidaMIXTA, NidaGL2, and NidaGL3.
Fig. 3: Expression patterns and functions of NidaMIXTA, NidaGL2, and NidaGL3.
Full size image

a–c Results of mRNA in situ hybridization for NidaMIXTA (a), NidaGL2 (b), and NidaGL3 (c) in Nigella damascena petals at different stages. For each gene, the first image shows the result using the sense probe. All results are representative of at least two independently repeated experiments. d–g The morphology and micromorphology of the mock petal (d, npetal = 7) and petals with strong phenotypic changes in TRV2-NidaMIXTA-NidaANS- (e, npetal = 3), TRV2-NidaGL2- (f, npetal = 6), and TRV2-NidaGL3- (g, npetal = 10) treated plants. In each treatment, both the adaxial side of the lower lip and the abaxial side of the upper lip (at the bottom left) are shown. Numbered regions (1–3) are magnified to show phenotypes of long hairs, short trichomes, and conical cells, respectively. Scale bars: (a–c) 100 μm; (d–g) 50 μm. See also Supplementary Figs. 4, 7 and Supplementary Data 3.

In contrast, in situ hybridization and VIGS results of NidaGL2 and NidaGL3 well supported our prediction of their roles in the development of STs and LHs. NidaGL2 was first expressed in emerging and developing LHs on S6 and S7 petals, and subsequently restricted to nascent and growing STs on petals at S9 and later stages (Fig. 3b and Supplementary Fig. 7a). When NidaGL2 was downregulated, both STs and LHs became significantly fewer and shorter. On petals with strong phenotypic changes, LHs were generally less than 200 μm, and most STs ranged from 3 to 30 μm (Fig. 3f, Supplementary Fig. 7c, d, g and Supplementary Data 3). On petals showing moderate phenotypic changes, the lengths of most STs and LHs were reduced to about half of those on mock petals (Supplementary Fig. 7e and Supplementary Data 3). NidaGL3 showed broader expression domains than NidaGL2. At early stages of petal development, NidaGL3 was strongly expressed in epidermal cells on the adaxial side of both upper and lower lips at S6 and then confined to emerging LHs at S7. Later, NidaGL3 was specifically expressed in the areas where STs would initiate and on the adaxial surface of the upper lip at S9 and S10 (Fig. 3c and Supplementary Fig. 7b). When NidaGL3 was knocked down, petals with strong phenotypic changes became balder than NidaGL2-knocked-down petals: LHs completely disappeared, whereas STs were noticeably reduced in both number and length (generally less than 20 μm) (Fig. 3g, Supplementary Fig. 7c, d, h and Supplementary Data 3). On petals showing moderate phenotypic changes, LHs were almost absent, and STs were fewer and shorter (mostly 20–60 μm) than those on mock petals (Supplementary Fig. 7f and Supplementary Data 3). Taken together, these results indicate that NidaGL2 and NidaGL3 indeed promote the formation of STs and LHs.

Expression patterns and functions of NidaTT8 and NidaMYB5s

We next focused on the remaining candidates, i.e., NidaTT8 and NidaMYB5s. NidaTT8 was a candidate for the specification of STs. In accordance with our hypothesis, NidaTT8 was expressed in STs of petals at S10 and later stages (Supplementary Fig. 8a). When NidaTT8 was knocked down, STs were missing or became extremely short (generally less than 40 μm) on petals with strong phenotypic changes (Fig. 4a, Supplementary Fig. 9a, b, h and Supplementary Data 3); relatively few and short STs (mostly 30–70 μm) were observed on petals showing moderate phenotypic changes (Supplementary Fig. 9c and Supplementary Data 3). Besides, simultaneous knockdown of NidaTT8 and its paralog NidaGL3 led to the complete absence of both STs and LHs (Fig. 4b, Supplementary Fig. 9b, d, i and Supplementary Data 3). Interestingly, we found that NidaTT8 was also expressed in the adaxial side of the upper lip and the abaxial side of the lower lip (Supplementary Fig. 8a). Knockdown of NidaTT8 alone or in combination with NidaGL3 affected pigmentation, resulting in green petals (Supplementary Fig. 9e–g). These results indicate that NidaTT8 not only promotes the initiation of STs redundantly with NidaGL3 but also regulates anthocyanin biosynthesis on petals.

