Mature flowers of Dasymaschalon trichophorum Merr. (X. Guo 20130517-1, Hainan, China [HKU]) and Friesodielsia desmoides (Craib) Steenis (X. Guo 20130630-1, Thailand [HKU]) were collected and preserved in freshly prepared formalin-acetic acid-alcohol (FAA; 1:1:18 mixture of 40% formalin, glacial acetic acid and 70% ethanol) for 24 h, and then transferred to 70% ethanol for long-term storage at room temperature. The specimens were subsequently dehydrated using a tertiary butyl alcohol (TBA) series (Ruzin, 1999) prior to embedding in paraffin wax and sectioning (12 μm thickness) using a rotary microtome. Serial sections were sequentially mounted onto clean slides using gelatin, dried overnight and dewaxed using Histochoice Clearing Agent (Sigma–Aldrich). The slides were then stained with Safranin O and Fast Green, mounted with DPX (Ruzin, 1999), and photographed using a Nikon 80i imaging system (voucher slides deposited in HKU herbarium).

A Bayesian molecular phylogeny of the Dasymaschalon alliance (Guo et al., 2017a) was used to reconstruct ancestral character states. Four perianth characters of diagnostic and functional importance were selected. Character states were scored based on species descriptions and direct observations from living and herbarium materials. The following characters and character states were studied: (i) number of petal whorls: 0 = one whorl; 1 = two whorls; (ii) pollination chamber type: 0 = pollination chamber absent, with petals spreading or loosely coherent (Type I); 1 = partially closed pollination chamber formed by basally constricted petals, with an apical aperture and three basal apertures (Type II); 2 = tightly closed pollination chamber formed by connivent inner petals, with three basal apertures periodically blocked by outer petals (Type III); 3 = tightly closed pollination chamber formed by connivent outer petals (Type IV); (iii) connivence of outer petals when immature: 0 = free; 1 = connivent; and (iv) opening of outer petals during anthesis: 0 = becoming fully open; 1 = remaining apically connivent. The taxon-character matrix for the reconstructions is given in Supplementary Table S1, with variation in floral chambers among the different genera illustrated in Figure 1.

Ancestral character state reconstructions were performed in Mesquite v.3.2 (Maddison and Maddison, 2017) by parsimony and likelihood methods using MCMC trees after initial 50% burn-in as input tree file. A 50% majority-rule consensus tree was summarized from post-burn-in trees and used as the topology for mapping the reconstructions. As the analyses primarily focused on Dasymaschalon and its close relatives Desmos and Friesodielsia, samples of Monanthotaxis, as well as phylogenetically more distant outgroups, were pruned from the tree. The final 46-accession phylogeny included 10 Desmos accessions, 18 Friesodielsia accessions, and 15 Dasymaschalon accessions, with single species representatives of Monanthotaxis, Sphaerocoryne, and Toussaintia as outgroups.

For the parsimony optimizations, character state changes were treated as unordered. The option “Trace over trees” was used to account for phylogenetic uncertainty. Optimizations of each character across the input trees were summarized at each node using the “Count Trees with Uniquely Best States” option, with reconstructions regarded as equivocal when two or more states were estimated as equally parsimonious for a particular node.

For the likelihood approach, the Markov k-state 1 parameter (Mk1) model (Lewis, 2001) was selected for ancestral character state reconstructions: character state changes are regarded as equally probable under this model, and the only parameter is the rate of change. Likelihood reconstruction results were summarized at each node using the “Average Frequencies across Trees” option, which estimates the average likelihood of each state across the input trees.

The vasculature of the second perianth whorl in Dasymaschalon trichophorum is clearly connected to the lateral bundles of the adjacent parts of the first perianth whorl (Figure 5B). The vascular traces leading to various perianth organs are intimately connected, collectively forming a perianth cortical vascular system (CVS) that correspond to the outer and middle whorls widely reported in most other Annonaceae flowers (Deroin, 1989, 1999; Deroin and Le Thomas, 1989; Xue and Saunders, 2013). The two perianth whorls in Dasymaschalon flowers are therefore interpreted here as homologues of the calyx and outer corolla whorl observed in the sister genus Friesodielsia and other Annonaceae (Figure 5).

Cases of reduction or increase in the number of whorls are rare within the Annonaceae, although two clear examples, Fenerivia heteropetala Diels and Dasymaschalon, were highlighted in a review by Saunders (2010). Fenerivia heteropetala flowers possess three highly reduced sepals and 12 petals, comprising: three ovate outer petals; three linear bracteopetals, homologous with inner petals of other Annonaceae species; and six linear andropetals, derived from the outermost stamens (Deroin, 2007; Saunders, 2010). This has been interpreted as the possible result of disruptions to the homeotic control of floral organ identity during development, with a centrifugal shift so that the outermost whorl of six stamens develop as inner petals (Saunders, 2010; Xue and Saunders, 2013).

