Taxon sampling and composition of datasets

We used four datasets to complete the objectives of this investigation. The first dataset comprises major entities of Gomphrena (12 species) and allied genera (Lithophila, Gossypianthus, Guilleminea, Philoxerus) and also covers the other lineages of the gomphrenoid clade sensu Sánchez-del Pino & al. (2009) such as Xerosiphon, Froelichia, Hebanthe and Pfaffia. Pedersenia as a representative of the Alternantheroid clade served as outgroup. We selected 27 taxa to generate a dataset of plastid regions (rpl16 intron, matK gene and trnK intron as well as the trnL intron and the trnL-F spacer; dataset A) and a corresponding matrix of morphological characters. In most cases, molecular and morphological character data were obtained from the same individuals. For Froelichia, Hebanthe, Lithophila, Philoxerus, and Xerosiphon and some species of Gomphrena s.str. (G. boliviana, G. haenkeana, G. pulchella) the trnL-F and rpl16 data came from Sánchez-del Pino & al. (2009). In two exceptions, data from two closely related taxa were concatenated to represent the respective lineages: Pfaffia fruticulosa (matK-trnK, morphology, this study), and P. tuberosa (rpl16 and trnL-F from Sánchez-del Pino & al. 2009), as well as Gomphrena mandonii (matK-trnK, morphology, this study), and G. elegans (rpl16 and trnL-F from Sánchez-del Pino & al. 2009).

The second dataset includes a much higher number of samples (80 for plastid matK-trnK = dataset B-1 and 82 for nrITS = dataset B-2) with the aim to illuminate the overall tree space of the “Gomphrena clade” (= 45 taxa of Gomphrena s.str. including Gossypianthus, Lithophila, and Philoxerus, plus Guilleminea). Sampling was guided by morphological diversity, the sectional classification recognized so far (Holzhammer 1956) and the distribution of species in different biogeographical regions of the Americas and Australia. Since species limits in many cases are not yet well understood, plastid and nuclear sequences were obtained from the same individuals. In addition, some previously published matK-trnK sequences (e.g. Müller & Borsch 2005) were used for some species (Gomphrena ferruginea, G. fuscipellita, G. pulchella, Guilleminea densa, Philoxerus vermicularis). Voucher information and EMBL/GenBank accession numbers are provided in Appendix 1.

A third, extended matK-trnK dataset (dataset C) of the Amaranthaceae–Chenopodiaceae alliance was used for molecular clock dating in order to accommodate fossil calibration points. The sequence matrix employed the same representatives as in Di Vincenzo & al. (2018) for Chenopodiaceae, other Caryophyllales and eudicot lineages. For the Amaranthaceae, the representation of the achyranthoids was reduced here whereas Gomphrenoideae were sampled as in dataset B-1 of this investigation with some terminals belonging to the same species not included.

A fourth dataset (D-1 for matK-trnK and D-2 for ITS) was created by extending datasets B to include the further C₄ genera of Gomphrenoideae (species of Alternanthera and Tidestromia) into the ancestral character reconstruction of photosynthesis types. The sampling followed Borsch & al. (2018), from which most of the sequences were taken, with additions in Alternanthera from Sage & al. (2007) and Sánchez-del Pino & al. (2012). The ITS sequence of Pseudoplantago was newly generated. Chamissoa was used to root the trees as it is an early branching lineage (Müller & Borsch 2005) with Ptilotus and Pandiaka representing the aervoid and achyranthoid clades of the Amaranthoideae, respectively. However, the greater distance in particular of the sequences from Alternanthera and Tidestromia led to unalignability in two hotspots (which therefore needed to be excluded); thus this dataset is inferior for calculating precise relationships within Gomphrena as it has fewer characters.

DNA isolation and sequencing

Genomic DNA was isolated from silica-gel-dried leaf tissue or herbarium specimens using a triple CTAB extraction method (Borsch & al. 2003). The matK-trnK, trnL-F and rpl16 regions were selected because of their high phylogenetic structure (Borsch & Quandt 2009; Korotkova & al. 2011) and to achieve consistency with other Amaranthaceae datasets (Müller & Borsch 2005; Di Vincenzo & al. 2018; Borsch & al. 2018). Primers were used as in Borsch & al. (2018) to amplify two overlapping halves of the matK-trnK region, whereas DNAs isolated from herbarium specimens often required to amplify even shorter fragments. Also the additional Gomphrena-specific primers ACmatK442F (5′-AGTCAAAAGAGCGATTGGG-3′), ACmatK602F (5′CTTGTTTTGACTGTATCGC-3′), ACmatK465R (5′-TCTTATAACAAAATAAGATGG-3′) and ACmatK631R (5′-ACAAAAGTAAAAATAGAGG-3′) designed here were used Primer ACmatK442F served as forward sequencing primer to complement the pherograms made with ACmatK1400R that could not read over a large microsatellite located in the trnK intron in several species of Gomphrena about 90 nt upstream of the matK start co-don. The rpl16 intron was amplified with flanking primers rpl16-1216F and rpl16-18R that were also used for sequencing along with the additional internal forward primer GOMrpl16-495F (Borsch & al. 2018). The trnL-F region was either amplified as a whole using primers trnTc and trnTf (Taberlet & al. 1991) or in two parts when DNA was isolated from herbarium material with primers trnTc and trnTd as well as trnL460F (Worberg & al. 2007) and trnTf. Primers trnTd and trnL460F were used for sequencing.

The ITS region was amplified and sequenced with the universal primers ITS4 and ITS5 (White & al. 1990). For some difficult samples from herbarium specimens, the new internal primers ACITS3F (5′-TTGGTGTGAATT GCAGAATCCC-3′) and AC-ITS2R (5′-GATGGTTCA CGGGATTCTGC-3′) were used to amplify and sequence the ITS region in two halves.

The PCR conditions were as described in Di Vincenzo & al. (2018). Sanger sequencing was performed by Macrogen Inc. (Seoul, South Korea) on an ABI 3730 XL capillary sequencer. Both strands were sequenced for ITS.

The sequences were submitted to ENA in bulk as annotated multiple sequence alignments by means of the program annonex2embl (Gruenstaeudl 2020).

Alignment and coding of length mutational events

Sequences were edited and aligned manually using PhyDE (Phylogenetic Data Editor) v. 0.9971 (Müller & al. 2012), using a motif alignment approach (Morrison 2009; Ochoterena 2009) following the rules by Löhne & Borsch (2005). Positions of uncertain homology (mutational hotspots) were excluded from the analysis (for exact delimitation see  Appendices S1 (wi.50.50301_Appendix_S1_matK-trnK_rpl16_trnL-F_dataset_A_alignment_and_matrix.nex) – S6 (wi.50.50301_Appendix_S6_ITS_dataset_D-2_alignment_and_matrix.nex) ). Indels were coded using the Simple Indel Coding method (Simmons & Ochoterena, 2000) as implemented in SeqState 1.40 (Müller 2005a).

Phylogenetic analyses

Maximum Parsimony (MP) analyses were performed using the Parsimony Ratchet (Nixon 1999) implemented in the software PRAP (Müller 2004) in combination with PAUP v.4.0b10 (Swofford 1998). Settings were 200 ratchet iterations with 25% of the positions randomly up-weighted (weight = 2) during each replicate and 10 random addition cycles. The command files generated with PRAP were then run in PAUP, using the heuristic search with the following parameters: all characters have equal weight, gaps are treated as “missing”, TBR branch swapping, initial swapping on 1 tree already in memory, Maxtrees set to 100 (auto increased by 100) and branches collapsed actively if branch length is zero. Jackknife (JK) support for nodes was also performed in PAUP with 10,000 replicates, using a TBR branch swapping algorithm with 36.788% of characters deleted and one tree held during each replicate, following Müller (2005b).

The substitution models for the individual data partitions were determined with jModelTest 2 (Darriba & al. 2012) and using the Akaike information criterion (Akaike 1974). The best-fitting model was TVM+Γ for trnL-F and rpl16 and GTR+Γ for matK-trnK (datasets A and B-1). The best-fitting substitution model found for ITS was SYM+Γ.

Bayesian inference (BI) was carried out using MrBayes v. 3.2 (Ronquist & al. 2012) using the specifications from the best fitting models. A sampling frequency of 1000 was applied with the first 25% discarded as burn-in, four independent runs were performed with 4 chains each and 10 million MCMC generations. The convergence and effective sample size (ESS) of each replicate were checked using Tracer v. 1.5.0 (Rambaut & al. 2013). The remaining population of trees was summarized as majority rule consensus tree also using MrBayes 3.2.

Maximum Likelihood (ML) analysis was performed using RAxML GUI 1.3 v. 7.2.8 with 1000 bootstrap replicates Stamatakis (2006). Searches were performed using the general time-reversible (GTR) model with among-site rate heterogeneity modelled by a GAMMA distribution with 25 rate categories based on the available model choice in RAxML.

