Evolutionary relationships, biogeography and morphological characters of Glinus (Molluginaceae), with special emphasis on the genus composition in Sub-Saharan Africa

Abstract Glinus is a small genus of Molluginaceae with 8–10 species mostly distributed in the tropics of the World. Its composition and evolutionary relationships were poorly studied. A new molecular phylogeny constructed here using nuclear (ITS) and chloroplast (rbcL, trnK-matK) markers confirmed the monophyly of the genus. Based on ITS analysis, the following well-supported lineages are present within Glinus: the G. bainesii lineage is recovered as sister to the remainder of the genus followed by G. oppositifolius. Three other clades are: G. hirtus with G. orygioides; G. radiatus and G. lotoides; the latter is represented by a sample from North America, and G. zambesiacus as sister to G. setiflorus + G. lotoides + G. dictamnoides. On the plastid gene tree, G. bainesii + G. oppositifolius form a sister clade to all other Glinus species. The next clade is formed by G. hirtus and G. orygioides followed by G. radiatus plus an American sample of G. lotoides. The next branch comprises G. setiflorus as sister to G. zambesiacus + G. lotoides + G. dictamnoides. Glinus seems to have originated from Africa around the Late Eocene or Early Miocene, with further radiations to Australia and the Americas during the Late Miocene or Late Pliocene. Compared with the previous limited character set used for the diagnostics, we have found ten new morphological and carpological traits distinguishing Glinus members. In both trees based on nuclear and plastid datasets, the major phylogenetic clades cannot be characterized by the peculiar morphological characters. Many shared character states leading to their contrasting pattern in the multivariate analysis model are interpreted as a high homoplasy in the phylogenetically distant species. We paid special attention to the composition of the genus in Sub-Saharan Africa, a region with the greatest species diversity. Our results provide new insight into the taxonomy of Glinus in this region. Glinus lotoides var. virens accepted in many previous works is a synonym of G. dictamnoides that is closely related to G. lotoides based on molecular analysis and morphological characters. The status of the American populations of G. lotoides needs further investigation due to different characters of the specimens from the Old and the New World. Many specimens previously identified as G. lotoides var. virens and as the intermediates G. lotoides × G. oppositifolius belong to G. zambesiacussp. nov. and G. hirtuscomb. nov. (≡ Mollugo hirta); the latter species is resurrected from synonymy after 200 years of unacceptance. In some African treatments, G. hirtus was known under the invalidly published name G. dahomensis. Glinus zambesiacus is distributed in the southern and eastern parts of tropical Africa, and G. hirtus previously assumed to be endemic to West Africa is indeed a species with a wide distribution across the tropical part of the continent. Glinus microphyllus previously accepted as endemic to West Tropical Africa together with other new synonyms (G. oppositifolius var. lanatus, G. herniarioides, Wycliffea rotundifolia) is considered here as G. oppositifolius var. keenaniicomb. nov. (≡ Mollugo hirta var. keenanii), a variety found across the entire distribution of G. oppositifolius (Australia, Asia, and Africa). The presence of the American G. radiatus in Africa is not confirmed, and all records of this species belong to G. hirtus. The lectotypes of some names (G. dictamnoides, G. herniarioides, Mollugo hirta, M. setiflora, Pharnaceum pentagynum, Wycliffea) as well as a neotype of G. trianthemoides are designated. A new key to the identification of all Glinus species in Sub-Saharan Africa is provided. A checklist is given of all accepted species in this region (G. bainesii, G. hirtus, G. lotoides, G. oppositifolius s.l., G. setiflorus, and G. zambesiacus) with their nomenclature, morphological description and geographical distribution.


Introduction
Glinus L. comprises six to ten species distributed in the warm regions of the world (Backer 1951;Endress and Bittrich 1993;Thulin et al. 2016). It includes annual and perennial herbs and rarely subshrubs covered with simple or stellate hairs, with exstipulate, lanceolate to obovate leaves arranged in false whorls of 3-5, several to many verticillate, sessile or pedicellate flowers, pentamerous perianth consisting of free segments that are green dorsally and white, pink or yellow ventrally, petaloid staminodes often present originating from the outer stamen whorl, 3-30 stamens and 3-or 5-valvate, strongly hydrochastic capsules with numerous small arillate seeds. Among Caryophyllales as well as Molluginaceae, Glinus is characterized by the remarkable seed aril divided into two parts: a white, usually well-visible hood covering the funiculus and a large ribbon-like appendage (Pitot 1965;Narayana and Lodha 1972;Endress and Bittrich 1993;Sukhorukov et al. 2018). Molecular phylogenetic analyses based on four species (G. lotoides L., G. oppositifolius (L.) Aug.DC., G. radiatus (Ruiz & Pav.) Rohrb. and G. setiflorus Forssk.) suggest the genus is monophyletic (Christin et al. 2011;Thulin et al. 2016).
Species delimitation within Glinus is usually based on the pubescence details (stellate vs. simple trichomes; presence of tiny prickles on stem and leaves), leaf dimensions and shape, number of flowers in the leaf axils, and number of stamens (e.g., Bogle 1970;Short 2002;Lu and Hartmann 2003;Vincent 2003). These character sets are sufficient for the delimitation of G. lotoides and G. oppositifolius, the most widespread species in the Old World (e.g., Hutchinson and Dalziel 1927;Backer 1951;Jeffrey 1961;Gonçalves 1978;Hargreaves 1995;Gilbert 2000;Short 2002; Retief and Meyer 2017). In some cases, seed size and ultrasculpture are useful tools for identification, e.g., in the characterization of G. setiflorus, G. orygioides F.Muell. and G. bainesii (Oliv.) Pax (Sukhorukov et al. 2018). Additional characters (pubescence density and pedicel length) were used for the infraspecific descriptions of the morphologically heterogeneous G. lotoides and G. oppositifolius (Gonçalves 1965(Gonçalves , 1970(Gonçalves , 1978. There was an attempt to divide Glinus into two subgeneric groups called Euglinus (≡ Glinus) and Pseudoglinus Endl. mainly based on stellate vs. non-stellate pubescence (Endlicher 1840), but this classification was never used.
The members of the genus are unevenly distributed across the tropics and subtropics of the world with most species present in Africa. In total, seven Glinus species are currently accepted in all parts of the continent APD 2019). Glinus lotoides is present throughout Africa (Jeffrey 1961;Maire 1962;Gonçalves 1978;APD 2019). Glinus oppositifolius possesses a similarly wide distribution, but is not reported from North Africa (Maire 1962;Boulos 1999;Hassan et al. 2005). Other species are considered to be restricted to smaller regions. Glinus setiflorus is widespread in E and NE Africa (Jeffrey 1961;Gilbert 1993aGilbert , 2000. Glinus dahomensis A.Chev. and G. microphyllus Hauman are considered to be local endemics to different territories of West Africa (Chevalier 1938;Lebrun and Stork 2003). Another African species, G. bainesii is restricted to Botswana, Mozambique, Zimbabwe and the eastern part of South Africa (Adamson 1961;Gonçalves 1978). A single adventive species G. radiatus (Ruiz & Pavon) Rohrb. from the Americas was referenced in some West African countries (Berhaut 1967(Berhaut , 1979Lisowski 2009;Thiombiano et al. 2012;Schmidt et al. 2017). Additionally, it is postulated that G. lotoides and G. oppositifolius seem to freely hybridize in Africa with a large number of intermediates (Adamson 1961;Jeffrey 1961;Gonçalves 1970Gonçalves , 1978. Numerous specimens usually labelled as "G. lotoides × G. oppositifolius" are present in many herbaria. The reason for this identification is because of the less prominent stellate pubescence in these Glinus specimens compared to the typical densely pubescent specimens of G. lotoides. However, such hybrids were not reported from other parts of the Old World. In a previous carpological study of all Molluginaceae (Sukhorukov et al. 2018), the disparity of many specimens belonging to the same Glinus species in the herbaria visited was noted. In fact, the limited number of diagnostic characters used and absence of detailed taxonomic treatment do not allow for an evaluation of the real diversity of Glinus in tropical Africa, the most species-rich region of the genus in the world. For this reason, a critical study of the taxonomy, morphology and distribution of the genus in Sub-Saharan Africa would be desirable. Therefore, the main aims of the present study are: (1) to conduct an expanded molecular phylogenetic analysis of Glinus worldwide, with further implications on its divergence and origin; (2) to deeply investigate the morphology and species distribution of Glinus in tropical Africa, the most species-rich region, and to provide precise diagnoses through an updated taxonomic treatment.