Fig. 4: Functions of NidaTT8, NidaMYB5-1, NidaMYB5-2, and NidaLMI1.
Fig. 4: Functions of NidaTT8, NidaMYB5-1, NidaMYB5-2, and NidaLMI1.
Full size image

a–f The morphology and micromorphology of Nigella damascena petals with strong phenotypic changes in TRV2-NidaTT8- (a, npetal = 6), TRV2-NidaTT8-NidaGL3- (b, npetal = 4), TRV2-NidaMYB5-1- (c, npetal = 3), TRV2-NidaMYB5-2- (d, npetal = 7), TRV2-NidaMYB5-1-NidaMYB5-2- (e, npetal = 3), and TRV2-NidaLMI1- (f, npetal = 4) treated plants. In each treatment, both the adaxial side of the lower lip and the abaxial side of the upper lip (at the bottom left) are shown. Numbered regions (1–3) are magnified to show phenotypes of long hairs, short trichomes, and conical cells, respectively. Scale bars: 50 μm. See also Supplementary Figs. 9, 10 and Supplementary Data 3.

Our hypothesis stated that NidaMYB5-1 was a candidate for both STs and LHs, whereas NidaMYB5-2 for STs only. As expected, NidaMYB5-1 was first expressed in emerging LHs on S7 petals, and then in ST precursors on S9 petals (Supplementary Fig. 8b). However, we found that NidaMYB5-2 was specifically expressed in developing LHs from S7 to S10, but not in STs (Supplementary Fig. 8c), against our prediction. When NidaMYB5-1 was knocked down, STs completely disappeared and LHs were reduced in length (ranging from 100 to 500 μm) on petals with strong phenotypic changes (Fig. 4c, Supplementary Fig. 10a, b, e and Supplementary Data 3), whereas a few STs arose sporadically on petals with moderate phenotypic changes (Supplementary Fig. 10c and Supplementary Data 3). Knocking down NidaMYB5-2 alone resulted in normal petals (Fig. 4d, Supplementary Fig. 10a, b and Supplementary Data 3). However, in TRV2-NidaMYB5-1-NidaMYB5-2-treated plants, petals with strong phenotypic changes became nearly glabrous, on which LHs almost disappeared and STs were completely missing (Fig. 4e, Supplementary Fig. 10b, f and Supplementary Data 3). On petals with moderate phenotypic changes, LHs were significantly fewer and shorter than those on the mock petals, and STs were absent (Supplementary Fig. 10d and Supplementary Data 3). These findings indicate that NidaMYB5-1 not only regulates the initiation of LHs redundantly with NidaMYB5-2, but also is responsible for the formation of STs.

To understand how the above functionally characterized genes cooperatively regulate the formation of STs and LHs, we examined the possible regulatory relationships among them. Through qRT-PCR assays, we found that when genes encoding the potential components of the MBW complex, e.g., NidaMYB5-1, NidaGL3, and NidaTT8, were downregulated, the expression level of NidaGL2 significantly decreased, indicating that NidaGL2 functions downstream of the hypothetical MBW complex to specify STs and LHs (Supplementary Fig. 11a–c, e). We also found that when the ST-specific gene NidaLMI1 was knocked down (Fig. 4f), NidaMYB5-1, NidaGL3, and NidaGL2 were all downregulated (Supplementary Fig. 11d). This result, plus the fact that the expression of NidaLMI1 in STs36 was earlier than the expression of other ST genes, i.e., NidaMYB5-2, NidaGL3, NidaTT8, and NidaGL2, indicate that NidaLMI1 may function as an upstream activator of the hypothetical MBW complex-GL2 module for STs (Supplementary Fig. 11e).

Ecological functions of STs and LHs in N. damascena

It is noteworthy that knocking down most of the candidate genes produced glabrous petals or those with only LHs, thus providing us an excellent opportunity to explore the potential ecological functions of STs and LHs. By using bumblebees (Bombus terrestris), one of the legitimate pollinators35,39, we conducted pollination studies on wild-type flowers (with STs and LHs), TRV2-NidaMYB5-1-treated flowers (with only LHs, but without STs), and TRV2-NidaTT8-NidaGL3-treated flowers (lacking both STs and LHs) (Fig. 5 and Supplementary Data 4). We found that 90% of naive bumblebees and 100% of experienced ones successfully opened the lips of all the petals in wild-type flowers and reached nectar (Fig. 5a–e, j, k and Supplementary Movie 1). In contrast, when naive and experienced bumblebees visited TRV2-NidaMYB5-1- and TRV2-NidaTT8-NidaGL3-treated flowers, they failed to open petal lips in most cases, with no significant differences observed between the two types of VIGS-treated flowers (Fig. 5f–k and Supplementary Movies 2, 3). This occurred because the gap between the upper and lower lips, naturally present in wild-type petals, was closed in the VIGS-treated petals due to the absence of STs (Supplementary Fig. 9e–g). In addition, when comparing the foraging time of the bumblebees, we did not find significant differences among different types of flowers regardless of whether naive or experienced pollinators were used (Fig. 5l, m). These results indicate that the presence of STs may serve as tiny pillars to prop open upper and lower lips, hence facilitating the access of pollinators to nectar, whereas the role of LHs in bumblebee pollination is negligible.