Assessment of morphological characters and ancestral character state reconstruction

A set of 21 morphological characters was defined covering vegetative, inflorescence and floral morphology. Therefore, characters used in previous studies of Amaranthaceae (e.g. Holzhammer 1955; Pedersen 1976, 1990, 1999; Eliasson 1988; Borsch & Pedersen 1997) were re-analysed and also new features and compared with the observed variation in the plant specimens in order to arrive at clear definitions of characters and their states reflecting hypotheses of homology (see Appendix 2). Also new characters on vegetative and inflorescences morphology were assessed. The selection of characters also considered their previous use as diagnostic features for the various generic concepts that were applied to Gomphrena and allies. Since the purpose of this investigation was not species delimitation, we did not include the usually much more variable characters with often many states (e.g. shape or texture of sepals) or quantitative data (e.g. length of sepals). These will be dealt with in future studies.

The 21 morphological characters were scored for the same species as in the 27-taxon combined plastid dataset based on herbarium specimens corresponding to the samples used for molecular analyses. To depict the ancestral states we used the maximum clade credibility tree from the Bayesian analysis, which was identical in topology to the Bayesian Majority rule consensus and the best ML trees. Ancestral states were reconstructed using BayesTraits, v. 2.0 (Pagel & al. 2004) sampling 1000 randomly selected trees after the burn-in. Commands for BayesTraits were generated by TreeGraph2 (Stöver & Müller 2010). Ancestral state probabilities were imported into TreeGraph2 to simultaneously visualize them on the branches.

Molecular clock dating

Dating was carried out with BEAST v. 1.8.0 (Drummond & al. 2012) using the broad Amaranthaceae–Chenopodiaceae matK-trnK dataset C. Because no fossils for the Gomphrenoideae are known, the same three fossil calibration points used by Di Vincenzo & al. (2018) in Chenopodiaceae were also employed here. A maximum age of 125 Ma was assumed which corresponds to the most likely age of the crown group of core eudicots (Bell & al. 2010). Age distribution priors for fossil primary calibration points were set as “exponential” (Ho & Philips 2009), whereas the three secondary calibration points (using age estimates of Bell & al. 2010) were included with “normal” age distributions equal to the 95% highest posterior densities interval (HPD) of Bell & al. (2010). Thus, our priors and calibration points for the dating of the gomphrenoid clade in Amaranthaceae were equal to the dating of Amaranthaceae with a focus on the achyranthoid clade as recently carried out by Di Vincenzo & al. (2018). A birth-death model was employed to model lineage diversification, using a random starting tree. Trees were sampled every 1000^th generation after a burn in of 50%, calculating a total of 50 million generations for two MCMC runs. Adequate parameter sampling was checked with Tracer v. 1.4.0 (Rambaut & Drummond 2007). The combined post-burn-in tree distribution of both runs was then summarized as a maximum clade credibility tree using TreeAnnotator v. 1.8.0 (Drummond & al. 2012).

Evolution of C₄ photosynthesis

To assess the type of the photosynthetic pathway, δC13 values were taken from Sage & al. (2007). Many of the specimens studied there are also included into our phylogenetic analysis. Some additional specimens (out of the mostly Andean clade of Gomphrena) were examined for Kranz anatomy, which all were close relatives of taxa previously studied for isotopes. Presence of C₄ versus C₃ was then coded as a binary character. Ancestral states were reconstructed as reported for the other morphological characters. An analysis of ancestral states was implemented both for the plastid and the ITS trees to account for the incongruence in the position of the G. mollis–G. rupestris clade using datasets D-1 and D-2.

Generation of the distribution map

The distribution map of the coastal species (Philoxerus portulacoides and P. vermicularis was produced in ArcGIS 10.3, based on the data obtained from the label of herbarium specimens reviewed in the National Herbarium of Bolivia (LPB), the Instituto Darwin-ion (SI) in Buenos Aires, and Berlin (B). In addition, the data entries from the online registers of TropicosMOBOT (MO), Jardim Botânico do Rio de Janeiro (JABOT) specimens from Brazil; and the African Plant Database ( https://www.ville-ge.ch/musinfo/bd/cjb/africa/recherche.php) were considered. The georeferenced data was used in “DECIMAL DEGREES” format and all coordinates were verified with Google Earth. To assess the distribution in Australia occurrence data were obtained by querying the GBIF Portal (GBIF 2016).

Characteristics of the sequence datasets

Statistics of the multiple sequence alignments for the 27 samples of trnL-F, rpl16, matK-trnK, the individual and consensus information are in Table 1. The length of the alignment of 27 samples of the regions trnL-F, rpl16, matK-trnK (dataset A) had 1082, 1150, 2499 characters, respectively, with eight hotspots (HS) of unclear homology excluded (most being poly A/T stretches). The number of parsimony informative characters were 91 in trnL-F, 118 in rpl16, and 178 in matK-trnK. The alignment of matK-trnK in the larger dataset B-1 had 2510 characters including 45 indels, excluding three HS. The ITS alignment (dataset B-2) had 722 characters including 81 indels, of which 269 were parsimony informative (See Table 1).

Molecular phylogenetic trees

The combined plastid tree (matK-trnK, trnL-F, rpl16; Fig. 1) depicts a basal split into a clade of Pfaffia and allies on one hand and Gomphrena s.str. and relatives on the other. Almost all nodes receive maximum support under MP, BI and ML. Several lineages were identified within a major clade that is here called “Gomphrena clade”: the first two branches are a Gomphrena prostrata–Guilleminea clade and a Gomphrena mollis–G. rupestris clade whereas all other lineages form the “core C₄ Gomphrena clade” (Fig. 1–3). Within the latter, a clade comprising G. meyeniana sister to G. radiata plus G. tomentosa, Gossypianthus and Gomphrena boliviana; a clade of Lithophila, Philoxerus and G. flaccida from Australia; a clade of G. haenkeana and G. celosioides; and a clade with G. agrestis, G. lanigera and G. pulchella were found. Xerosiophon and Froelichia are successive sisters to the “Gomphrena clade”.

The trees inferred from the extended matK-trnK dataset B-1 (Fig. 2) show the same principal lineages but several of them are revealed with more diverse crown groups. These are a central and southern Andean dry Puna and Prepuna Gomphrena radiata–G. umbellata–G. tomentosa clade, and a G. boliviana–G. martiana clade. All species sampled from Australia are resolved together with G. flaccida in a Philoxerus+Australian Gomphrena clade. The clade with G. agrestis and G. lanigera from the Cerrado of Bolivia and Brazil, G. pulchella ranging from the Cerrado to the Chaco is extended by G. cardenasii, an endemic from the Cerrado-Chiquitania of Bolivia and thus comprises species from the lowlands of southeastern South America. The most noteworthy result is that G. haenkeana represents a species-rich Andean clade with two well supported subclades (A and B), whereas the Mexican G. nitida appears unresolved to them (Fig. 2).

Table 1.

Sequence statistics. Taxon datasets: A: 27 taxa combined matK-trnK+rpl16+trnL-F; B: 80 sequences of matK-trnK of Gomphrena and relatives; C: 82 sequenes of ITS of Gomphrena and relatives; D: matK-trnK of Amaranthaceae–Chenopodiaceae with representation of Gomphrena for molecular-clock dating.

The ITS data (Fig. 3) recover the same principal clades and also largely the same topology. Nevertheless, the position of the Gomphrena mollis–G. rupestris clade is in-congruent in the ITS trees, where it diverges before the G. prostrata–Guilleminea clade. The Australian species of Gomphrena clade are also retrieved with high support as close relatives to Philoxerus and Lithophila (Fig. 3).

Morphological characters

Six characters describe the habit and vegetative morphology in Gomphrena and allies as defined in Appendix 2. Character 1 (life cycle) is a complex trait, which is relevant because of many species being annuals. Characters 7 to 10 relate to inflorescences. The paracladia in Gomphrena (Fig. 4) and allies are complex structures that resemble different levels of expansion of a complex synflorescence architecture, including the reduction of internodes, and a specific whorl-like arrangement of the paracladia (character 8, state 3; Fig. 4). In particular, character 9 (apical leaves subtending paracladia) is grouped into different states that resemble different forms of transition of cauline-like to very specialized pseudanthial leaves (state 2). Characters 12 and 13 refer to floral morphology, whereas characters 14 to 21 describe the variation of the androecium in detail (Appendix 2). The androecia of all 27 species included in dataset A are also illustrated schematically in Fig. 6. The matrix of morphological characters and their states is found in  Appendix S7 (wi.50.50301_Appendix_S7_matrix_morphological_characters_1-21.xlsx) .