Conversely, a centripetal change in the homeotic control of floral organ identity has been hypothesized in Dasymaschalon (Saunders, 2010). The absence of inner petals and the reduction in the number of outermost stamens can be explained by reference to putative organ homologies, with the three outermost stamens in Dasymaschalon possible homologues of the three inner petals in the related genera Desmos and Friesodielsia. If this interpretation of organ homology is correct several structural features are expected, including: (1) fusion of the vascular traces of the outermost stamens with the CVS bundles; (2) three vascular bundles supplying the outermost stamens, similar to that observed for inner petals in contrast with typical stamens, which have a single trace; and possibly also (3) sterility of the outermost stamens (i.e., staminodes that lack functional pollen). Serial sections of D. trichophorum reveal that the vasculature of the outermost stamen whorl does not fuse with the CVS, but instead splits from the basal traces of the free stamen bundles (Figures 3I–K), providing no evidence for homology with the inner corolla whorl of Friesodielsia. The outmost stamens have a single vascular trace (Figures 4A,B) and possess four pollen sacs (Figures 4A,B) with fully developed pollen grains that are likely to be functional (Figure 4C). These features do not support the interpretation that the outermost stamens of Dasymaschalon are homologous with the inner petals of Friesodielsia.

According to the classic ABC model (Bowman et al., 1991; Coen and Meyerowitz, 1991) and modified ABCDE model (Rijpkema et al., 2010, and references therein) of flower development, floral organ identity in eudicots can be determined by several classes of homeotic genes (essentially MADS-box genes). The ABC model is also likely to be applicable to the family since it has been tentatively identified in other members of the Annonaceae (Asimina longifolia Kral: Kim et al., 2005): B class genes AP3 and PI were expressed in petals and stamens, and not expressed or only weakly expressed in sepals. In Dasymaschalon, however, the inner petal whorl is likely lost and hence a homeotic conversion in identity is probably not involved. Instead, the likely cause might be the loss of genes that are responsible for determining inner petals or expression failure of these genes. PETAL LOSS genes may influence a petal initiation signal in the Arabidopsis flower (Griffith et al., 1999). In the buttercup family (Ranunculaceae), disruption of the petal identity gene APETALA3-3 is highly correlated with loss of petals (Zhang et al., 2013), and the underlying genetic mechanisms may be complex, with disruption of AP3-3 being either cause or effect (organ identity function likely not involved). The genetic control of floral morphological changes of the Dasymaschalon alliance is largely unknown, and further study involving flower development gene expression approaches (transcriptomes, RT-PCR, and in situ hybridization) is needed to investigate whether similar genes are likely associated with the inner petal loss in Dasymaschalon.

Most Annonaceae species possess partially enclosed pollination chambers with apertures that remain open throughout anthesis so that pollinators can freely enter or depart from the flowers during any phenological stage (Saunders, 2010). Tightly enclosed pollination chambers associated with circadian trapping are rare in the family, only having been reported in Dasymaschalon (Pang and Saunders, 2014; Lau et al., 2017), Friesodielsia (Lau et al., 2017), Goniothalamus (Lau et al., 2016, 2017) and possibly also Artabotrys (Lau et al., 2017).

Dasymaschalon is distinctive amongst Annonaceae species reported to show circadian trapping: the outer petals of Friesodielsia (Figures 7A–K) and Goniothalamus species periodically reflex during anthesis to control the opening and closing of apertures, whereas Dasymaschalon petals remain firmly connivent throughout the anthesis, with closure of apertures due to only minimal lateral petal expansion (Figures 7L,S). Our reconstructions of ancestral character states (Figures 6A,D and Supplementary Figures S1A,D) indicate that loss of inner petals in Dasymaschalon may be associated with evolution of the connivence of the remaining petals during sexually functional stages. If the petals of Dasymaschalon were to reflex as in Friesodielsia, the pollination chamber would lose its ability to prevent beetles from leaving before pollen release. As discussed above, connivence of outer petals during sexually functional stages in Dasymaschalon varies in the degree of contact between adjacent petals, ranging from firmly convergent to only weakly connivent over a small area at the apex. The accumulation of small changes presumably provides further evidence for adaptation in response to selective pressures.

Circadian trapping likely evolved twice in the Dasymaschalon alliance, and multiple times independently in Annonaceae (tribe Annoneae: Goniothalamus; tribe Uvarieae: Dasymaschalon, Friesodielsia; and possibly tribe Xylopieae: Artabotrys), indicating considerable selective advantages of the mechanism. If circadian trapping already evolved in the stem lineage of the Dasymaschalon–Friesodielsia clade, the loss of inner petals in the Dasymaschalon ancestor would have disrupted it. The subsequent changes in Dasymaschalon, with petals that do not fully separate, were possibly adaptations lessening detrimental effects of inner petal loss resulting in re-establishment of the trapping mechanism and its associated selective advantages.