Fig. 5: Ecological functions of short trichomes and long hairs in Nigella damascena.
Fig. 5: Ecological functions of short trichomes and long hairs in Nigella damascena.
Full size image

a–i The visiting behaviors of bumblebees on wild-type (a–e, petals with short trichomes and long hairs), TRV2-NidaMYB5-1- (f, g, petals with only long hairs) and TRV2-NidaTT8-NidaGL3- (h, i, petals lacking both short trichomes and long hairs) treated flowers. Movie frames showing the moments before (b), during (c), and after (d) the bumblebee foraged for nectar in a petal. The yellow arrow indicates the entrance to the petal nectary. The purple arrow indicates the proboscis of the bumblebee. Illustrations showing the accessibility of bumblebees to nectar in different petals (e, g, i). ms: millisecond. See Supplementary Movies 1–3 for details. j, k Percentages of petals opened by naive (j) and experienced (k) bumblebees in wild-type and VIGS-treated flowers. l, m Foraging times of naive (l) and experienced (m) bumblebees on wild-type and VIGS-treated flowers. In (j–m), the box is bounded by the first and third quartiles with a horizontal line at the median, and whiskers extend to the minimum and maximum values. **P < 0.01, ***P < 0.001, n.s., not significant (Kruskal-Wallis test). Each experimental group comprised 21 independent replicates, each involving one bumblebee and one flower. Source data for (j–m) are provided in Supplementary Data 4.

Evolution of STs and LHs and their underlying mechanisms

To understand how STs and LHs have evolved in Nigella, we carefully observed the cellular features and developmental processes of hairs on petals of N. integrifolia, which are unicellular, with a wide distribution and relatively uniform lengths (Fig. 6a–c and Supplementary Fig. 12a, b). For the convenience of comparison with N. damascena, we divided the hairs into two types, i.e., Type 1 hairs (T1Hs) and Type 2 hairs (T2Hs), with T2Hs specifically referring to hairs positionally corresponding to STs on petals of N. damascena, and T1Hs representing hairs in other regions. When comparing T1Hs on the adaxial side of the lower lip (positionally corresponding to LHs on N. damascena petals) with T2Hs, we found that T2Hs (median number: 211; median length: 261.74 μm) are denser but slightly shorter than T1Hs (median number: 129; median length: 289.60 μm) (Fig. 6d, e and Supplementary Data 5). Nuclear size and ploidy level comparison revealed that the nuclei of T1Hs and T2Hs were approximately twice as big as the PC nuclei (Fig. 6f–i and Supplementary Fig. 12c, d) and had a higher ploidy level of 4C than the dominant 2C cells (Supplementary Fig. 12e–g), suggestive of the occurrence of one round of endoreduplication in the hair cells. By tracing the development of T1Hs and T2Hs, we found that T1Hs initiate continuously from S6 to S9, and reach their final lengths by S12, while T2Hs initiate at S10 and continue elongating until S12 (Fig. 6j, Supplementary Fig. 12h and Supplementary Data 5). A comparison of these results with those in N. damascena indicates that the evolution of obviously different STs and LHs was likely caused by the elongation of T1Hs on the adaxial side of the lower lip, but the shortening of T2Hs on the contact surfaces between upper and lower lips.

Fig. 6: Cellular features and developmental processes of hairs and expression patterns of key candidate genes in Nigella integrifolia petals.
Fig. 6: Cellular features and developmental processes of hairs and expression patterns of key candidate genes in Nigella integrifolia petals.
Full size image

a The distribution of Type 1 hairs (T1Hs), Type 2 hairs (T2Hs), and pavement cells (PCs) on a mature petal under stereomicroscope and scanning electron microscope. Numbered regions (1–2) are magnified to show T1Hs/PCs and T2Hs, respectively. b, c Semi-thin sections of regions containing T1Hs/PCs (b) and T2Hs (c). d Total numbers of T1Hs and T2Hs on each petal. e Lengths of T1Hs and T2Hs. f, g Propidium iodide (PI) staining of T1Hs/PCs (f) and T2Hs (g). The white arrowhead points to the nucleus. h The largest PI fluorescence areas in T1Hs, T2Hs, and PCs. i Volumes of reconstructed three-dimensional nuclei in T1Hs, T2Hs, and PCs. In (d, e, h) and (i), violin outline width shows the density of the data; the thick bar represents the interquartile range (IQR) between the first and third quartiles, with the median marked by a white dot and labeled with the corresponding value; whiskers extend up to 1.5 times the IQR. **P < 0.01, ***P < 0.001, n.s., not significant (two-sided Mann-Whitney U test). Source data are provided in Supplementary Data 5. j The development of T1Hs and T2Hs. The images below petals show the magnified views of the boxed regions containing T2Hs (dark blue). Pink and dark blue arrowheads point to the emerging T1H and T2H, respectively. For each stage, at least 3 petals were observed. k Expression profiles of key candidate genes across eight developmental stages of petals. Pink and dark blue bars indicate the stages when T1Hs and T2Hs arise, respectively. Source data are provided in Supplementary Data 6. l–o Results of mRNA in situ hybridization for NiinMYB5-1 (l), NiinGL3 (m), NiinGL2 (n), and NiinLMI1 (o) in N. integrifolia petals at different stages. All results are representative of at least two independently repeated experiments. See also Supplementary Fig. 14. Scale bars: (a left) 1 mm; (a right, b, c, f, g) 50 μm; (j, l–o) 100 μm. p A summary of gene expression patterns throughout petal development of N. integrifolia. The expression levels are determined based on the FPKM values and the intensities of in situ hybridization signals (the darker, the higher).