Evolution of morphological characters

Ancestral state probabilities for vegetative characters are depicted as pie charts (chars 1–6, 8, 9, 12, 13 are illustrated in Fig. 5; characters 7 and 10 see  Appendix S8 (wi.50.50301_Appendix_S8_ancestral_states_morphological_character_7.pdf) ), Note that heteromorphic sepals with the inner two smaller than the outer and strongly compressed in fruit (character 12, state 1 in grey) is a synapomorphy for the core C₄ Gomphrena clade. Pseudanthia (in Fig. 5 character 8, state 3 in blue for the stellately arranged paracladia; character 9 state 2 in green for the specialized leaves) is derived three times, in G. meyeniana, G. boliviana and G. haenkeana. The evolution of the androecium (characters 14, 16, 17, 19; for chars 15, 18, 20, 21 see  Appendix S9 (wi.50.50301_Appendix_S9_ancestral_states_morphological_character_15.pdf) ) is shown in Fig. 6. The size of the androecial tube varies considerably. Long androecial tubes appear in two lineages and are associated with showy, often colourful inflorescences. Staminal tube appendages alternating with the filaments only appear in Pedersenia (alternantheroids) and Froelichia but otherwise are completely lacking in the gomphrenoid clade (Fig. 6).

Diversification of Gomphrena and allies in time

The age of the crown group corresponding to the monophyletic subfamily Gomphrenoideae was inferred to be 30.9 Ma (Fig. 7; 21.16–43.22 95% HPD, node number 14 in  Appendix S10 (wi.50.50301_Appendix_S10_BEAST-tree_node_numbers.pdf) ), whereas the Gomphrena clade (see Fig. 1, 2) (15.1 Ma, 8.92–23.6 95% HPD, node 21) and the less inclusive core C₄ Gomphrena clade (node 23; 11.4 Ma, 6.69–18.78 95% HPD) are much younger. The Australian subclade has a stem age of 10 Ma (5.7–16.9 95% HPD) and thus started to diversify at a similar time to the speciose Andean subclades A and B (Fig. 1, 3, 7).

Evolution of C₄ photosynthesis

The reconstruction of ancestral states using the Bayesian trees inferred from the extended datasets D-1 (for matK-trnK) and D-2 (for ITS) is depicted for all nodes in  Appendices S11 (wi.50.50301_Appendix_S11_matK-trnK_extended_MCCtree_MCMC_character_11.pdf) and  S12 (wi.50.50301_Appendix_S12_ITS_extended_MCC-tree_MCMC_character_11.pdf) , respectively. The plastid trees show C₄ photosynthesis to have originated in the common ancestor of the Gomphrena clade plus Froelichia, and two more times independently in a sublineage of Alternanthera and in Tidestromia (Fig. 7). The plastid topology also suggests a reversal from C₄ to C₃ (Fig. 7). To the contrary, the nuclear ITS tree, that resolves the G. mollis–G. rupestris clade in a different position not as second but as first branch of the Gomphrena clade (Fig. 3), suggests a single gain of C₄ photosynthesis in Gomphrena and allies ( Appendix S12 (wi.50.50301_Appendix_S12_ITS_extended_MCC-tree_MCMC_character_11.pdf) ).

Phylogeny of the Gomphrena clade and relationships of Lithophila, Gossypianthus, Guilleminea and Philoxerus

All plastid trees converge on a deep split into a lineage comprising Hebanthe, Pfaffia and allies (the Pfaffia clade) and a lineage including Gomphrena and allies, with Xerosiphon and Froelichia branching as successive sisters to the remaining taxa (Fig. 1, 2). The deep nodes received maximum support under MP, BI, and ML in matK-trnK alone (Fig. 2) as well as plastid regions combined (Fig. 1). The same topology for the principle lineages of the gomphrenoids as one of the tree major clades of the monophyletic subfamily Gomphrenoideae (in addition to alternantheroids and iresinoids) was found by Sánchez-del Pino & al. (2009) based on trnLF and rpl16 sequences alone. The trees recovered from nrITS show the same branching order of Froelichia and Xerosiphon (Fig. 3) although support of the respective nodes is less. The clade that includes the majority of Gomphrena species in addition to Xerosiphon and Froelichia depicts Guilleminea, Gossypianthus, Philoxerus and Lithophila nested among other species of Gomphrena (Fig. 1–3). This major clade is characterized by metareticulate pollen with the tectum reduced to distal bands (see also Borsch & al. 2018 for a detailed analysis of pollen evolution where this clade is represented by Froelichia and Gomphrena lanigera). The clade of Gomphrena including Guilleminea, Gossypianthus, Lithophila and Philoxerus is here annotated as the “Gomphrena clade” (Fig. 1–3) and receives maximum support in all analyses both from plastid and nrITS.

The taxon sampling of the Gomphrena clade in this investigation is several times higher and now includes a representative sampling of Gomphrena compared to the only 13 species in Sánchez-del Pino & al. (2009). Our trees inferred from plastid and nrITS sequence data congruently reveal ten lineages within the Gomphrena clade (see Fig. 1–3). The earliest diverging lineages are a Gomphrena prostrata–Guilleminea clade and a Gomphrena mollis–G. rupestris clade. These are inferred as successive sisters from the plastid data, although the combined analysis of trnK-matK+rpl16+trnL-F sequence data shows only moderate support for the second-branching position of the G. mollis–G. rupestris clade under MP (71% JK) and ML (89% BS; Fig. 1). The signal for this topology of the trnK-matK partition alone (Fig. 2) is much stronger. The nrITS partition depicts the G. mollis–G. rupestris clade in a switched position, branching before the G. prostrata–Guilleminea clade. Further analysis with a spectrum of loci from nuclear and organellar genomic partitions will be needed to test if this is hard incongruence, being the result of a reticulate evolutionary event.

Fig. 1.

Cladogram reconstructed from the combined matrix of matK-trnK+rpl16+trnL-F sequence data including indels (dataset A). The majority rule consensus tree obtained from MrBayes is depicted. Support values shown above branches are posterior probabilities; below are JK percentages from parsimony (left) and BS percentages from likelihood reconstruction (right).

The relationship of Guilleminea and Gomphrena prostrata receives maximum support in all plastid trees, whereas nrITS is less conclusive (0.89 PP, 80% JK, 69% BS). Guilleminea densa has a procumbent habit and is characterized perigynous flowers in terminal paracladia, with the androecium united in the sepals, and the five sepals of similar size, like the other species in the genus (Eliasson 1988, Pedersen 1990). Whereas most of the six species occur in the Chaco, Puna, and Andean dry valleys of Bolivia, Peru, Argentina and Paraguay (365 to 5000 m), G. densa ranges throughout the neotropics to the southern United States, and can be found as an introduced plant in Africa. Gomphrena prostrata has procumbent to ascending stems, flowers in terminal paracladia, a largely fused androecial tube, the anthers attached between two small filament appendages, and five sepals of similar size, being densely pubescent in the dorsal part. The plant occurs in Caatinga and Cerrado habitats of Brazil, where it is adapted to sandy soils. Gomphrena mollis and G. rupestris have detrital trichomes in the leaves and stems, while the rest of species in the Gomphrena clade have uniseriate and linear trichomes. The genera Philoxerus and Lithophila are deeply nested within the Gomphrena clade and appear closely related to the Australian species of Gomphrena in a well-supported clade in all analyses (Fig. 1–3).

Evolution of vegetative morphology in Gomphrena and allied genera

Vegetative characters tend to be homoplastic rather than exhibiting synapomorphies for major clades (Fig. 5). The annual life cycle (character 1, state 0) thereby was derived multiple times from perennial ancestors and is depicted as an independent transition in four lineages of the core C₄ Gomphrena clade, so in G. haenkeana, G. flaccida, G. radiata, G. boliviana, but also in Froelichia floridana. The annual life form therefore is mostly associated with plants occurring in dry (G. flaccida, G. radiata) or seasonally dry environments (G. haenkeana) indicating the adaptive nature with plants only appearing in the wet season. Multiple origins of annuals have also been observed in other lineages of angiosperms radiating in areas with specific geographical patterns of dry environments such as Nemesia Vent. (Scrophulariaceae; Datson & al. 2008). The evolution of root types shows a similar homoplastic adaptive pattern, linked to different survival strategies in the different environments where species grow. For example, G. meyeniana is distributed along the Andes (1890–4800 m) and presents taproots that represent 82% of its total biomass (Patty & al. 2010), whereas aerial parts can be largely lost during the dry season. These features occur in many unrelated plant lineages and are an adaptation to high mountain ecosystems (Körner 2003). In a similar way, taproots of Guilleminea or G. tomentosa and relatives are tuberose as a drought adaptation strategy. Whereas taproots originated in the ancestor of the Gomphrena clade, this state got lost again with shifts to annual life forms (e.g. G. radiata). The taproots of Cerrado species like G. pulchella or G. lanigera have been observed to store of fructanose (Fank-de-Carvalho & al. 2015), which is a further response to the environmental stress of the dry season and to fires where the soil temperature can raise to > 70°C. The adventitious roots of Philoxerus (character 2, state 2) evolved independently, and allow the plants to spread in the wet sand of coastal habitats. Roots of this species show tolerance to salinity (Bove 2011), and Cordazzo and Seeliger (2003) have proven the regeneration capacity of P. portulacoides from fragments of the adventitious root.

Fig. 2.