Taxon sampling
In the molecular phylogenetic analysis, 149 accessions representing 31 species and 10 genera including outgroups were sampled. Except for the Indian G. ononoides where samples could not be amplified because of no recent collection, all other Glinus species were sampled. In total, nine Glinus species representing their entire distribution were sampled. A list of all samples used in this study is presented in Table 1. Table 1. Voucher information and GenBank accession numbers for the species of Molluginaceae and outgroups included in the phylogenetic analysis (arranged in alphabetical order). The newly sequenced samples are highlighted in bold. DNA extraction, amplification and sequencing 5-10 mg of dried herbaria leaf samples was used to isolate DNA using the CTAB protocol (Doyle and Doyle 1987). One nuclear (the nuclear ribosomal internal transcribed spacer, nrITS) and two plastid markers (the coding gene rbcL and the region encompassing trnK introns and matK coding genes, trnK-matK) were selected for phylogenetic analysis. The primers ITS4 and ITS5 (White et al. 1990) were used to amplify the ITS region. The rbcL and matK-trnK primer sequences were taken from Christin et al. (2011). Due to the fact that the samples were taken from herbaria and had a rather long storage term, various combinations of the forward primers trnK-matK_For A, C, G, and H and the reverse primers trnK-matK_Rev C, D, F and I were used. For rbcL, two external rbcL_4_For and rbcL_1353 and two internal primers rbcL_629_For and rbcL_760_Rev were used (Christin et al. 2011). PCR reactions for all primers were performed in a total volume of 25 μl, using 5 μl DNA (10 ng/μl), 1 μl of each primer, 0.5 μl Encyclo polymerase (Evrogen, Russia), 0.5 μl 50× dNTP, 5 μl 10× Encyclo buffers and 14.5 μl mQ. PCR amplification of nrITS primers was performed under the following conditions: initial denaturation for 3 min at 94 °C; 37 cycles of 1 min denaturation at 94 °C, 30 s annealing at 51 °C, 150 s extension at 72 °C, and final extension of 10 min at 72 °C (Thulin et al. 2016). PCR amplification for rbcL was performed under the following conditions: initial denaturation for 10 min at 94 °C; 34 cycles of 30 s denaturation at 94 °C, 30 s annealing at 48 °C, 150 s extension at 72 °C, and final extension of 10 min at 72 °C. The following PCR program was used for matK-trnK primers: initial denaturation for 10 min at 94 °C; 34 cycles of 30 s denaturation at 94 °C, 30 s annealing at 51 °C, 2 min extension at 72 °C, and final extension of 10 min at 72 °C.
PCR products were cleaned with Cleanup Mini BC023S Kit (Evrogen, Russia). Sanger sequencing was carried at Evrogen JSC (Moscow, Russia) using the same primers as in the PCR.