To gain some insights into the molecular mechanisms underlying the evolution of STs and LHs in Nigella, we retrieved putative orthologs of genes that play critical roles in LHs and/or STs of N. damascena petals (i.e., MYB5-1, MYB5-2, GL3, TT8, GL2, and LMI1) from the reference transcriptome of N. integrifolia40. Phylogenetic analyses revealed that while MYB5-1 and MYB5-2 were duplicated before the divergence of the Ranunculaceae, all examined genes were single-copied in Nigella species (Supplementary Fig. 13). DGE profiling of N. integrifolia petals across eight developmental stages comparable to N. damascena revealed that while all examined genes were expressed at varying levels throughout petal development, they showed expression peaks at S8/9 (Fig. 6k and Supplementary Data 6). mRNA in situ hybridization results further showed that NiinMYB5-1 and NiinGL3 were expressed in regions containing emerging hair cells; NiinGL2 was specifically expressed in hair cells; NiinLMI1 was only expressed inside the nectary chamber and the surrounding region where T2Hs would arise; NiinMYB5-2 and NiinTT8, however, were not expressed in petal hairs (Fig. 6l–p and Supplementary Fig. 14). These results suggest that in N. integrifolia, NiinMYB5-1, NiinGL3, and NiinGL2 are all required for the formation of petal hairs, and NiinLMI1 is only involved in the initiation of T2Hs. Interestingly, when comparing the programs for petal trichomes/hairs between N. integrifolia and N. damascena, we found that MYB5-1, GL3, GL2, and LMI1 showed similar expression patterns and possibly conserved functions, but MYB5-2 was specific for LHs, and TT8 was specific for STs (Fig. 7a, b). These findings suggest that independent recruitment of TT8 and MYB5-2 into the trichome specification program, followed by refining of their expression patterns, may be the key to the evolution of STs and LHs in Nigella (Fig. 7c, d).

Fig. 7: Possible evolutionary scenarios for the evolution of short trichomes and long hairs on Nigella petals.
Fig. 7: Possible evolutionary scenarios for the evolution of short trichomes and long hairs on Nigella petals.
Full size image

a, b Schematic diagrams showing expression patterns of MYB5-2 (a) and TT8 (b) in S7 and S10 petals of Nigella damascena and N. integrifolia. The expression signals are marked by brown shadings. Scale bars: 200 μm. c Evolutionary history of petal hairs/trichomes in Nigella. The simplified phylogenetic tree and divergence times were adopted from a previous study32. Schematic diagrams (I)–(V) show petals of five representative Nigella species. From nodes -1 to 1, independent co-option of TT8 and MYB5-2 has resulted in the origination of STs and LHs, respectively. Scale bars: 1 mm. d Molecular mechanisms underlying the formation of trichomes/hairs on Nigella petals. Top: a refined functional model for the formation of LHs and STs on petals of N. damascena. Bottom: a deduced model for the formation of T1Hs and T2Hs on petals of N. integrifolia. The interactions among proteins of genes of interest are illustrated in reference to Arabidopsis thaliana. LHs, long hairs; STs, short trichomes; T1Hs, Type 1 hairs; T2Hs, Type 2 hairs.

Discussion

In this study, by applying multiple strategies, we for the first time provided a comprehensive understanding of differences and relationships among different types of protrusive cells on petals. We found that: (1) while the CCs, STs, and LHs on N. damascena petals are all unicellular, they differ not only in shape, length, number, distribution pattern, surface ornamentation, relative nuclear size, and possibly ploidy level, but also in developmental process and molecular basis; (2) NidaMIXTA is involved in CC development, but not in the formation of STs and LHs; (3) the identities of STs and LHs are specified by different combinations of genes encoding components in the hypothetical MBW complex-GL2 module; (4) in contrast to LHs, which are dispensable for the visit of bumblebees, STs serve as tiny pillars to prop open the petal lips and facilitate the access of pollinators to nectar; and (5) STs and LHs may have evolved from LH-like ancestors in Nigella through independent co-option of TT8 and MYB5-2, respectively. These results provide new insights into the developmental and evolutionary mechanisms and the adaptive significance of protrusive cells.