Extended phylogeny of Gomphrena based on matK-trnK sequence data including indels (dataset B-1). The Bayesian topology is shown as phylogram. Support values shown above branches are posterior probabilities; below are JK percentages from parsimony (left) and BS percentages from likelihood reconstruction (right). Major lineages are annotated as well as the geographic origin (country, department) of the respective samples. Note that there is a diverse clade that is mostly constituted by Andean species.

Basal leaves have evolved independently three times in different lineages of the Gomphrena clade. We coded two different states regarding to their different morphology in plants with prostrate versus erect stems. Gomphrena tomentosa and relatives as well as Guilleminea have prostrate or decumbent stems, but a few leaves develop from the reduced main axis. On the other hand, species such as G. lanigera, G. agrestis and G. flaccida have erect stems which are hardly branched, but develop a radical rosette of long-lived leathery leaves, often with a dense indumentum.

Ancestral character state reconstruction shows that procumbent and decumbent stems are derived from erect ones, and evolved two times independently within the core C₄ Gomphrena clade (in the Lithophila+Philoxerus clade; and in the ancestor of G. tomentosa, G. meyeniana plus Gossypianthus) and a third time in the ancestor of Gomphrena prostrata and Guilleminea (Fig. 5). In the latter two clades plants grow in dry and hot environments where procumbent and decumbent stems protect them from wind, and reduce evapotranspiration. This corresponds to the evolution of radical leaves. In Guilleminea they are small and lost when the plants grow (Mears 1967), whereas in Gossypianthus they are persistent and more prominent, a similar quality also present in Gomphrena tomentosa, G. radiata and G. meyeniana which all belong to the same clade (Fig. 2, 3, 5).

Fig. 3.

Extended phylogeny of Gomphrena based on nrITS including indels (dataset B-2). The Bayesian topology is shown as phylogram. Support values shown above branches are posterior probabilities; below are JK percentages from parsimony (left) and BS percentages from likelihood reconstruction (right). Major lineages are annotated as well as the geographic origin (country, department) of the respective samples.

Complex branched main stems are the rule in woody species (character 5, state 1) such as Hebanthe, Pedersenia or Pfaffia fruticosa and are ancestral in Gomphrenoideae (Fig. 5). Nevertheless, stems with secondary and tertiary branches also occur in perennial or annual herbaceous plants. In the latter this is a strategy to produce numerous axes that can quickly develop flowers and thus foster reproduction. Those species that only have unbranched stems generally possess a paracladium or at or near the terminal or node of the stem. This feature (state 0) has evolved independently several times (in Xerosiphon aphyllus, Froelichia floridana, Gomphrena flaccida, G. mollis, G. agrestis and G. meyeniana). Among the species of Gomphrena, G. tomentosa presents indeterminate growth, and as well Gossypianthus and Guilleminea. According to our ancestral character state reconstruction, this feature (character 6, state 0) in only a few cases, where it seems to be associated with plants possessing a tap root and procumbent stems.

Evolution of inflorescence and floral morphology, especially the androecium in Gomphrena and allied genera

The ancestral inflorescence in Gomphrenoideae was a globose to subglobose paracladium with densely adjacent solitary flowers (character 7, state 0) appearing terminal on a more or less elongated axis, without any specialized subtending leaf organs (character 9, state 1; Fig. 5). The head-like shape was maintained in the Gomphrena clade whereas elongate to cylindric paracladia with remote flowers were independently derived in Hebanthe and Froelichia (not illustrated). The arrangement of paracladia into more complex synflorescences (character 8; partial florescences) is variable in Gomphrenoideae (Acosta & al. 2009) and also considerably differs in Gomphrena and allies depending on the branching system of more or less complex synflorescences. Because no detailed insights on the development or anatomy of inflorescences in Amaranthaceae are available that would support further hypotheses on homology, we coded the different arrangements of paracladia as unordered multiple states (character 8 with 6 states). Whereas all states reach nearly equal probabilities to be ancestral at deeper nodes (Fig. 5), it can be shown that the very specific architecture of a terminal florescence surrounded by usually 5 paracladia in a whorl-like structure (character 8, state 3) evolved two times independently in G. boliviana and G. haenkeana (Fig. 5). In fact, all relatives of G. boliviana (the G. boliviana–G. martiana clade) and of G. haenkeana in the two Andean subclades A and B (see Fig. 2, 3) possess the same states, so that the two species depicted in Fig. 5 are good representatives for their respective clades with a whorl-like arrangement of paracladia. It will be interesting to study the ontogeny of this structure to determine in how far it is the result of a strong condensation of a thyrsoid branching system or also involves collateral duplication of buds developing into paracladia.

Fig. 4.

Morphological diversity in flowers and inflorescences of Gomphrena and allies. – A: inflorescence of Gomphrena spec. 7 with less prominently visible paracladia but bright yellow pseudanthial leaves (Ortuño & al. 1677); B: Gomphrena meyeniana with pseudanthial leaves from lower surface (Borsch & Ortuño 3561); C: inflorescence of G. fuscipellita with six paracladia supported by broadly rounded, green pseudanthial leaves (Borsch & Ortuño 3594); D: Gomphrena haenkeana showing long, pink sepals and androecial tubes with reflexed pale yellow filament appendages, whereas stamens are dark yellow (Borsch & Ortuño 3627); E: Philoxerus vermicularis without pseudanthial leaves in the terminal inflorescence and two opposite leave-like ones in the lateral inflorescence, and a pair of succulent cauline leaves (Borsch & al. 3444); F: Lithophila muscoides, part of a cushion-like plant in a limestone crevice (Greuter & al. 26915); G: Terminal part of flowering stem of Guilleminea densa (Borsch & al. 3437). – Photo credits: A: T. Ortuño; B–G: Th. Borsch.

Fig. 5.

Evolution of vegetative (characters 1 to 6), inflorescence (characters 8 and 9) and floral morphology (characters 12 and 13) in Gomphrena and relatives (see Appendix 2 for the definition of characters and their states as well as the full matrix in  Appendix S7 (wi.50.50301_Appendix_S7_matrix_morphological_characters_1-21.xlsx) ). The pie charts represent probabilities for ancestral states as reconstructed with BayesTraits and depicted on the maximum clade credibility tree obtained by Bayesian Inference of the combined plastid dataset. Pie charts were omitted for some deep nodes with minor changes and for characters 7 and 10 where changes appeared not to be relevant for Froelichia, Hebanthe, Pedersenia and Xerosiphon, all of which are well-defined distant genera (see  Appendix S8 (wi.50.50301_Appendix_S8_ancestral_states_morphological_character_7.pdf) ).

This feature is correlated with the presence of pseudanthial leaves that are arranged in a whorl (character 9, state 2), earlier noted by Holzhammer (1956) and Siqueira (1992) as “involucral leaves”. Whereas two opposite, small subtending cauline leaves (state 1) is reconstructed as the ancestral state in the core C₄ Gomphrena clade, whorl-like pseudanthial leaves are derived even three times. Once in the speciose clade of G. haenkeana and relatives to which also G. fuscipellita and G. pallida belong, in the G. boliviana–G. martiana clade, and a third time in G. meyeniana that grows with a short upright stem at high elevations in the Andes (Fig. 4, 5). In the case of G. meyeniana pseudanthial leaves surround a single main florescence with relatively large white flowers appearing in good contrast to the dark pseudanthial leaves. Some populations of G. pallida have even been found to possess yellow-coloured pseudanthial leaves (Fig. 4). Conspicuously coloured leaves also serve as optical attractants in inflorescences of other Caryophyllales lineages such as Bougainvillea Comm. ex Juss. (Nyctaginaceae; Brockington & al. 2009). Gomphrena pulchella represents a further specialization (autapomorphy, character 9, state 3), where an increased number of up to 10 leaves subtend a terminal solitary paracladium. The flowers in this species are the largest in Gomphrenoideae (>45 mm in length), and also androecial tube is very long (Fig. 6), leading to very conspicuous flower heads. These specialized leaves could also serve as a protection against fire before anthesis and during fruit development in G. pulchella which is a typical Chaco species (Pedersen 1976). This pattern of inflorescence evolution indicates that there is a trend toward the formation of pseudanthia in the core C₄ Gomphrena clade. Compared to several other families of flowering plants such as Apiaceae, Asteraceae, Euphor-biaceae (their cyathium), or Rubiaceae, the formation of pseudanthia in Amaranthaceae so far did not receive much attention. Classen-Bockhoff (1990) just mentioned their existence in Gomphrena and Ptilotus.

Gomphrena and allies as all Caryophyllales possess five perianth parts with opposite stamens, located in individual primordia in quincuncial order independently of the stamens (Vrijdaghs & al. 2014). Their arrangement results in two external, one intermediate, and two inner sepals, the latter of which can vary strongly in size and texture (Character 12). The state with the inner two sepals smaller than the outer two, which are cymbiform, compressed and carinate in fruit is resolved as synapomorphy for the core C₄ Gomphrena clade (Fig. 5). The G. mollis–G. rupestris clade, which also exhibits smaller inner sepals differs in that these inner sepals do not become thickened at maturity (state 2), which could represent a transitional state if the G. mollis–G. rupestris clade is assumed to be the sister of the core C₄ Gomphrena clade. Further analyses using reductive coding of this complex character (Torres-Montúfar & al. 2018) will be interesting but will require detailed anatomical analyses of the tissues. In Guilleminea perigynous flowers are derived as the only genus in Gomphrenoideae (character 13, state 0).