Phylogenetic inference and molecular dating
Sequences were aligned using MUSCLE v. 3.5 (Edgar 2004) and the alignment was adjusted manually using PhyDe (version 0.9971; Müller et al. 2010). Three separate analyses were performed for the nuclear and plastid DNA datasets using maximum parsimony (MP), Bayesian inference (BI) and maximum likelihood (ML). Due to a conflict between G. zambesiacus and G. setiflorus, all subsequent analyses were conducted using the separate datasets. Models of nucleotide substitution were chosen according to the Akaike information criterion using jModelTestv.2.1.7 (Guindon and Gascuel 2003;Darriba et al. 2012) for each gene separately. The best-fit substitution model for both the nuclear and plastid datasets was GTR + G. For the ML analyses, we used RAxML Version 8 (Stamatakis 2014). Bootstrap analyses were conducted with 2500 replicates for ML. Parsimony analyses were conducted in PAUP* 4.0a162 (Swofford 2002) with the following settings: all characters have equal weight, MaxTrees set to 1000 (auto increased by 1000), TBR branch swapping and with 20000 jackknife (JK) replicates to calculate node support. Final trees were edited in TreeGraph ver. 2.14.0 (Stöver and Müller 2010).
Divergence times for Glinus were estimated using a Bayesian uncorrelated lognormal relaxed clock under a birth-death speciation process (Gernhard 2008) for the nuclear and plastid datasets separately. We selected a normal distribution for the secondary calibration with a standard deviation of 8.5, equivalent to the 95% HPD estimate of Yao et al. (2019) for the crown of Molluginaceae. Four independent MCMC analyses were run, each of 20 million generations sampling every 2000. The analyses were run using BEAST 2.4.5 (Bouckaert et al. 2014) on the CIPRES Science Gateway 3.3 (https://www.phylo.org; Miller et al. 2010). Output log files were analyzed using TRACER 1.6 (Rambaut et al. 2014) to assess convergence and ESS of all parameters. As "burn-in", 10% of samples were removed prior to combining the independent runs using LOGCOMBINER 2.4.5 (Bouckaert et al. 2014). The MCC tree was generated using TREEANNOTATOR 2.4.5 (Bouckaert et al. 2014).

Biogeographical analysis
Geographical distributions of all species were compiled from herbarium specimens and field work (see section "Field and herbarium studies" below). Due to the wide distribution of some species (G. hirtus, G. lotoides, G. oppositifolius, G. radiatus), the biogeographical analysis is based on the continents and not the floristic provinces. Four geographical areas were identified: A -Africa including Madagascar; B -Asia; C -Australia, and D -America (including Galapagos Islands). The BI gene trees were pruned to remove all duplicate accessions and G. dictamnoides using the drop.tip function in the package ape (Paradis et al. 2004). The nuclear and plastid gene trees used for the analyses had 32 accessions each corresponding to 31 species. Two accessions of G. lotoides were included representing the Old World and the American populations. The coded geographic data is available in Table 2.
Ancestral range estimation (ARE) was conducted using the R package "BioGeo-BEARS" (Matzke 2013(Matzke , 2014. Out of the six models explored in this study, the DEC+J model was the best fit based on the AIC and likelihood ratio test (LRT) results (see Table 3). The analyses were unconstrained (without possible dispersal routes or ancestral areas assumed a priori). We allowed the inferred ancestor to occupy a maximum of three areas, corresponding to the maximum number of areas occupied by any extant species.
Based on the likelihood and AIC values, the best fit model was the DEC +J model for both nuclear and plastid datasets (Table 3). Both datasets had the same results and only varied at the divergence times.

Multivariate analysis
The same species set of Glinus as in the molecular phylogenetic analysis were included in the character matrix. The varieties of G. oppositifolius (var. glomeratus and var. keenanii) that deviate in some states of the studied characters as well as G. ononoides and G. sessiliflorus were not included. The multivariate analysis aims to test whether the morphological and carpological character subdivision corresponds with the phylogenetic reconstructions. In our previous papers, multivariate analysis provided a good support for the non-stochastic distribution of the characters in major clades of the entire Molluginaceae (Sukhorukov et al. 2018) and the genus Microtea, Microteaceae (Sukhorukov et al. 2019).
Different Glinus species were classified by group average linkage algorithm of cluster analysis constructed on a Gower similarity matrix (Gower 1971) based on seventeen characters including general morphology (life history, pubescence, leaves) and reproductive traits. This approach recognizes the species grouping based on similar characters, but does not provide a true phylogenetic context. The reliability of grouping was assessed at the level p<0.05 using SIMPROF algorithm (Clarke 1993;Clarke and Warwick 2001). Calculations were performed using PRIMER 6.1.6 statistical software (Clarke and Gorley 2006).

Carpological study
The seeds were investigated using scanning electron microscopy (SEM) and anatomically. The hard seed coat does not require any special preparation prior to SEM due to absence of any trichomes on its surface. The seed colliculae if present are the thickenings of the outer walls of the testa cells. After sputter coating the material with goldpalladium, the SEM observations were made with a JSM-6380 microscope (JEOL Table 4. List of species and vouchers used in the carpological analysis. The specimens used for both anatomy and SEM analyses are marked with an asterisk (*) after the herbarium acronym. The samples of the widely distributed species G. lotoides and G. oppositifolius originated from different regions of the World.
Ltd., Japan) in the Laboratory of Electron Microscopy of Moscow State University. The anatomical cross-sections of seeds were prepared using a rotary microtome Microm HM 355S (Thermo Fisher Scientific, USA). Before sectioning, the seeds were soaked in water:alcohol:glycerin (1:1:1) solution, dehydrated in an ethanol dilution series and embedded in Technovit 7100 resin (Heraeus Kulzer, Germany). The crosssections were observed using a Nikon Eclipse Ci microscope and photographed with a Nikon DS-Vi1 camera (Nikon Corporation, Japan) at the Department of Higher Plants (Moscow State University). The list of samples used for the SEM is provided in Table 4.

Choosing the territory for taxonomic study
Due to the fact that Glinus has the highest taxonomic diversity in the tropics of Africa but is represented only by a single widespread species (G. lotoides) in North Africa (Maire 1962;Boulos 1999) and a poorly known local endemic in Luxor (G. runkewitzii Täckh. & Boulos), which is closely related to G. lotoides (Täckholm and Boulos 1972), we thus exclude the northern part of the continent (Algeria, Egypt, Libya, Morocco and Tunisia) from our investigations. The territory under study corresponds to the geographical concept designed for the framework for the investigation of biodiversity and conservation of Sub-Saharan plants Klopper et al. 2006). . Only a part of the images seen from BOL, LISC and PRE were exactly identified and cited in the present article; other specimens from these herbaria requiring more detailed analysis were not cited here. No Glinus specimens were seen from Equatorial Guinea, Lesotho and West Sahara. Distribution maps are based on the specimens cited in the text and were prepared using SimpleMappr online tool (http://www.simplemappr.net).

