Inflorescence axes show decussate branching, the lateral axes occurring two-by-two, subtended by pairs of opposite bracts. Each pair of bracts and lateral axes is at 90° with respect to the preceding pair. This pattern is continued by the inflorescence unit/flower subtending bracts, also occurring in pairs that are perpendicular to each other (Fig. 2A). The pedicel of each flower has two opposite bracteoles which may be slightly displaced with respect to each other (Fig. 2A). Bracteoles may each subtend (or not) a lateral flower (Fig. 2B), forming together with the first/terminal flower a cymosely branched inflorescence unit (Fig. 2B). Each flower originates from a flower meristem in the axil of a bract(eole) (Fig. 2B–F). Flower primordia are initially covered by a pair of bracteoles (Figs 2B, C, 5A). All perianth parts appear pairwise, continuing the decussate organisation of the inflorescence within the flower. First, two pairs of opposite sepals appear consecutively (Figs 2D, E, 5B), followed by two pairs of opposite petals (Figs 2F, 3A). Within the bract that subtends a flower/inflorescence unit, hairs and colleters develop (e.g. Fig. 2E). Simultaneously, in the flower a floral cup develops (Figs 3A, B, 5C), raising the calyx and corolla. Simultaneously, at the adaxial side of the central floral cup, one by one individual stamen primordia appear in centripetal order (Fig. 5D–F), eventually resulting in two ‘whorls’ of stamens, the distal stamens more developed relative to the proximal ones (Figs 3B, C, 5E). With the proximal stamens appearing, at the bottom of the central depression, two bulges appear surrounding a cavity (Figs 3C, 5E). The bulges develop into a single style with two stigmatic branches (Fig. 3D–H) and simultaneously two locules are formed (Figs 3E, 4A–E), each with a U-shaped placenta with multiple unitegmic ovules (Fig. 4A–E). Simultaneously with the development of the gynoecium, the initial stamen primordia develop into short filaments and basifixed, tetrasporangiate, introrse anthers (Fig. 4A, B, D, F). The bases of the filaments of the stamens are fused (Fig. 3I). Later in the development, a dense whorl of hairs is formed at the base of the single style (Fig. 4F).

In a first stage of the floral development, a calyx with abaxially large trichomes and five petals develop, the latter united at the base (Fig. 6A, B). Simultaneously, five stamens develop (Fig. 6A, B, E, H, I), initially with short filaments. The dorsifixed anthers are tetrasporangiate and introrse and surround the equally developing single style with two stigmatic branches (Fig. 6H, I). With the anthers well developed, a stamen-corolla tube starts developing fastly (Fig. 6C–F) as well as a corolla tube sensu stricto (Fig. 6C–F), which raises the initial petals, now corolla lobes. A dense whorl of trichomes at the bases of the filaments gradually develops, separating the stamen-corolla tube and corolla tube sensu stricto (Fig. 6B–F). Until semi–maturity, the filaments stretch somewhat slower than the corolla tube sensu stricto (Fig. 6C–G). Meanwhile a four-lobed annular nectary surrounding the base of the single style develops (Fig. 6I–K), and below it two locules, each with a U-shaped placenta (Fig. 6I, K) with multiple unitegmic ovules (Fig. 6K, L).

In (semi-)mature flowers, a calyx is present consisting of a short calyx tube and five calyx lobes (Fig. 7A, B). Alternating with the calyx lobes, five free petals are inserted on the hypanthium (Fig. 7A, B). Alternating with the petals, five stamens with dorsifixed, tetrasporangiate and introrse anthers are inserted on the hypanthium, ‘below’ the protruding rim of a conspicuous annular nectary that surrounds the base of a single style (Fig. 7B, C). Below the nectary, the inferior ovary consists of two locules, each with a U-shaped (not shown) placenta with multiple unitegmic ovules (Fig. 7C). At maturity, the stamens are stretched with a long filament (Fig. 7D). The anther is curved on top of the filament, releasing pollen through longitudinal slits (Fig. 7D, E). The style ends in two short stigmatic branches (Fig. 7F).