Fig. 6.

Evolution of androecium characters 14 (length of androecial tube), 16 (shape of androecial tube), 17 (presence of androecial tube appendages), 19 (presence of filament appendages) in Gomphrena and related genera (see Appendix 2 for the definition of characters and their states as well as the full matrix in  Appendix S7 (wi.50.50301_Appendix_S7_matrix_morphological_characters_1-21.xlsx) ). The pie charts represent probabilities for ancestral states as reconstructed with BayesTraits and depicted on the maximum clade credibility tree obtained by Bayesian Inference of the combined plastid dataset. Ancestral states of characters 15, 18, 20 and 21 are shown in  Appendix S9 (wi.50.50301_Appendix_S9_ancestral_states_morphological_character_21.pdf) ).

Differences in the morphology of the androecium were used in the circumscription of Gomphrena and other genera of the Amaranthaceae (Schinz 1893; Eliasson 1988; Townsend 1993). Nevertheless, the homology of interstaminal and staminal appendages has been a subject of debate. Eliasson (1988) suggested a stepwise character state transition from so-called “pseudostaminodes” alternating with stamens as present in Alternanthera and Pedersenia (Fig. 6) to “apical filament lobes” present in many species of Gomphrena (e.g. G. haenkeana), Pfaffia and Xerosiphon (Fig. 6). Vrijdaghs & al. (2014) investigated the ontogeny of the androecium in Gomphrenoideae and found that the androecium tube develops from a circular intercalary meristem, without any indication of postgenital fusion. Also the appendages develop from this meristem without any traces of veins, suggesting their interpretation as stamen-tube appendages without any organ identity. Moreover, Sánchez-del Pino & al. (2019) concluded that the lateral filament appendages in Gomphrena and Pfaffia are de novo organs, thus not being homologous to the stamen tube appendages. This is consistent with the observation that the Amaranthaceae–Chenopodiaceae clade generally has a single whorl of antesepalous stamens without any trace of staminodes (Ronse de Craene 2013). Our reconstruction of ancestral states shows that the presence of androecial tube appendages (character 17, state 0; Fig. 6) is ancestral in Gomphrenoideae. They were lost in the gomphrenoid clade but apparently re-gained in Froelichia. In line with our results (Fig. 6), Eliasson (1988) illustrated organs like stamen tube appendages in Froelichia. However, in the absence of any ontogenetic information for Froelichia, their homology may be further investigated as we cannot exclude any lateral fusion of de novo lateral stamen appendages that may have occurred together with the evolution of a completely closed androecial tube in that genus. Sánchez-del Pino & al. (2019) annotated the presence of stamen tube appendages on a tree of Gomphrenoideae and consistently show their absence in the gomphrenoid clade, indicating their loss after the divergence of Pseudoplantago. However, their character coding of Froelichia differs from ours. Other major lineages of the Amaranthaceae also exhibit similar variation patterns in the androecium such as the aervoid clade (Hammer & al. 2019).

Fig. 7.

Overview on the evolution of Gomphrena in time. The tree represents the clade of the Gomphrenoideae from the overall divergence time estimation of Amaranthaceae and Chenopodiaceae calculated with BEAST and 95% HPD intervals as grey bars (see  Appendix S10 (wi.50.50301_Appendix_S10_BEAST-tree_node_numbers.pdf) for the full tree with precise divergence times). Pie charts depict the evolution of C4 photosynthesis in Gomphrena in the wider context of the Gomphrenoideae (C3 in blue and C4 in red). The core Gomphrena clade including Lithophila, Gossypianthus and Philoxerus constitutes the most diverse C4 lineage in Amaranthaceae. Note the young crown group ages of the Philoxerus+Australian Gomphrena clade in a similar range than the mostly Andean clade.

The evolution of lateral filament appendages (character 19) shows strong homoplasy. They have arisen twice within the core C₄ Gomphrena clade but were lost again in the common ancestor of the Philoxerus+Australian Gomphrena clade (Fig. 1, 6). And they arose independently in G. prostrata in the Pfaffia clade (Fig. 6). Sánchez-del Pino & al. (2019) also showed the presence and absence of lateral filament appendages in Gomphrena and relatives, but their tree did not sample the genus Gomphrena in a representative way and also did not provide the necessary resolution of nodes to make more specific conclusions. Vrijdaghs & al. (2014) suggested the androecial tube appendages to be part of an insect-pollination syndrome, functioning as floral nectaries. This may be true for androecial tube appendages in Iresine or Pedersenia but not for the lateral stamen appendages. These were attributed to attract pollinators by taking over this function from the corolla (Sánchez-del Pino & al. 2019). To the contrary, our results show that the gain of filament appendages (character 19, Fig. 6) in species of Gomphrena coincides with either the evolution of large and showy flowers or pseudanthia (e.g. G. haenkeana, G. mollis; see Fig. 5).

Androecial tubes have extended in length two times during the evolution of Gomphrena (character 16, Fig. 6), to the extreme of the up to 30 mm long tubes in G. pulchella. Floral evolution in the G. mollis–G. rupestris lineage is therefore convergent to G. haenkeana, G. pulchella and allies (mostly Andean clade; see Fig. 1). Long androecial tubes also correlate to showy flowers, which appears to indicate a specialized pollination syndrome. However, the pollination biology of Gomphrena is not well known. Based on scattered observations, Gomphrena and allies seem to be pollinated by insects. During field work for this study, moths were observed visiting flowers of G. pallida that introduced their proboscis through the androecium tube, and butterflies were reported to pollinate the Australian G. splendida (Barrett & Palmer 2015). The position, length and shape of filament appendages (characters 20 and 21) is highly homplastic ( Appendix S9 (wi.50.50301_Appendix_S9_ancestral_states_morphological_character_21.pdf) ). On the other hand, Lithophila muscoides has true staminodes, which result from the reduction of the androecium to two or three functional stamens (Fig. 6; Eliasson 1988). This species was classified as an own genus based on this feature and its specific Caribbean habitat (Swartz 1788) but it is deeply nested within Gomphrena.

Overall phylogenetic relationships and diversification of the core C₄ Gomphrena clade

The extended taxon sampling based on plastid matKtrnK (Fig. 2) and nrITS (Fig. 3) sequence data revealed the same principal lineages of Gomphrena and allies as in the analysis of combined plastid markers (Fig. 1) but indicates that the majority of species of the more extensive taxon set belong to a large “mostly Andean clade” (Fig. 2, 3) that receives maximum statistical support in all analyses. In the 27-taxon set this clade is just represented by G. haenkeana (Fig. 1). The increased taxon sampling also reveals the G. boliviana–G. martiana clade, the high-Andean G. meyeniana clade and a more extensive central to south Andean G. radiata–G. tomentosa clade also comprising G. umbellata based on evidence from matK-trnK (Fig. 2) and nrITS (Fig. 3). Sequence data from matK-trnK alone show the G. boliviana–G. martiana clade in a tritomy with the other two before mentioned clades but like nrITS provide strong support for the monophyly of each of the three clades. Gossypianthus is part of this major lineage and may be the sister group to both the Gomphrena radiata–G. tomentosa clade and the G. meyeniana clade as depicted in the combined analyses of plastid data (1.0 PP, 98% MP-JK, 91% ML-BS).

Three further clades diverge after the lineage with Gomphrena boliviana, G. meyeniana, G. radiata and relatives: one is the G. pulchella–G. cardenasii clade that consists of lowland species occurring in the Chaco (G. pulchella; Pedersen 1976), Cerrado (G. agrestris, G. demissa; Siqueira 1992). Gomphrena lanigera occurs in the Campo Rupestres in Brazil and reaches the eastern parts of Bolivia (Siqueira 1992; Borsch & al. 2015b), although ITS inconsistently depicts the latter species as sister to G. celosioides (the respective nodes are not well supported, Fig. 3). Furthermore, the Bolivian G. cardenasii, which is a small subshrub endemic to the Serrania de Chiquitos (eastern lowlands of Departamento La Paz) belongs here. Gomphrena cardenasii is a well-defined entity in terms of morphological characters and its samples appear on a long branch in the matK-trnK tree (Fig. 2). The other two major lineages within the core Gomphrena C₄ clade are the clade comprising all Australian species of Gomphrena plus Philoxerus and Lithophila (the Philoxerus+Australian Gomphrena clade) and G. celosioides, the specimens of which appear as a rather isolated clade (Fig. 2, 3). A sister group relationship of G. celosioides to G. haenkeana representing the mostly Andean clade is resolved in the analyses from the combined plastid dataset (Fig. 1), albeit with only weak support.