Nomenclature
Protologues of each plant name involved were examined for valid publication, legitimacy, and other nomenclatural issues in agreement with the International Code of Nomenclature for algae, fungi, and plants (Turland et al. 2018). As far as possible, original material was traced and type specimens were cited or designated here.

Molecular phylogeny and dating
The combined plastid dataset of rbcL and matK-trnK comprises 4009 aligned bp and 48 accessions while the nrITS dataset has 760 aligned bp and 47 accessions. The ML and BI analyses revealed identical topologies, although slightly different from the MP (see Figs 1,2). In all three analyses, Glinus is resolved as monophyletic. In the nuclear gene trees, Glinus is sister to Trigastrotheca F.Muell. (Fig 1; BSL 89; PP 0.97) while in the plastid gene trees it is recovered as sister to Mollugo (Fig 2; BSL 100; PP 1). In the parsimony analyses, Glinus is sister to Mollugo in both the plastid (BSP 100) and nuclear (BSP 100) gene trees.  In the plastid gene trees, G. bainesii + G. oppositifolius form a clade sister to all other Glinus species. In this clade, G. hirtus + G. orygioides are in turn sister to the remaining species. Glinus radiatus plus the northern American G. lotoides sample are sister to a clade composed of G. setiflorus, G. dictamnoides, G. zambesiacus and G. lotoides (Fig. 2).
In the nuclear gene tree, G. bainesii is recovered as sister to the remaining species. The G. oppositifolius lineage is recovered as sister to a well-supported clade of (1) G. hirtus + G. orygioides; (2) G. radiatus and the North American G. lotoides and (3) G. zambesiacus as sister to G. setiflorus + G. lotoides + G. dictamnoides (Fig. 1).
The plastid and nuclear gene trees had a conflict regarding the position of G. zambesiacus and G. setiflorus. In the plastid gene tree, G. setiflorus is well-supported (BSL 100; BSP 100; PP 1) as sister to a clade composed of G. zambesiacus, G. dictamnoides and G. lotoides whereas in the nuclear gene tree G. zambesiacus is well-supported (BSL 99; BSP 100; PP 0.98) as sister to a clade composed of G. setiflorus, G. dictamnoides and G. lotoides (Figs 1,2).
Except for the crown node of Molluginaceae, the other nodes show very different ages for both the nuclear and plastid gene trees. The nuclear gene tree had much older node ages compared to the plastid gene tree (Fig. 3). Molluginaceae started to diversify during the Late Paleocene at ~58.97 mya (95% HPD 42.07-76.03 mya) or ~58.81 mya (95% HPD 41.8-76.03 mya) based on the nuclear and plastid trees, respectively. The diversification of Glinus started around the Late Eocene ~39.5 mya (stem age, 95% HPD: 13.33-30.85) or around the Middle Miocene ~13.47 mya (stem age, 95% HPD: 7.82-20.06 mya) for the nuclear and plastid trees, respectively (Fig. 3).

Biogeographical analysis
In the biogeographical analysis, the nuclear and plastid gene trees based on the reduced data showed the same topology but only varied with the divergence times. The nuclear and plastid trees based on the reduced data showed the same topology, as such similar biogeographic results only differing by node ages. The ancestral area of Glinus is uncertain (see Fig. 4 ACD: p = 0.21; AD: p = 0.18; ABD: p = 0.16; A: p = 0.10). The ancestral area of the crown node of Glinus remains uncertain ( Fig. 4; ABC: p = 0.23; A: p = 0.19; AC: p = 0.19; AC: p = 0.19; AB: p = 0.14) even though it seems to be connected to Africa. There have been two shifts from Africa, one to Australia for G. orygioides and the other to America for G. radiatus and the American G. lotoides clade.

Diagnostic characters revisited
We coded 17 characters (14 morphological and 3 anatomical characters) used for the multivariate analyses (see Table 5). Out of the 17 characters coded, 10 are used for species delimitation for the first time (characters 4, 7, 8, 10, 11, 13, 14, 15, 16 and 17; see Table 5). The following characters and states were coded.   Table. To be consistent with the molecular phylogeny, the same species set was used for the multivariate analysis (G. ononoides and G. sessiliflorus were included in the Table, but they are absent in both molecular and multivariate analyses).
Note. Two seed types were observed in G. hirtus: seeds with smooth surface found in several specimens only (Fig. 6C, D) and more common colliculate ones (Fig. 6E, F). The colliculate ultrasculpture is the most common type in almost all species and their varieties (Figs 6E, F, 7A-D, 8A-F, 9C-H). All Glinus radiatus samples have a smooth seed surface (9A, B), as well as those of G. ononoides (Fig. 6G, H) and some specimens of G. hirtus (Fig. 6C, D).
Note. Due to the presence of the colliculae making the seed coat more robust in these areas, the thickness of the seed coat was measured between them.

Multivariate analysis
The data of the multivariate analysis were evaluated using the matrix of the characters and their states provided in the Table 5. The results of cluster analysis of all characters suggest the existence of four significantly different groups within Glinus, these branches are highlighted in black ( Fig. 11) The groups are significantly (p<0.05) distinct on different levels of Gower's index.

Figure 10. Seed coat cross-sections
zambesiacus. Scale bar: 10 μm. Glinus dictamnoides has the same seed coat structure as G. lotoides and is not illustrated here. Origin of the material is provided in the Table 4 and is designated with an asterisk (*).