The inflorescence is essentially decussate and dichasial, with successive pairs of opposite bracts, each subtending a lateral inflorescence axis (Figs 2A, 8A). According to Piesschaert et al. (1997: 205), each inflorescence axis ends in a terminal bud: “...that at first sight may be confused with a single terminal flower. Dissection... reveals a floral meristem where acropetal inceptions of lateral flowers occurs”. In our opinion, this implies that each branch eventually must end in a terminal flower, though Piesschaert et al. (1997) also mentioned that the distal flowers are poorly developed. We conclude that each branch ends in an underdeveloped cymosely branched inflorescence unit. In a typical Rubiaceae inflorescence, usually two oppositely situated bracteoles occur on the pedicel of each flower. From a floral ontogenetic viewpoint, the term ‘bract’ has a relative meaning as it refers to any appendage that subtends a vegetative or generative axis. In this sense, in Rubiaceae, each bracteole on the pedicel of the first flower of an inflorescence unit may act (or not) as a bract by subtending a lateral flower (Weberling 1992; Robbrecht 1988). As a result, such an inflorescence unit ideally consists of a first or terminal flower with a younger, lateral flower in each of its bracteoles. In D. fuscescens, similar inflorescence units occur (Figs 2B, 8A).

Flowers in D. fuscescens at first view show a quite idiosyncratic development, compared to other Rubiaceae, as already mentioned by Piesschaert et al. (1997). In Rubiaceae, a flower is mostly 4- to 5-merous, and usually develops a calyx tube with lobes (the calyx may appear late in the development or not at all), a tubular corolla with distinct lobes, epipetalous stamens in the same number as the calyx and corolla lobes (Fig. 8B, left). In contrast, our present study as well as previous studies (e.g. Piesschaert et al. 1997) show that in Dialypetalanthus, during early development, it is difficult to distinguish between the bracteoles and the first perianth parts, all of which appear as similar pairs of opposite scales (Fig. 2). In total, five of such pairs are formed, which in the course of the floral development reveal their ‘identity’ as bracteoles (one pair), sepals (two pairs), and petals (two pairs), respectively. The bracteoles soon show a slight metatopic displacement with respect to each other (Fig. 2A, E–G). The four sepals and four petals eventually seem to constitute two whorls, respectively a greenish calyx of four small sepals and a corolla of four white petals (Fig. 2A, F), instead of the four whorls of two perianth parts each as described by Piesschaert et al. (1997). Whether the calyx and corolla each consist of two whorls of two members, or of one whorl of four members is rather semantic; at anthesis, the positions and identities of the four petals and sepals suggest two whorls, despite the pairwise development of the perianth members. In this context, we refer to the perianths in e.g. Amaranthaceae but also in many other families, where the five perianth part primordia have a quincuncial aestivation (sequence of appearance of perianth parts and consequent arrangement). Nevertheless, they are considered to constitute one whorl (e.g. Flores Olvera et al. 2011). In that sense, Dialypetalanthus can be compared with the corolla in some Papaveraceae and the calyx in Brassicaceae with four sepals, where the apparent tetramery results from a rather dimerous or successive appearance of floral primordia (e.g. Payer 1857: 210, 218, plates 44, 48). On the other hand, the apparent dimery in Dialypetalanthus seems systemic in the vegetative as well as the generative parts of the plant, which suggests a deep regulatory origin which goes beyond so-called isomerous changes (Ronse De Craene 2016) within the flower. Without a doubt, in Dialypetalanthus, the opposite arrangement of petals and sepals contributes to the atypical floral morphology for a genus in Rubiaceae.

Moreover, during floral development, the initially slightly convex floral apex (Fig. 2B) becomes conspicuously concave. Actually, a floral cup sensu Leins and Erbar (2010) is formed by the development of a stamen-corolla-calyx tube (Figs 3A, 8B right; see also figure 99 in Leins and Erbar 2010: 102). Sepals and petals are lifted up by the development of this stamen-corolla-calyx tube. We consider the presence of a stamen-corolla-calyx tube to be the morphological link, hitherto missing, with the other Rubiaceae, since strictly speaking, the sepals should be seen as calyx lobes and the petals as corolla lobes. Hence, it can be argued that also in Dialypetalanthus, there are no free sepals nor petals since the sequence of appearance and subsequent arrangement of the calyx and corolla lobes is not related to the appearance of individual perianth part primordia. The pairwise sequence of appearance may be due to spatial constraints.

During the development of the perianth lobes, individual stamen primordia appear in a more or less centripetal order at the adaxial side of the central floral cup (Figs 3B–C, 8B right). Eventually, two ‘whorls’ of stamens develop and fill up the initial floral cup. There is no visible primary androecium primordium but individual stamen primordia seem to originate from an annular meristem surrounding the floral apex. Initially, we were inclined to interpret the androecium as consisting of four parts, each one positioned below the base of a corolla lobe (petal) upon the common stamen-corolla-calyx socle. However, our observations do not support this hypothesis (Fig. 3). Moreover, Piesschaert et al. (1997) already mentioned that the androecium falls off as a whole. Since indeed, all stamens appear to be connected to each other at the very base of the filaments, we conclude that they originate from an annular meristem at the inner (adaxial) side of a short stamen-corolla-calyx tube. The fact that at later developmental stages there are two ‘whorls’ of stamens is rather a consequence of the spatial constraints within the central cavity in the flower. Consequently, although the stamens seem free-standing at first glance, they should be considered as epipetalous on the stamen-corolla-calyx tube. If we consider the usual epipetaly in Rubiaceae with stamens positioned on a stamen-corolla tube as epipetaly sensu stricto, the epipetaly in Dialypetalanthus can be considered as another character state (Ochoterena et al. 2019).