Trees inferred from plastid and nuclear partitions converge in depicting two sublineages (A and B) of a clade of mostly Andean species (Fig. 2, 3), both of which predominantly comprise perennial and annual species from dry valleys and Prepuna habitats (from 2500 to 4000 m a.s.l.) of the central Andes, with the exception of Gomphrena perennis, which is a morphologically variable species extending from the Chaco and Bosque Tucumano-Boliviano to the Andean dry valleys (Borsch & al. 2015b). Gomphrena cf. nitida from Oaxaca in Mexico appears in a relatively isolated position unresolved in a polytomy. However, further individuals representing the central American populations (Borsch 2001) need to be sampled to clarify if the morphologically allied plants from the region are closely related.

Our trees include samples of the annual Gomphrena phaeotricha from the South of Bolivia and the North of Argentina, depicted in sublineage A. The species was described by Pedersen (1976) based on differences in floral morphology in comparison to G. pallida. Our results agree with Pedersen (1976) by providing molecular evidence that G. phaeotricha and G. pallida are only distantly related, essentially belonging to two different sublineages A and B. However, the circumscription of both species in the sense of Pedersen (1976) and as currently accepted may not reflect natural entities. Pedersen recognized several infraspecific taxa in G. pallida to accommodate some of the morphological variation but at his time did not apply any evolutionary method. Further work is therefore underway on species limits in the mostly Andean clade (Ortuño & Borsch, unpubl. data). One other annual species was already described based on its deviant floral morphology (Ortuño & Borsch 2005), G. mizqueensis from dry valleys in the province of Mizque (department of Cochabamba). The matK-trnK tree depicts it as closely related to samples of G. pallida within subclade B but also the perennial G. fuscipellita, a morphologically easily recognizable species described from rocky outcrops with herbaceous vegetation in the same area (Ortuño & Borsch 2005) but apparently extending to the Toro Toro National Park in northern Potosí (T. Ortuño, pers. obs.). Other perennials also belong to subclade B such as G. oligocephala, G. bicolor and G. potosiana according to both ITS and matK-trnK (Fig. 2, 3). On the other hand, subclade B comprises the perennials G. perennis, G. trollii and the annuals G. haenkeana and G. ferruginea alongside with G. phaeotricha. Our earlier hypothesis that the Andean annuals could represent a lineage that has adapted to survive the dry season with seeds as diaspores and then radiated in the inner Andean dry valleys (Ortuño & Borsch 2005) can thus not be fully accepted. The annual species of Gomphrena grow and flower in March and April after the rains (López 2003). Both Andean sublineages A and B comprise annuals and perennials, indicating multiple shifts in life forms, which in the case of Gomphrena could have occurred in conjunction with the origin of dry valleys, resulting in a strongly geographically governed pattern of diversity. The two Andean subclades of Gomphrena started to diversify about 2.6 Ma (0.7–7.1 and 0.2–5.0 95% HPD, respectively; Fig. 7,  Appendix S10 (wi.50.50301_Appendix_S10_BEAST-tree_node_numbers.pdf) ) and the annual lineages appear to be not much older than 1 Ma.

The Gomphrena meyeniana clades comprises plants growing at high altitudes that all possess a characteristic and unique morphology with leaves arranged in a rosette that arises from a reduced main axis, flower heads few to solitary on inflorescence axes arising from a stout, almost invisibly short stem, bracteoles without crest and stamens almost completely fused into a staminal tube lacking filament appendages Holzhammer (1955, 1956; see Fig. 4, 6). Our molecular phylogenetic data that depict the G. meyeniana clade as a highly supported lineage with a crown group diverging on a long stem (Fig. 2, 3) are in line with this. The latest treatment at species level is by Pedersen (1990), who only recognized G. meyeniana with several infraspecific taxa that reflect the morphological and ecological variation within the clade. Our results indicate significant phylogenetic diversity within the G. meyeniana clade but strongly diverging ITS ribotypes visible as polymorphic sites in our ITS pherograms within most individuals (Ortuño & Borsch pers. obs.) show that hybridization occurs within this clade that is likely to contribute to the phenotypic variation observed. Nevertheless, all ITS copies found within G. meyeniana and allies are clearly different from all other lineages of Gomphrena, underscoring that the role of reticulate patterns and incomplete lineage sorting needs to be analysed to delimit species within the clade but will not influence our view on the composition of the G. meyeniana clade.

The Gomphrena radiata–G. umbellata–G. tomentosa clade appears with maximum support in plastid and nuclear trees (see Fig. 2, 3) and encompasses species adapted to dry habitats. Trees show two subclades, one of which includes all the G. tomentosa specimens, whereas the other subclade includes G. umbellata and G. radiata. The main morphological difference between these subclades is the life form present. Gomphrena tomentosa and close relatives are perennials (Fries 1920; Hunziker & Subils 1977), whereas G. radiata and G. umbellata are annuals (Pedersen 1976). Gomphrena umbellata is extremely adapted to sandy places at high elevations (2800 to 4400 m a.s.l.; Borsch & al. 2015b). Its leaves are reduced to the uppermost cauline leaves and the apical leaves subtending paracladia whereas the stems are under the sand. In all species of the G. radiata–G. umbellata–G. tomentosa clade the filaments are fused into a tube for their most part lacking lateral filament appendages (Fig. 6). Based on the absence of these appendages Fries (1920) proposed the “Chnoanthus group”, later formally described as a section by Holzhammer (1956). Fries (1920) hypothesized that Gomphrena comprises a series of species, where the filament appendages were reduced in steps during the evolution of the genus, and G. tomentosa appears at the end of this series. Although the evolution of lateral filament appendages is homoplastic (Fig. 6,  Appendix S9 (wi.50.50301_Appendix_S9_ancestral_states_morphological_character_21.pdf) ), the section Chnoanthus appears as a natural group. Recently, Bena & al. (2017) generated trnL-F sequences of an individual of each G. umbellata, G. radiata, G. cladotrichoides and G. mendocina, the latter two of which are close relatives of G. tomentosa. They also recovered these four taxa in a clade, albeit without statistical support of any of the nodes.

All three before mentioned clades within Gomphrena diversified at higher elevations in the Andes (Fig. 7). The mostly Andean clade diverged around 8 Ma (4.0–13.6 95% HPD) from lowland ancestors whereas the crown group has an age of just 4.3 Ma (1.8–10.3 HPD). The G. meyeniana clade (stem group age 7.8 Ma; 2.9–13.5 HPD) and the G. radiata–G. tomentosa clade (stem group age 8.8 Ma; 3.5–15.8 HPD) must have colonized the rising Andes independently. Considering that the central Andes had only about half of its modern elevation by the late Miocene (Gregory-Wodzicki 2000; Garzione & al. 2008; Jiménez & al. 2009) followed by a continuous uplift in the last 10 Ma to reach the modern elevation of the central Andean plateau of ∼4 km further exceeded by the Eastern and Western Cordilleras, the dates estimated for the origin of the Andean clades of Gomphrena fit well to the geological history. Compared to the radiation of the Andean clade of Lupinus L. (Fabaceae) with a crown group age of about 1.5 Ma (Hughes & Eastwood 2006) the Andean diversification within Gomphrena started somewhat earlier (crown group ages of 2.6 Ma [0.7–7.1 and 0.2–5.0 95% HPD, respectively] for the Andean subclades A and B, and 2 Ma for the G. meyaniana clade). The youngest crown group within Gomphrena (G. meyeniana and allies) exclusively grows above 3000 m (up to 4700 m; Borsch & al. 2015b) whereas members of Andean subclades A and B are often range-restricted in dry inner Andean valleys or at elevations >3000 m. This underscores the importance of valley systems with dry climates that originated as a consequence of the Andean uplift (Strecker & al. 2007) in particular in the Eastern Cordillera (Bolivia and northwestern Argentina) and probably fuelled reproductive isolation as well as adaptation to dry habitats. However, better resolved species trees and a more thorough understanding of species limits and their exact geographical distribution will be needed to further illuminate the diversification of Gomphrena in the Andes (e.g. testing if the morphologically variable and widespread G. perennis is in fact a single species). Nevertheless, Gomphrena is an interesting model to study plant diversification with the uplift of the Andes as it contains at least three independent ascents onto the Andes.

The Philoxerus+Australian Gomphrena clade comprises all sampled Australian species as well as Lithophila and Philoxerus, the latter two of which are plants adapted to coastal environments with fleshy leaves (Fig. 4). The species of Lithophila and Philoxerus appear morphologically similar but Lithophila differs in the reduced number of two stamens (Eliasson 1988, Fig. 6). The filaments of Lithophila and Philoxerus are united only basally, forming a cup, and lateral filament appendages are absent. To the contrary, the Australian species have morphologically variable filaments, some with lateral filament appendages (Townsend 1993). Some Australian species differ from the neotropical species by longer stigmas (Palmer 1998), but this variation will be analysed in more detail in a future study. Nevertheless, all the species of the Philoxerus+Australian Gomphrena clade share the main morphological characteristics of the core C₄ Gomphrena clade (Fig. 5, 6) so that it is proposed here to include the two genera into Gomphrena.