Morphological interpretation of the phylogenetic results
Monophyly of Glinus is not surprising because all species share the same unique trait (presence of seed aril) not encountered in other genera of Molluginaceae. Mollugo s.str. was suggested to be closely related to Glinus after the first molecular studies (Christin et al. 2011;Thulin et al. 2016), but this is only partially supported by our results. The merging of Glinus into Mollugo s.l. previously undertaken by several authors (e.g., Bentham in Bentham and Mueller 1866; Oliver 1871; Trimen 1894; Clifton 2003) cannot be accepted, even though species of both Mollugo and Glinus share similar morphological and carpological characters, e.g., whorled leaves, leafy inflorescences, multiseeded capsules, colliculate seeds with a relatively thick seed coat with stalactites. The carpological differences between the related genera were determined by Sukhorukov et al. (2018) based on an extended seed analysis.
The close relationship between the Australian G. orygioides and G. hirtus is unexpected from a morphological point of view. These species share only five character states (4:0, 10:1, 12:0, 16:1, 17:1). Compared with other Glinus species, G. orygioides differs by being a subshrub, while G. hirtus has no peculiar character states. Surprisingly, G. hirtus shares eleven character states (mostly gross morphological) with the unrelated G. radiatus. From the six states distinguishing these two taxa, only one (seed colour) is visible to the naked eye; however, some G. hirtus specimens have yellow seeds (13:0), a usual character state in G. radiatus. The other five character states are micromorphological (length of anthers and stylodia, seed ultrasculpture, presence of stalactites in the testa and its thickness). This similarity in the gross morphology has caused the misidentification of both species.
The results of the molecular analysis support the polyphyly of G. lotoides; a sample from the New World forms a clade together with G. radiatus. The specimens of American G. "lotoides" seen by us have yellow or bright brown seeds, like G. radiatus (not dark red or almost black as in G. lotoides s.str.: state 13:2 in the Table 5), and differ from it by colliculate seed sculpture (e.g., Thieret 1966) and larger perianth segments (Christy 1998). We provisionally accept only one native species in the Americas (G. radiatus s.l.). The close relationship between the Old World G. lotoides specimens and G. setiflorus is supported by many identical character states (Table 5; Figs 7A-D, 10C, H), and they both share nine states with G. zambesiacus. In the absence of a well-resolved relationship between G. lotoides and G. dictamnoides in the plastid and nuclear gene trees, we suggest that G. dictamnoides be treated as a synonym to G. lotoides, as was proposed in the earlier studies (e.g. Backer 1951;Zohary 1966;Hassan et al. 2005).
Due to the discordance between the gene trees and multivariate analysis of morphological characters, we cannot propose any infrageneric groups within Glinus. We assume that character states shared between phylogenetically distant taxa should be interpreted as homoplasies. Similarly, the former genus subdivision proposed by Endlicher (1840) and based on the pubescence details is also not supported.

Biogeography
Both the plastid and nuclear gene trees suggest that Glinus started to diversify during the Neogene. Even though our results do not indicate a clear origin of Glinus, it seems to be connected to Africa (Fig. 4). Origin and adaptation to Neogene aridification in Africa has also been reported in many other plant lineages such as Acridocarpus Guill. & Perr., Malpighiaceae (Davis et al. 2002), Coccinia Wight & Arn., Cucurbitaceae (Holstein and Renner 2011), Guibourtia, Fabaceae (Tosso et al. 2017), Manilkara Adans., Sapotaceae (Armstrong et al. 2014) and the tribe Melastomateae, Melastomataceae (Veranso-Libalah et al. 2018). The Australian Glinus orygioides and the American G. radiatus group probably originated during the Late Miocene and Pliocene based on the plastid and nuclear gene trees, respectively. Long-distance dispersal might be the most appropriate explanation for migration of the species to Australia and America during the Neogene.

Extant geographical distribution
Two species from the basal clade(s) -G. oppositifolius and G. bainesii -prefer different climates. Glinus bainesii is well adapted to the hot semi-arid climate [climate classification used here is according to Köppen (1936), with additions by Geiger (1961)].
Glinus oppositifolius is more frequently found in regions with tropical rainforest and savanna climates. In the regions with hot semi-arid or desert climates, it clearly prefers the habitats near water sources (e.g., river banks). Another widely distributed species, G. lotoides, as well as East African G. setiflorus, are drought-adapted species and avoid territories with tropical rainforest and monsoon climates.
Altogether, we accept 9-10 species in (sub)tropical parts of the World. These can thrive in different habitats (riversides, deserts, stone outcrops, sandy coastal areas) and sometimes are noxious weeds, especially in the tropics with a humid climate. The species number is unevenly distributed around the World (Fig. 12). Four species occur in Australia (G. lotoides, G. oppositifolius, G. sessiliflorus and G. orygioides, endemic to N & C Australia: Short 2002Short , 2011, with the northern regions being the most speciesrich. The Americas are reported to have two species: G. lotoides, considered to be an alien from the Old World, and G. radiatus (Grayum and Koutnik 1982;Boetsch 2002;Vincent 2003;Vigosa Mercado 2015). However, the presence of G. lotoides in the New World is doubtful. Further taxonomic studies are needed to decide whether the carpological and chorological data support the acceptance of the second (presumably native) Glinus species in North America. The species number in Asia is two or three: poorly known Indian G. ononoides with only two old collections seen by us (G! K!), G. lotoides and G. oppositifolius (e.g., Backer 1951;Hedge and Lamond 1975;Heller and Heyn 1994;Lu and Hartmann 2003;Townsend 2016;Byalt and Korshunov 2020) with synonyms or insufficiently studied taxa described from Asia (see Taxonomic section below). Only one species (G. lotoides) is present in S Europe (Paradis 1993;Tutin et al. 1993) as well as in North Africa (Maire 1962;Boulos 1999). On the contrary, the Sub-Saharan area incorporates two to six species, depending on the regions (Fig. 12). The most species-rich region (5 spp.) is East Africa (Kenya and Tanzania). A large region with a tropical savanna climate is the second richest region with four species.

Possible mode of species dispersal
All Glinus species have hydrochastic capsules which open when triggered by rain drops (ombrohydrochory). This seems to be a somatic response of the plants to the climates characterized by alternating dry and wet periods. It is also known in many members of the Aizoaceae (Parolin 2006;Kurzweil and Burgoyne 2009) from areas with hot desert and semi-arid climates. In light of the presumable African origin of Glinus, such disseminative adaptation allows for a rapid seed dispersal during the rainy season.
In all species of the genus, the dispersal unit is a seed. The rains enable rapid dehiscence of the capsules and further dispersal of the seeds with surface water runoff. Additionally, the dry seeds due to their tiny size can easily roll over the substrate (Sulakshana and Raju 2018). However, it is unlikely that the seeds can move long distances in this manner, and, at least in G. lotoides, they remain viable only for several months (Bhatia 1987;Teshome and Feyissa 2015). We suggest that epizoochory may play a significant role in the dissemination, whereby moist substrate particles with seeds attached may inadvertently be carried by animals or humans. Figure 12. The number of Glinus species around the world. Areas coloured in mauve -one species, blue -two species, green -three species, yellow -four species, red -five species. Area boundaries are approximated. G. ononoides is a poorly known Indian taxon that needs to be studied further.