The polyandry exhibited by D. fuscescens can theoretically find its origin in either ‘dédoublement’ of initially individual stamen primordia or polygenesis (Ronse De Craene 1988; Ronse Decraene and Smets 1993), or by the development of an annular primary androecium primordium, from which secondarily stamen primordia originate. In the first case, one might expect that the stamens are grouped in four ‘groups’ of stamens, the position of each corresponding to the position of the initial stamen. Our results show that this is not the case. In contrast, our observations suggest that the many stamens in D. fuscescens originate from an annular meristem on the adaxial flanks of the central floral cup (which can also be considered as a primary androecium primordium since a particular structure, the central floral cup, is formed preceding the appearance of the stamens). The androecial development in D. fuscescens resembles the development of the numerous perianth parts, often called silky hairs, in flowers of the Cyperaceae genus Eriophorum (Vrijdaghs et al. 2005), which also originate one by one and centripetally, from an annular perianth meristem. Apparently, within angiosperms, there is a potential to switch from a limited default number of individual floral part primordia within a whorl to a common annular meristem/primary primordium from which a large number of floral parts secondarily develops.

Only after the appearance of most stamens, the development of the gynoecium starts, similar as in other Rubiaceae (De Block and Vrijdaghs 2013; Vrijdaghs et al. 2020), though without gynoecial nectary (Fig. 3C; Figueiredo et al. 2017). Perhaps, investing in polyandry instead of a nectary gives an ecological advantage to Dialypetalanthus, but this is a subject for another study.

The floral morphology in M. beccariana differs from that in most Rubiaceae studied by the absence of a tubular corolla and epipetaly. However, in contrast to the flowers of Dialypetalanthus, the flowers of M. beccariana have floral features that are considered to be common in Rubiaceae, such as a calyx tube, the same number of stamens as corolla lobes (no polyandry) and the presence of an annular nectary surrounding the base of the single style.

The development of the flower in P. exsertus concurs with the floral development as described by Vrijdaghs et al. (2020) for Crucianella Boiss. and Phuopsis (Griseb.) Hook.f., both genera belonging to the Rubioideae-Rubieae. The floral ontogeny in P. exsertus is characterised by the development of a stamen-corolla tube as well as a corolla tube sensu stricto and corresponds with figure 12C in Vrijdaghs et al. (2020: 484). However, in P. exsertus, the development of the stamen-corolla tube is initially somewhat delayed compared to the flowers in Crucianella and Phuopsis. Adaxially and at the base of the stamens, a thick whorl of hairs develops. Also the development of the androecium and inferior and bilocular gynoecium occurs as observed in all other Rubiaceae studied.

According to Puff and Igersheim (1994), the corolla of the two then known Mussaendopsis species consists of a tube and lobes. This is in contrast to our observations in semi-mature flowers. We cannot exclude that very late in the floral development, shortly before anthesis, a stamen-corolla tube is still formed, which would explain the observations of Puff and Igersheim (1994). However, this seems improbable, since no indications of a common (meristematic) base of petals and stamens were found as for example is the case in the genera Pentodon (Spermacoceae) and Sacosperma (Knoxieae) (Vrijdaghs et al. 2015). Of the two taxa compared here with Dialypetalanthus, P. exsertus is also neotropical. Yet it is the floral morphology of the Asiatic M. beccariana that shares two unusual features with that of D. fuscescens, namely the free petals and free stamens, while the floral morphology of P. exsertus is similar to what may be expected from a Rubiaceae flower.

Following the hypothesis of Vrijdaghs et al. (2020), Table 3 compares the floral development and morphology in D. fuscescens with that in M. beccariana and P. exsertus. It follows that: 1) the typical and quite common floral Bauplan in Rubiaceae is maintained, albeit by means of a stamen-corolla-calyx tube instead of a stamen-corolla tube, 2) the development of floral whorls (in casu, the androecium) from a fixed number of individual primordia can be reorganised into an annular meristem from which multiple whorl members secondarily originate. This was also observed in other families than Rubiaceae (e.g. Ronse De Craene 1988; Vrijdaghs et al. 2005; Ronse Decraene and Smets 1993).