An Australian lineage derived recently from South American ancestors

The crown group of the Australian clade is of very recent origin (3.9 Ma, 1.3–8.6 95% HPD) as is the more inclusive clade comprising Philoxerus and Lithophila (crown group age of 5.7 Ma [2.0–12.4 95%], Fig. 7 and  Appendix S10 (wi.50.50301_Appendix_S10_BEAST-tree_node_numbers.pdf) ). Therefore, a Gondwanan vicariance hypothesis (Upchurch 2008; Barker & al. 2007a) for the origin of the Australian Gomphrena clade can be clearly rejected. Our results further show that this Australian clade is very deeply nested among South American and Caribbean ancestors, underscoring that long distance dispersal is the only plausible explanation for the origin of the Australian Gomphrena clade.

In Fig. 8 we present an overview on the global distribution of the core C₄ Gomphrena clade (pale yellow signature) as well as Philoxerus (dot map for each of the three currently accepted species). Gomphrena is not native in Africa (Townsend 1985) and therefore is not recorded for this continent although our two samples of G. celosioides are of African origin (Fig. 2, 3). The possible first branch of the Philoxerus+Australian Gomphrena clade is Lithophila (Fig. 2, 3), which is endemic to all Caribbean islands (Acevedo-Rodríguez & Strong 2012; not illustrated), where it grows in limestone crevices at the coast. Gomphrena and Philoxerus do not occur in the southernmost part of South America, nor on Tasmania or New Zealand, indicating that a trans-Tasmanian dispersal route is very unlikely for Gomphrena and relatives. Sanmartín & al. (2007) inferred this as a frequent pattern, apparently facilitated by westward directed circumpolar currents, but more in temperate plant groups.

A striking result of this investigation is to find the Australian Gomphrena clade nested among species that inhabit coastal habitats. Their adaptations such as fleshy leaves, resistance to high concentrations of salt and adventitious roots allowing vegetative reproduction from broken off stems (Cordazzo 2007; Cordazzo & Seeliger 2003, T. Borsch pers. obs.) have apparently led to marine dispersal. Philoxerus (= Blutaparon) has reached the west coast of tropical Africa (see Fig. 8) but also the islands of Galapagos. Blutaparon rigidum is reported as a Galapagos endemic, which has some similarities to Lithophila in floral morphology (Eliasson 1990) but is a morphologically distinct plant with upright habit. However, it is only known from two historical collections and now extinct because it was covered by the eruption of the volcano on the island of Santiago (Eliasson 1971). But Galapagos harbours at least two other species of Lithophila (L. radicata Standl., L. subscaposa Hook. f.). And interestingly, other lineages of Amaranthaceae also reached Galapagos, so species of Alternanthera, which have a derived position among neotropical ancestors within this monophyletic genus (Sánchez-del Pino & al. 2012).

Based on the above, we hypothesize that Philoxerus (= Blutaparon) or Lithophila-like plants adapted to coastal environments were the ancestors of the Australian Gomphrena. They were dispersed to Australia across the Pacific Ocean through marine currents (Fig 8, based on Kasang 2018) such as the North and South Equatorial current, which then leads over into the East Australian current and which were present since about 6 Ma. Interestingly, the dispersal of the ancestor of Philoxerus wrightii could be explained with the Kuroshio current (Fig. 8). Grehan (2001) reported Galapagos-Central-America-Caribbean tracks were also reported in other organisms such as isopods (Nesophilosia, Troglo-philoscia), snakes (Antillophis) or beetles (Ablechrus) and even Galapagos-Australia tracks in angiosperms (Nicotiana), termites (Insitermes) and beetles (Pitinus). The latter dispersals may have occurred in the late Miocene, which roughly corresponds to the stem age of the Philoxerus+Australian Gomphrena clade dated 10 Ma (5.7–16.9 95% HPD). A detailed phylogeographic analysis as well as the inclusion of all available specimens of this clade into a well-resolved dated phylogeny will certainly further illuminate this scenario in the future.

Fig. 8.

Distribution of Gomphrena sensu stricto (the core C4 Gomphrena clade including Gossypianthus, Lithophila and Philoxerus as well as Gomphrena mollis and G. rupestris) in the Americas and Australia. Only the native range is shown. Specimen records corresponding to the geographic occurrence of species so far classified as Philoxerus and now found to be part of Gomphrena are shown as dots. Gomphrena vermiculare is widespread along the coasts of the Caribbean islands and adjacent mainlands, as well as along the coast of tropical East Africa. Gomphrena portulacoides extends further south along the eastern coast of South America. Other species have reached the Galapagos Islands (G. rigida) as well as the islands extending from southern Japan to Taiwan (G. wrightii). This pattern can best be explained by long-distance dispersal through sea currents (Kasang 2018), which are therefore hypothesized to have transported a coast-adapted ancestor of the Australian Gomphrena lineage to Australia. The blue line shows the extension of major surface currents of the oceans that appear to be relevant in this context, such as the Antarctic circumpolar (1a/b), the South Pacific (2), the Humboldt (3), the South (4a) and North Equatorial (4b), the Kuroshio (5), the East Australian (6), the South Atlantic (7), the Benguela (8), part of the South Equatorial in the Atlantic (9) and the Brazil current (10).

C₄ evolution in Gomphrena and allies

Within the subfamily Gomphrenoideae the C₄ photosynthetic pathway was reported from a large number of species of Gomphrena, as well as from several members of Alternanthera and the genera Blutaparon, Froelichia, Guilleminea, Lithophila and Tidestromia (Sage & al. 2007). Our results show that the C₄ species of Gomphrena (Fig. 2, 3, 7) all belong to a single clade. The members of this core Gomphrena clade are characterized by metareticulate pollen where the tectum is reduced to a distal band (Borsch 1998; Fig. 7). On the other hand, most C₃ species of Gomphrena (e.g. G. elegans, G. mandonii) have pollen with closed tecta similar to Pfaffia. They are resolved in a distant lineage together with the C₃ genera Hebanthe and Pfaffia in both chloroplast and nrITS trees (Fig. 2, 3, 7 and  Appendix S10 (wi.50.50301_Appendix_S10_BEAST-tree_node_numbers.pdf) ). The only exception are the two C₃ species G. mollis and G. rupestris that belong to the core Gomphrena clade in which they share the morphology of pollen with reduced tecta (Ortuño & Borsch, unpubl. data).

Ancestral character state reconstruction using the chloroplast trees (Fig. 7 for matK-trnK; MCMC model; the same results are found with the 27-taxon set, not shown) indicates that C₄ photosynthesis arose in the common ancestor of the core Gomphrena clade plus Froelichia but reversed back to C₃ in the clade of G. mollis and G. rupestris. Analyses under a ML model in BayesTraits slightly favour a reversal in the common ancestor of the core C₄ Gomphrena clade plus the G. mollis–G. rupestris clade and a regain of the C₄ syndrome by the ancestor of a core C₄ Gomphrena clade ( Appendix S11 (wi.50.50301_Appendix_S11_matK-trnK_extended_MCCtree_MCMC_character_11.pdf) ). However, this would be a much more complicated scenario that therefore seems less probable. To the contrary, the ITS trees favour a single origin of C₄ photosynthesis in a common ancestor of the core C₄ Gomphrena clade plus the Gomphrena prostrata–Guilleminea clade ( Appendix S12 (wi.50.50301_Appendix_S12_ITS_extended_MCC-tree_MCMC_character_11.pdf) ). The about ten species of the core C₄ Gomphrena clade, for which leaf anatomy has been investigated so far (Carolin 1978; Estelita-Teixeira & Handro 1984; Fank-de-Carvalho & al. 2015), show thick-walled parenchymatous bundle-sheath, and in most cases dimorphic chloroplasts corresponding to an NADP-ME subtype. Similar anatomy was also found in Froelichia (Carolin 1978) and Guilleminea (Filippa & Espinar 1993). This was called “Gomphrena type” (Carolin 1978; Kadereit & al. 2003), and considered to be present in all C₄ Gomphrenoideae. Gomphrena prostrata has thickened and lignified bundle sheath cell walls (Estelita-Teixeira & Handro 1984), but this may be rather seen as an adaptation to xeric environments like white sand habitats. Leaf anatomical characters seem therefore not informative indicating a sole in the whole subfamily, contrary to the strong differentiation of leaf anatomy in Chenopodiaceae (Kadereit & al. 2003). It is nevertheless remarkable that a potential C₃ reversal coincides with a possibly reticulate origin of the respective reversed C₃ lineage (G. mollis–G. rupestris clade). An alternative explanation would there be a hybridization of a C₃ ancestor with an early branch of the Gomphrena C₄ lineage as a maternal parent, so we actually see a captured C₄-type chloroplast. Experimental hybrids of C₃ and C₄ species in Atriplex and Flaveria (see Kadereit & al. 2017 for review) rather showed differential segregation of different characters of photosynthesis types, leading to C₂-like offspring. Further biochemical and physiological analysis of G. mollis and G. rupestris would be needed to see if they differ from other C₃ Gomphrenoideae. Nevertheless, the δ13C values of G. mollis and G. rupestris (Sage & al. 2007) are not different from those C₃ species.