Taxonomic treatment of Glinus in Sub-Saharan Africa
The following treatment provides a new insight into the identification, taxonomic composition and distribution of Glinus in Sub-Saharan Africa where the genus is represented by 6 species. Description. Annual, rarely perennial, prostrate herbs or erect subshrubs with a rootstock, covered with stellate or simple (multiseriate, soft and crispate) hairs, in the latter case additionally with multiseriate, stout thick-walled and broad-based hairs (prickles). Stems branched from the base, often forming mats, rarely erect (G. orygioides). Leaves in false whorls, lanceolate to obovate, entire or denticulate (mostly in their upper half), with several lateral nerves that can be adaxially recessed and abaxially prominent. Flowers usually several to many (up to 20) in leaf axils forming loose or rarely dense inflorescences, bracteate, sessile or pedicellate. Perianth of 5 free oblong, ovate or roundish segments, green (brown) abaxially and white, pinkish or yellowish adaxially, with a green or brown midvein, horizontally spreading when fully opened, sometimes white petaloid staminodes present, always shorter than perianth segments. Stamens (2-4)5-30, outer stamen series corresponding to another staminode whorl (if stamen number is more than 5) often with filaments terminating with teeth (and without anthers), 0.3-1.8 mm, oblong or roundish; anthers yellow; pollen tricolpate (studied in G. lotoides: Perveen and Qaiser 2000). Stylodia 3-5, free or united in lower half into a style. Anthocarp (fruit enclosed by perianth) cylindrical or ovate to roundish. Fruit a hydrochastic loculicidal capsule. Ovary three-or five-carpellate, ovules arranged in two rows. Seeds usually more than 50, yellow, red, brown or black, up to 1.0 mm long, ovoid or reniform, smooth or with numerous colliculae; seed aril divided into two parts: a white, very noticeable hood covering the funiculus and a large ribbon-like appendage, sometimes the hood is reduced. Embryo curved; perisperm abundant and easily visible (in larger seeds) or scanty (in small seeds). The basic chromosome number in G. radiatus is x = 9 (Lane and Keil 1976), which corresponds with that of other Molluginaceae (Bogle 1970, with references therein). However, Mitra and Datta in Löve (1967) indicated the basic number x = 18 for the Indian populations of Glinus lotoides and G. oppositifolius.
Remarks. In the herbaria G. bainesii (Fig. 14) is often confused with G. oppositifolius, but it is differentiated from the latter species by having small prickles, larger perianth and anthers, and seeds with longitudinal ridges (Table 5).  General distribution. Endemic to Zambezi floristic province (according to Takhtajan 1986). Reported from Okavango region, NE Namibia (Friedrich 1966), but the cited specimen ("Lightfoot 65") has not been found by us (SAM?). Description. (Figs 16, 17). Annual, highly branched, with prostrate or ascending stems up to 100 cm long, covered with stellate (sometimes very scattered) and branched hairs, prickles absent. Leaves rosulate, short-lived, and cauline, green or grayish-green turning red at senescence, sparsely to moderately pubescent, rarely glabrous, petiolate (petioles up to 10.0 mm), entire or slightly crisp or scarcely denticulate, ovoid, obovate or oblong-spatulate, 10 Note. Mollugo hirta described from South Africa was very rarely accepted in old taxonomic literature, and only cited as a poorly known species . Otherwise, it has been commonly considered a synonym of G. lotoides (e.g., Fenzl 1840; Harvey and Sonder 1860;Just 1879;Adamson 1961;Hedge and Lamond 1975;Gonçalves 1978). We state for the first time that (1) Mollugo hirta must be resurrected in specific rank based on both molecular and morphological studies as G. hirtus, and (2) this species is conspecific with the name G. dahomensis. Specimens of this species were misidentified in collections with various names, particularly G. lotoides, G. lotoides var. virens, G. lotoides × G. oppositifolius, G. spergula, and Mollugo glinoides (both latter names belong to the synonymy of G. oppositifolius). Almeida (1998) referred the Indian plants to G. lotoides subsp. hirtus in the belief that they differ from the European populations (G. lotoides subsp. lotoides), but he did not indicate any differences between them. However, his description of the subspecies rather belongs to G. lotoides based on the number of stamens (>10) and stigmas (5), while G. hirtus has up to eight stamens and three stigmas.
The name G. dahomensis was originally introduced (Chevalier 1938) with a description in French, whereas the nomenclatural code required a description or diagnosis in Latin. To date the name G. dahomensis remains invalidly published because of having been commonly placed in the synonymy of G. lotoides var. virens. While considering G. hirtus distinct from G. lotoides, we resurrect the name published by Thunberg because it is the only one available for the species as circumscribed in our study.
Glinus hirtus is morphologically very close to G. ononoides Burm.f. described from India (Burman 1768), a species completely forgotten in the past (Dizionario… 1842) and present. The differences between both species are minor and limited to the following characters: (1) the type of G. ononoides has stellate hairs that are mostly localized on the young stem parts and leaves and additionally very short (up to 0.4 mm) prickle-like simple hairs present on the stem. In contrast, G. hirtus has well-expressed stellate pubescence on stem and perianth, and no prickle-like hairs, and (2) the seeds of G. ononoides are smooth, whereas those of G. hirtus are usually colliculate (however, the seeds of the specimens from Senegal, [without date] Perrottet; Botswana, 1982, P.A. Smith;and Somalia, 1893, D. Riva, are smooth). The collections of G. ononoides are very scarce, and the variability of the characters mentioned and the distribution of the species in Asia require further investigations. The morphologically very similar, but phylogenetically distant G. radiatus distributed in South and Central America has shorter anthers (0.35-0.6 mm long) and stylodia (0.2-0.6 mm long) as well as smooth seeds; by contrast, G. hirtus has longer anthers (0.7-0.9 mm) and stylodia (0.5-1.0 mm) and its seeds are usually colliculate (rarely smooth). Also, the distribution areas of both species do not overlap.
Habitat. River beds, wetlands, damp areas or as a weed, mostly on sandy soils at elevations 0-2400 m a.s.l. Sometimes, G. hirtus is found growing together with G. Distribution (Fig. 18). The species was originally known as Mollugo hirta from South Africa (Thunberg 1794) and not reported as a distinct species from any other African territory. Glinus dahomensis was described from Benin (Chevalier 1938), and later reported from Belgian Congo (DR Congo) (Hauman 1951). Hauman (1951) also noted the presence of the species in other territories of tropical Africa ("Du Dahomey au Transvaal"). However, he probably was not sure of that and did not reidentify the specimens from any other countries. We confirm that the range of G. hirtus is not restricted to Benin and DR Congo, but the species is distributed in all sub-Saharan Africa and seems to be a common weed in many regions according to the collector's observations. Glinus hirtus has not been previously reported for almost any Sub-Saharan countries (e.g., Adamson 1961;Jeffrey 1961;Berhaut 1967;Gonçalves 1970Gonçalves , 1978Sita and Moutsambote 1988;Barry and Celles 1991;Mapaura and Timberlake 2004;Phiri 2005;Setshogo 2005;Hassan et al. 2005;Sosef et al. 2006;Germishuizen et al. 2006;Darbyshire et al. 2015). In some West African checklists and manuals it was confused with the American G. radiatus (Berhaut 1967(Berhaut , 1979Boudet et al. 1986;Akoegninou et al. 2006;Lisowski 2009;Thiombiano et al. 2012;Brundu and Camarda 2013;Schmidt et al. 2017). Only a few specimens from Burkina Faso, DR Congo, and Nigeria were correctly identified as G. dahomensis.
Angola ( figure B). Superseded neotype (Jeffrey 1961: 15): [Italy] Sicily, Boccone (OXF, n.v.). Note. The protologue of Glinus lotoides L. is based on two main elements, an illustrated treatment of Alsine lotoides sicula from Sicily (Boccone 1674: 21, tab. 11, figure B) and another illustrated treatment of Portulaca baetica, luteo flore, spuria aquatica from Spain (Barrelier 1714a: 47, 1714b: figure 336). Linnaeus had not seen any specimen of the species prior to the publication of the protologue (Adamson 1961), in particular the collection of P. Boccone at OXF (Jarvis 2007). The herbarium collections of J. Barrelier are no longer extant (cf. Morton 1970) and had seemingly never been consulted by any botanist because all Barrelier's legacy but drawings was destroyed by fire after his death (Barrelier 1714a). There were two attempts to lectotypify the name Glinus lotoides. Jeffrey (1961) designated a specimen collected by Boccone and kept at OXF; although this specimen is associated with the illustration in Boccone (1674), it was not examined by Linnaeus and therefore is not part of the original material. Jeffrey's lectotypification is in effect neotypification. Adamson (1961)  [icon] Plate 356, figure 6 in Plukenet (1705). Note. The name G. dictamnoides was erroneously attributed by Fenzl to Linnaeus (1771); however, the latter clearly refers to Burman's "Flora Indica" where G. dictamnoides was described (Burman 1768). Burman (1768) cited no specimens in the protologue. Both Burman (1768) and Linnaeus (1771) cited Plukenet (1705) who depicted this plant in the Plate 356, figure 6. This image shows a hairy shoot with the rounded leaves in whorls and almost sessile verticillate flowers. The obovate vs. orbicular leaf shape was the main character known to Burman to distinguish G. lotoides and G. dictamnoides in situ (Burman 1768; see also a drawing of G. lotoides in the table 36, figure 1). The orbicular and greyish-green leaves of G. dictamnoides were similar to those of Dictamnus creticus Garsault (≡ Origanum dictamnus L.), and such plants were found in Madras [now Chennai], India: "dictamni cretici facie, maderaspatana". Based on the protologue of G. dictamnoides, Merrill (1921) synonymized this species name with G. lotoides.
In agreement with Merrill's opinion, we designate the cited illustration as lectotype and retain Burman's name in the synonymy of G. lotoides. The specimens seen from India usually have rounder leaves with ± scattered pubescence and shorter perianth (usually 5.5 mm long in fruiting) compared to the European populations, which have obovate and usually hirsute leaves and longer perianth reaching 6.5-8.5 mm in length. These characters were presumably the main argument to consider the Indian plants as G. lotoides subsp. hirtus (Almeida 1998) based on the name Mollugo hirta described from South Africa but erroneously applied to Indian plants (Clarke 1879). Mollugo hirta represents the plants with different characters (see also notes under G. hirtus). The African plants corresponding with G. dictamnoides are present in eastern and southern parts of the continent. It should be noted that the density of pubescence is very diverse in African plants, and those growing in a humid climate (e.g., Nigeria, Cameroon) usually have green leaves with scattered hairs. The intermediate forms in leaf shape and pubescence degree were frequently seen in the herbarium collections. In light of our molecular studies showing a mixed position of G. lotoides and G. dictamnoides, and scarce morphological differences between them, we prefer to synonymize G. dictamnoides with G. lotoides. = Tryphera prostrata Blume, Bijdr. Fl. Ned. Ind. 11: 549 (1826) In the type citation, the locality information is added from Drège (1843: 92) and the collection date is complemented from Gunn and Codd (1981). When establishing G. lotoides var. virens, Fenzl (1836) cited several validly published plant names as synonyms (G. dictamnoides, Pharnaceum pentagonum, Physa madagascariensis), and also illustrations and specimens. Since this new name was at the rank of variety and the synonyms were at the rank of species, and Fenzl provided his own description and used his own original material, his variety does not need to be treated as based on any of the synonyms included. Indeed, Adamson (1961) designated a separate type for this varietal name, a specimen collected by Drège in South Africa and cited by  in the protologue. This lectotypification is technically correct and should therefore stand.  established his variety for the plants that are less villose (or glabrescent at maturity) than the type variety of G. lotoides, with the perianth 5 mm long, and without petaloids. The distribution area of this new variety was circumscribed as East India, Timor, Arabia, Madagascar, and South Africa . This variety was accepted by the later authors, and its range was widened to include many regions of tropical Africa (Oliver 1871 sub Mollugo glinus var. virens;Adamson 1961;Gonçalves 1970Gonçalves , 1978Figueiredo and Smith 2008;Klaassen and Kwembeya 2013).
The plants with green or glabrescent leaves are more common in the wet climate (e.g., West Africa), and the densely pubescent populations are most frequent in drier conditions (Fig. 20).