CCs, STs, and LHs: differences and relationships

Early studies have proposed that petal trichomes and CCs are closely related and that the relative timing of MIXTA expression may determine their final fates20,21,22,23, but direct evidence is still lacking. In this study, we revealed that, on petals of N. damascena, while the CCs and the two types of trichomes (i.e., STs and LHs) are unicellular, they are apparently different at multiple levels. First, they differ in cellular features, including shape, length, number, cuticular ornamentation, relative nuclear size, and possibly ploidy level. Second, they experience distinct developmental trajectories, with LHs starting to emerge from S6, but STs from S10, and CCs from S11. Last but not least, while the CC identity gene in N. damascena remains to be identified, CCs and STs/LHs are specified by totally different genetic programs. We found that NidaMIXTA, a homolog of the well-known CC/trichome identity gene, is involved in the cuticle development of CCs, but does not affect STs and LHs. In contrast, LHs and STs are specified by related programs, with NidaMYB5-1, NidaMYB5-2, NidaGL3, and NidaGL2 for LHs, whereas NidaLMI1, NidaMYB5-1, NidaGL3, NidaTT8, and NidaGL2 for STs. By taking the expression patterns and functions of all trichome identity genes into account, we propose that the common regulators NidaMYB5-1, NidaGL3, and NidaGL2 constitute the core of the program, defining the trichome identity of a cell. The developmental fates of STs and LHs mainly depend on the expression timing of the core regulators and the spatial expression of ST- or LH-specific genes.

Interestingly, when we compared the trichome identity networks in N. damascena with those identified in other species, we found two special aspects. One concerns the contribution of MYB family genes to trichome specification. In N. damascena, NidaMYB5s, rather than the homologs of known trichome identity genes MIXTA and GL1, are involved in the specification of STs and LHs. However, MYB5-like genes are usually involved in flavonoid accumulation41,42,43,44, or occasionally participate in trichome development, as reported in A. thaliana and G. hirsutum45,46. The other peculiarity is related to the mechanisms underlying the fate determination of different types of trichomes on one organ. Available data have shown that functional divergence of duplicate genes (e.g., MIXTA duplicate copies in G. hirsutum) 8,9,47 and differential regulation of certain genes (e.g., Woolly [Wo] and GLAND CELL REPRESSORs [GCRs] in S. lycopersicum) 48,49 play critical roles. Yet, our findings revealed a different mechanism in N. damascena. We showed that the component differences in the potential MBW complex underlie the divergence of STs and LHs, with LHs requiring two MYB genes (NidaMYB5-1 and NidaMYB5-2) and one bHLH gene (NidaGL3), but STs relying on one MYB gene (NidaMYB5-1) and two bHLH genes (NidaTT8 and NidaGL3). These findings expand our knowledge of the conservativeness and flexibility of the trichome specification programs across angiosperms.

Evolutionary mechanisms of STs and LHs in Nigella

The evolution of trichomes on Nigella petals involves the concurrent origination of STs and LHs and the formation of their spatial patterns (Fig. 7c). In this study, by comparing the copy number and expression patterns of the six key trichome genes between N. damascena and N. integrifolia, we found that all the investigated genes remained single-copied and that the orthologs of all but two (i.e., TT8 and MYB5-2) showed similar expression patterns. In N. damascena, NidaTT8 and NidaMYB5-2 play essential roles in the formation of STs and LHs on petals, respectively, but in N. integrifolia, neither NiinTT8 nor NiinMYB5-2 was expressed in petal hairs. Notably, TT8-like genes have an ancient origin traceable to the most recent common ancestor (MRCA) of seed plants50, while MYB5-like genes originated later and can be traced back to the MRCA of angiosperms51. Moreover, the general functions of both TT8- and MYB5-like genes in angiosperms are associated with flavonoid biosynthesis41,42,43,52,53,54, with a few exceptions in A. thaliana and G. hirsutum45,46,55. Presumably, the independent co-option of two evolutionarily conserved pigment regulatory genes (i.e., TT8 and MYB5-2) into the existing trichome specification program led to the origination of STs and LHs on petals in Nigella.