Key to the varieties
Lectotype (Jeffrey 1961: 15)     Originally (Gonçalves 1965), the name G. oppositifolius var. glomeratus was invalidly published with a diagnosis in Latin based on two gatherings without explicit type designation. It was validated later (Gonçalves 1970) by indicating a "lectotype" and providing a full and direct reference to the validating diagnosis. This variety is characterized by the clusters with >10 flowers, and sessile or shortly (up to 4.0 mm) pedicellate flowers (Fig. 27). It may represent a montane variety of G. oppositifolius growing at higer altitudes (1000-2000 m), because one of the specimens cited by Gonçalves (1965) from Bié province, Angola, was collected at the elevation of 1360 m a.s.l. However, we did not include the samples of this variety into the molecular analysis, and further studies are needed to prove its varietal rank.
The plants of G. oppositifolius var. glomeratus look similar to G. hirtus. However, at least two characters distinguish it from the latter species: simple curved hairs vs. stellate pubescence, and shorter (0.3-0.6 mm long) stylodia (as in the type variety) vs. 0.5-1.0 mm long stylodia in G. hirtus. This variety was reported from Angola (Gonçalves 1970), but here we extend its distribution into DR Congo, Tanzania and Zambia.
Distribution (Fig. 26) Note. Jeffrey (1961: 16) indicated the "type" of Glinus setiflorus at C. He had not seen the material, and failed to distinguish between several sheets in Forsskål's collection. We designate one of these specimens as lectotype, which was seen and annotated by P.Ascherson in 1881. The locality information and collection date are complemented from Hepper and Friis (1994).
In the herbaria, G. zambesiacus was identified as G. lotoides, G. lotoides var. virens, G. lotoides × G. oppositifolius, and sometimes as G. bainesii. Glinus zambesiacus, G. oppositifolius, G. bainesii and G. hirtus are found in similar habitats. All character sets for each species are indicated in Table 5. The new species can be easily differentiated from both G. oppositifolius and G. bainesii by indumentum consisting of stellate trichomes, while G. hirtus has much shorter perianth and fewer (up to 8) stamens.  Note. The American Glinus radiatus was reported for Senegal and some other African countries (Central African Republic, Mali, Nigeria, and South Africa) as an alien species originating from tropical America (Berhaut 1979). It was later reported for Mali, Niger, Central African Republic, and repeatedly for Senegal by Boudet et al. (1986), for Benin by Akoegninou et al. (2006), for Guinea, Sierra Leone, Ghana, Ivory Coast by Lisowski (2009)   al. (2017). We did not see any specimens of true G. radiatus from the whole of Africa, and almost all specimens identified as G. radiatus belong to the morphologically similar G. hirtus. The differences between them are mentioned in the notes under G. hirtus. One of the specimens from Burkina Faso (Gnagna prov., Bogande, FR-0019665) identified as G. radiatus is indeed Zaleya pentandra (L.) C.Jeffrey (Aizoaceae). Herb. Forsskål 544 (C10002319). Note. The protologue (Forsskål 1775) was based on the only specimen indicated by Hepper and Friis (1994) as holotype. Hartmann (2001: 31) designated figure 14 in Forsskål (1776) as lectotype but this choice has no standing.

Glinus chrystallinus
The locality information and the collection date are complemented from Hepper and Friis (1994). Type material. n.v. Probably described from Mozambique.

Glinus mozambicensis
Note. No specimens were cited in the protologue (Sprengel 1825), and the description of G. mozambicensis is short and cannot be evaluated properly. According to Don (1834) and Endlicher and Fenzl (1839), it is a synonym of Gisekia pharnaceoides L. sensu lato [incl. G. africana Kuntze] (Gisekiaceae).  Note. Roth (1821) is an account of plants collected by Benjamin Heyne in India. The corresponding herbarium specimens were donated to Roth, whose collection was purchased by the Berlin Botanical Garden (B) and largely destroyed together with the holdings of that Herbarium in 1943 (Hiepko 1987). Heyne made no records on the exact provenance of his specimens, and "India Orientalis" (the Indian Peninsula) is the only information published and known.

Glinus trianthemoides
In the absence of any surviving original material, we used for neotypification one of the classical specimens assigned to the species by , who was the first to interpret the name.
Heyne (in Roth 1821) described G. trianthemoides as a plant with glabrous stems, obovate leaves with rounded tips, and lax, almost dichotomous inflorescences. Such characters are not found in any Glinus. Wight (Wight and Walker-Arnott 1834) accepted the placement of Roth's species in Glinus but doubted its taxonomic position in view of major morphological differences from the latter genus. Under the influence of that account, Fenzl (1836) established a separate genus for this taxon, Axonothechium Fenzl, still being uncertain about its precise taxonomic identity. Recent treatments (Gonçalves 1978;Sukhorukov and Kushunina 2015) placed Glinus trianthemoides into the synonymy of Corbichonia decumbens; this placement agrees with the material used by Fenzl and designated as neotype here.

Conclusions
Glinus is a monophyletic genus, presumably originating in tropical Africa, with predominant species diversity in Sub-Saharan Africa. Altogether, we accept six species for Sub-Saharan Africa, and none of them can be considered as locally endemic. Only G. bainesii and G. zambesiacus are restricted in their distribution to the southern and eastern parts of tropical Africa. A wide range of morphological characters can be used for the identification of Glinus species. In total, Glinus comprises 8-9 species (G. bainesii, G. hirtus, G. lotoides, G. oppositifolius, G. orygioides, G. radiatus, G. setiflorus, G. zambesiacus, and probably G. ononoides), and further research is needed to clarify the status of the American plants labelled as Glinus "lotoides".