Ancestral character-state reconstruction indicated that the loss of hairs on the abaxial side of lower lip lobes was the key to the spatial patterning of STs and LHs on Nigella petals (Fig. 7c). Despite lacking direct evidence, a study in A. thaliana has shown that the abaxial polarity factor KANADI1 (KAN1) represses the trichome initiation on the abaxial side of leaves during the juvenile stage56. It is plausible that a similar mechanism may have evolved during the trichome evolution of Nigella. In addition, based on current and previous studies, we propose that the recruitment of LMI1 into the trichome specification program in the MRCA of Nigella may be responsible for the positioning of STs, for four reasons. First, as revealed in N. damascena and N. integrifolia, the LMI1 genes were specifically expressed surrounding the nectary chamber, overlapping with the region where STs are located36. Second, NidaLMI1 has been reported to be critical for the formation of STs; knockdown of it resulted in the loss of STs36. Third, NidaLMI1 may function as an upstream regulator of the trichome identity genes in N. damascena (Supplementary Fig. 11). Fourth, and most importantly, in Delphinium anthriscifolium (Delphinieae), a species close to Nigella, the LMI1 ortholog was found to regulate the differential growth between the adaxial and abaxial sides of lateral petals, rather than the development of trichomes57. We propose that the LMI1 genes prepattern the localization of trichomes around the nectary chamber in Nigella, providing a positional signal. Further studies to explore how TT8 and MYB5-2 were recruited, and how the regulatory links between the trichome identity program and potential positioning factors (e.g., abaxial polarity genes and LMI1) were established, will provide deeper insights into petal trichome evolution and pattern formation in Nigella.

Adaptive significance of STs and LHs on Nigella petals

Several scenarios have been proposed regarding the possible functions of STs on Nigella petals: filtering pollen, reducing nectar evaporation, and preventing unfavorable visitors32,33. Notably, the prerequisite of these hypotheses is that the gap between the upper and lower lips remains in the absence of STs. However, in this study, we unexpectedly found that the absence of STs caused the closing of petal lips. As a result, most naive and experienced bumblebees failed to open the lips and reach nectar when visiting flowers without STs on petals. Our findings neither support nor reject the previous hypotheses, but instead provide direct experimental evidence that STs act as tiny pillars in maintaining petal lip opening and facilitate the access of suitable pollinators to nectar. Notably, the function of trichomes in regulating the structures of flowers or floral organs has been reported in other species, usually as “linkers”8,58,59. For instance, in G. hirsutum, trichomes on the petal blade link adjacent petals together through a mechanical entanglement8. In cultivated tomato, the interlocking trichomes at anther margins can unite neighboring stamens to generate a closed anther cone and cleistogamy58. Interestingly, trichomes that function as linkers are generally soft, relatively dispersed, and closely intertwined. In contrast, STs that act as pillars on Nigella petals are rigid, densely arranged, and barely intertwined. These findings imply that the functions of trichomes in regulating the structures of flowers and floral organs are dependent on their cellular features.

Notably, while the current study revealed a negligible role of LHs in bumblebee pollination, we cannot rule out the potential functions of LHs for other effective pollinators. It has been shown that when honey bees (Apis mellifera) landed on wild-type flowers of N. damascena trying to gather nectar from petals, they retracted their forelegs and gripped petal lobes with middle legs35, indicating that LHs could potentially provide grip for honey bees. Besides, in N. damascena and N. orientalis, LHs are shiny and reflective under ultraviolet light like pseudonectaries and are mainly distributed on the distal part of the petal35,37. Therefore, they may offer visual signals to pollinators, similar to the function of trichomes on tepal tips of Narcissus tazetta60. Of course, there are other possibilities. LHs might be able to offer tactile signals to stimulate pollinators for further exploration. An interesting example from the deceptive orchid, Ophrys fusca, showed that trichomes on the labellum mimic hairs of female insects, which provide male pollinators with real tactile senses for pseudocopulation61. Alternatively, since most wild Nigella species blossom during hot, dry seasons, LHs may reduce evaporation and maintain moisture on the petal epidermis, helping plants adapt to arid environments, similar to the hairs protecting sunken stomata in desert plants62.

Methods

Plant materials and growth conditions

Seeds of N. damascena and N. integrifolia were purchased from B & T World Seeds (Paguignan, France). They were sown in a 2:1 mixture of vermiculite and nutrient soil and grown at 24 °C, 60% relative humidity, and a 12-h-light/12-h-dark photoperiod. The voucher specimens of N. damascena and N. integrifolia have been deposited in the Chinese National Herbarium (PE), Institute of Botany, Chinese Academy of Sciences, under accession numbers 02604841 and 02604842.

Microscopy and histology

The stereomicroscope images of mature petals were obtained by using a Leica DVM6 digital microscope. For scanning electron microscopy (SEM), the samples were fixed in FAA, containing 3.7% (v/v) formaldehyde, 5% (v/v) acetic acid, and 50% (v/v) ethanol, at 4 °C overnight, and then dehydrated in a gradient water-ethanol series. After being critical-point dried by CO2 and sputter-coated by gold, the samples were observed under a Hitachi S-4800 field emission scanning electron microscope. For histological observations, mature petals were fixed in 4% (w/v) paraformaldehyde (pH 7.2) at 4 °C overnight and embedded in SPI-PON 812 resin. Semi-thin sections (1.5 μm) were prepared on a Leica EM UC7 ultramicrotome, stained with 0.33% (w/v) toluidine blue for 30 s and imaged under a Leica DM5000B microscope equipped with a Leica DFC450 C camera.

Nuclear size measurement and ploidy level determination

To measure nuclear sizes of different types of cells on petals, mature petals of N. damascena and N. integrifolia were fixed in FAA at 4 °C overnight and immersed in ddH2O containing 100 μg mL−1 PI or 5 μg mL−1 DAPI for 10 min, and rinsed in ddH2O for three times. Stained samples were imaged under a Zeiss LSM 980 Elyra 7 confocal laser scanning microscope. PI was excited by a 488 nm diode laser and detected in a 617 nm emission spectrum. DAPI was excited by a 405 nm diode laser and detected in a 465 nm emission spectrum. Optical sections were reconstructed in ZEN 2012. The largest PI fluorescence areas of nuclei were measured in ImageJ. The volumes of reconstructed three-dimensional nuclei were measured using MorphoGraphX as follows63. First, the “Stack/Filters/Gaussian Blur Stack (x, y, z sigma = 0.7)” process was applied. Second, edge detection was performed using “Stack/Morphology/Edge Detect (threshold 10,000)”. The resulting stack was then manually corrected with the Pixel Editor tool to remove noise and misrepresented nuclei. The mesh surface was generated via “Marching Cubes Surface (cube size = 1)” followed by smoothing and subdividing the mesh. Each nucleus was individually seeded, and watershed segmentation was executed. Finally, nuclear volumes were calculated using “Heat Map (Heat Map type: volume; signal average; global coordinates)”.

To determine the ploidy levels of cells in petals of N. damascena and N. integrifolia, we conducted flow cytometry analysis. Petals were dissected under a stereomicroscope, and only regions covering STs or LHs in N. damascena or T1Hs or T2Hs in N. integrifolia were collected. Samples were chopped with a razor blade in 0.5 mL nuclei isolation buffer containing 30 mM sodium citrate, 45 mM MgCl2, 20 mM MOPS, 20 mM NaCl, 20 mM EDTA Na2·2H2O, 0.1% (v/v) Triton X-100, 0.5% (v/v) Tween-20, and 2% PVPK12 (pH 7.0)64. The nuclei were filtered through a 48-μm filter and stained with 5 μg mL−1 DAPI. Nuclear DNA content was analyzed with an LSR Fortessa flow cytometer (BD Biosciences), and then analyzed with FlowJo software. Data were collected for approximately 10,000 nuclei per sample and presented on a linear axis.

Digital gene expression profiling

Reference transcriptomes of N. damascena and N. integrifolia as well as DGE profiles and gene co-expression modules across eight petal stages of N. damascena were retrieved from previous studies36,40. DGE profiles for the four parts of S10 petals of N. damascena (each with four biological replicates) and those for petals at eight stages of N. integrifolia (each with three biological replicates) were generated in this study. The developmental stages of petals were defined according to a previous study32. For each sample, the total RNA was extracted using the SV Total RNA Isolation System (Promega) and a library was constructed and sequenced on Illumina HiSeq2000 (Novogene) for generating paired-end reads of 150 bp. Clean reads were mapped to the corresponding reference transcriptome, and fragments per kilobase per million mapped reads (FPKM) values were calculated by RSEM65. The average FPKM value of replicates was used as the gene expression level in the corresponding sample. Genes having FPKM ≥ 1.0 were defined as expressed.

mRNA in situ hybridization

Total RNAs were extracted from floral buds of N. damascena and N. integrifolia, respectively, as described above and reverse-transcribed using SuperScript III First-Strand Synthesis System (Life Technologies). cDNA fragments of candidate genes were amplified and cloned into the pEASY-Blunt Zero cloning vector (TransGen) for sequencing. Gene-specific fragments were then amplified and used as templates for synthesizing antisense and sense digoxigenin-labeled RNA probes using a DIG RNA labeling kit (Roche). For mRNA in situ hybridization, petals or floral buds of N. damascena and N. integrifolia at different developmental stages were immersed in fresh 4% (w/v) paraformaldehyde (pH 7.2) and embedded in Paraplast (Sigma). Serial sections (8 μm) were prepared and mounted on slides, which were then placed on a slide warmer at 42 °C overnight. After dewaxing, rehydration, and dehydration, the sections were hybridized with RNA probes at 55 °C overnight. After hybridization, the slides were washed in saline sodium citrate (SSC) and incubated with an anti-digoxigenin-AP antibody (Roche) for 2 h at room temperature. The hybridization signals were detected by NBT/BCIP (Roche) color reaction at room temperature overnight in the dark. Images were captured under a Leica DM5000 B microscope equipped with a Leica DFC450 C camera. Primers used for cDNA amplification and probe preparation are provided in Supplementary Data 7; relative positions of the probes to their cDNA sequences are shown in Supplementary Figs. 4, 6, 7, 9, 10, and 14.

Virus induced gene silencing

Fragments used for VIGS overlapped with those used for in situ hybridization (Supplementary Data 3, Supplementary Figs. 4, 6, 7, 9 and 10), which were introduced into the tobacco rattle virus (TRV2)-based pYL156 vector and transformed into Agrobacterium tumefaciens strain GV3101. The strains carrying either TRV1 or TRV2 (containing the target fragment) were cultured separately in LB culture medium (50 μg mL−1 Kanamycin, 50 μg mL−1 Gentamycin, and 25 μg mL−1 Rifampicin) overnight in a 28 °C shaker at 220 rpm. Bacteria were harvested by centrifugation at 4000×g for 15 min and resuspended in MMA buffer (10 mM MgCl2, 10 mM MES, and 200 μM acetosyringone) till the optical density (OD600) of approximately 2.0. Seedlings at the 5–7 true leaf stage were rinsed and immersed in a 1:1 (v/v) mixture of TRV1 and TRV2 bacterial suspensions. After vacuum infiltration for 2 min, the seedlings were transplanted and completely covered with black plastic for 48 h. They were then grown under a 12-h-light/12-h-dark photoperiod until flowering. As a negative control (mock), parallel infiltrations were performed using a 1:1 (v/v) mixture of TRV1 and the empty TRV2 strains. The morphology and micromorphology of petals with potential phenotypic changes were investigated under a stereomicroscope and SEM as described above. To evaluate the silencing efficiency of the candidate gene(s) and the regulatory relationships among interested genes, gene expression levels in VIGS-treated petals were determined by quantitative reverse transcription-polymerase chain reaction (qRT-PCR). Petals at stage 10 were collected from wild-type, mock, and VIGS-treated plants. For each VIGS treatment, two or three biological replicates were included. All reactions were performed with three technical replicates. The housekeeping gene NidaACTIN was used as an internal quantitative control. The relative expression values were calculated using the comparative CT (2−ΔΔCT) method. Primers used for vector construction and qRT-PCR are listed in Supplementary Data 7.

Pollination studies

Bumblebee colonies were purchased from Biobest company (Shandong, China) and maintained in a greenhouse at 25 °C. They were fed on a provided sugar solution at all times, supplemented with pollen every two days. On the day before the pollination study, the bumblebees were separated into different plastic jars, fed with 350 μL 20% (w/v) sucrose solution and isolated overnight in the dark. Of the bumblebees used, the naive ones were those that had never visited flowers of N. damascena, whereas the experienced ones were those that had been trained with wild-type flowers and were capable of opening petal lips. Pollination assays were set in a greenhouse under sunny conditions (25 °C) and performed in a flight cage with the following dimensions: 180 cm (length) × 80 cm (width) × 120 cm (height). For each observation, only one bumblebee was released into the flight cage, where only one plant with a flower at 1 day post-anthesis was displayed. During each foraging bout, i.e., the interval between landing on and leaving a flower, the number of petals with their lips opened and unopened by the bumblebee and the foraging time were recorded (Supplementary Data 4). To avoid repeated use, all experimental bumblebees were marked before returning to the hive.

Phylogenetic analysis

Sequences of MYB5-1, MYB5-2, GL3, TT8, GL2, and LMI1 of N. damascena and N. integrifolia were obtained from the reference transcriptomes. Their homologs from other species were retrieved from publicly available databases by BLAST searches (Supplementary Data 8). For each gene lineage, protein sequences were aligned with Clustal X 2.0.666 and manually adjusted using MEGA X67. Phylogenetic analyses were performed on matrices containing only alignable coding sequences in IQ-TREE 1.6.1068 using the maximum-likelihood method. Analysis was carried out using 10,000 bootstrap replicates. Trees were rooted with genes from Amborella trichopoda.

Statistics and reproducibility

In general, the two-sided Mann-Whitney U test was used to assess significant differences between two samples, while the Kruskal-Wallis test was used in pollination studies, and the unpaired two-tailed Student’s t test was used in qRT-PCR assays. Numbers of samples and repetitions for each experiment were indicated in the figure legends and methods.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.