Sage & al. (2007) mapped the taxonomic distribution of C₄ photosynthesis using a matK-trnK tree of the Amaranthaceae under parsimony and depicted three C₄ clades in subfamily Gomphrenoideae: (1) a clade of Froelichia–Guilleminea–Blutaparon–Gomphrena (species with distal tectal bands in their pollen grains), (2) a sublineage of Alternanthera, and (3) Tidestromia. The plesiomorphic condition in the monophyletic genus Alternanthera was later confirmed to be C₃ in a much more detailed study by Sánchez-del Pino & al. (2012) whereas C₄ was considered to be derived within their clade B3. This clade is represented in our dataset D by A. microphylla and A. pungens, which is congruently inferred as nested within Alternanthera in our analyses (Fig. 7, Appendices  S11 (wi.50.50301_Appendix_S11_matK-trnK_extended_MCCtree_MCMC_character_11.pdf) ,  S12 (wi.50.50301_Appendix_S12_ITS_extended_MCC-tree_MCMC_character_11.pdf) ). However, the position of C₄ Tidestromia is inconsistently shown as sister to Pedersenia (with ITS,  Appendix S12 (wi.50.50301_Appendix_S12_ITS_extended_MCC-tree_MCMC_character_11.pdf) ), in a grade with Alternanthera and Pedersenia as successive sisters (with matK-trnK; Fig. 7) or as sister to Alternanthera (Sánchez-del Pino & al. 2009). This makes a precise dating of the age of Tidestomia as a C₄ lineage difficult, but otherwise does not change the picture of three independent C₄ origins in the Gomphrenoideae (Sage & al. 2007). Nevertheless, the early C₄ evolution of the Gomphrena clade is more complex as evidenced by the improves sampling in this investigation. Amaranthaceae s.str. like the chenopod lineages (Kadereit & al. 2012) therefore exhibits multiple gains and losses of C₄.

Recently, Bena & al. (2017) studied macroclimatic niche limits and C₄ evolution in Gomphrenoideae. The authors compiled available plastid sequences from GenBank (mostly from Müller & Borsch 2005; Borsch & al. 2012; Sage & al. 2007; Sánchez-del Pino & al. 2009, 2012) and added new trnL-F sequences from ten species. However, they did not obtain any significant statistical support on nodes within in the gomphrenoids (sensu Sánchez-del Pino & al. 2009) which may be explained by the patchy matrices, and also they did not sample relevant taxa like Gomphrena mollis or G. rupestris, thus limiting conclusions on the origin of C₄ in Gomphrenoideae. Nevertheless, the general conclusion of Bena & al. (2017) that C₄ Gomphrenoideae, unlike C₄ grasses, shifted their niches into regions with colder winter climates is in line with the results presented here.

The age of the crown group of the speciose core Gomphrena C₄ clade plus Froelichia is inferred as 18 Ma (10.2–28.4 95% HPD; the stem diverged from C₃ Xerosiphon at 20.6 Ma [11.8–32.9 95% HPD]), and thus is slightly younger than the earliest inferred gain of C₄ in the crown group of Salsoloideae/Camphorosmoideae (Kadereit & al. 2012). The stem of the C₄ clade within Alternanthera originated at 18.5 Ma (7.8–33.7 95% HPD) and the C₄ genus Tidestromia diverged at 27.7 Ma (19.3–38.9 95% HPD). Tidestromia is very different morphologically to the other extant genera and C₄ photosynthesis might have evolved at some along the stem later (we have only included one of the 14 species here; and morphologically intermediate ancestors might now be extinct). Thus, we can therefore assume that C₄ originated in Gomphrena and other Gomphrenoideae in roughly the same time interval than in Chenopodiaceae s.str. subsequently to the drop of CO₂ concentrations at Eocene-Oligocene boundary (Christin & al. 2011; Kadereit & al. 2012). Nevertheless, the core Gomphrena clade stands out by being a C₄ clade that diversified at least twice into high elevation Andean environments. Species such as G. umbellata (3800–4400 m a.s.l.) and G. meyeniana (2600–4000 m a.s.l.) and allies probably constitute the highest populations of any C₄ plant, thereby being competitive in sub-arid as well as humid Puna vegetation.

Toward a phylogeny-based circumscription of Gomphrena

The genus Gomphrena has a complex taxonomic history. It was established by Linnaeus (1753) with G. globosa L. and G. flava L. [≡ Alternanthera flava (L.) Mears]. In later publications, Linnaeus added nine more species which now belong to four genera: Gomphrena, Philoxerus, Froelichia and Alternanthera. The type of the genus Gomphrena was chosen later as G. globosa with the typification effectively published by Hitchcock (Hitchcock & Green 1929). Gomphrena globosa is included in the molecular trees of Sánchez-del Pino & al. (2009) in a position that corresponds to the mostly Andean clade (Fig. 2, 3) and thus belongs to the core C₄ Gomphrena clade.

Martius (1826) then treated 57 species of Gomphrena with detailed descriptions and also described the genera Pfaffia, Hebanthe and Serturnera. Endlicher (1837) reduced these genera to sections of Gomphrena. This extended generic concept was largely followed by Moquin-Tandon (1849), who in addition recognized the section Xerosiphon, and also reduced the genera Ninanga and Wadapus described by Rafinesque (1837) to sectional level within Gomphrena. The currently used generic circumscription goes back to Seubert (1875) who excluded the sections Hebanthe, Pfaffia and Serturnera from Gomphrena. This was followed by Schinz (1893), Holzhammer (1955; 1956) and Townsend (1993), and confirmed by more recent investigations using molecular and morphological characters (Borsch & Pedersen 1997; Müller & Borsch 2005; Sánchez-del Pino & al. 2009). The resurrection of Xerosiphon as an own genus contrary to its treatment as a section of Gomphrena since Seubert (1875) was proposed by Pedersen (1990), and is in line with results of this study as well as Müller & Borsch (2005) and Sánchez-del Pino & al. (2009). Siquera (1992) included Pseudogomphrena R. E. Fr. which has pollen and floral morphology similar to Pfaffia and therefore is probably distantly related to the Gomphrena clade (Eliasson 1988; Borsch 1998). The same applies to some other species currently included in Gomphrena with Pfaffia-type pollen (here represented by G. mandonii; see Fig. 1–3).

Generic concepts should be based on phylogenetic insights, rendering genera monophyletic, but there should also be morphological characters that allow to recognize genera. This investigation provides robust evidence from nuclear and plastid genomic compartments that Lithophila, Philoxerus and also Gossypianthus are derived within the core C₄ Gomphrena clade. As currently classified (Eliasson 1988; Townsend 1993; Hernández-Ledesma & al. 2015) they represent cases of segregate genera that show morphological adaptions to specific habitats. Morphological synapomorphies of the core C₄ Gomphrena clade including Lithophila, Gossypianthus and Philoxerus (see previous paragraphs) further underscore that the species of these three genera should in fact be classified within Gomphrena.

The situation in Guillemina is different. The species of Guilleminea could represent a convergent origin of a procumbent habit in relation to Gossypianthus and the Gomphrena radiata–G. tomentosa clade. Whereas Guilleminea stands out by a perigynous flowers (Fig. 6, character 13; Eliasson 1988), its phylogenetic position and origin is not yet clear. This unique floral morphology historically always led authors to maintain Guilleminea at generic rank. Guilleminea could therefore (together with G. prostrata) be viewed as the earliest diverging branch in an even more widely circumscribed genus Gomphrena or maintained as an own genus. The latter would imply to find morphological characters that would support an inclusion of G. prostrata into Guilleminea or a generic status of G. prostrata. Our case of Gomphrena also exemplifies the stepwise progress in recognizing natural entities that can only be followed by a stepwise adjustment of classification systems to better reflect the natural history of the group (Borsch & al. 2015a). Based on the results of this investigation which consistently reveal the core Gomphrena clade as a distinct entity we propose to include Blutaparon, Gossypianthus and Lithophila into Gomphrena whereas any change of classification appears to be premature for Guilleminea.

The relationships of the Australian species of Gomphrena were debated since the early days of botanical systematics, even affecting the use of Philoxerus R. Br. and Blutaparon Raf. as generic names. Robert Brown (1810) created an own genus Philoxerus to accommodate P. conicus and P. diffusus. Standley (1917) lectotypified the genus name with P. conicus, which is a species of sandy soils close to coasts in Australia. This was in line with Hooker (1880) who had defined Philoxerus as a genus of coastal plants from the Americas, Africa and Australia rather than using discrete morphological characters to support his genus concept. When Mears (1982a, b) proposed to use the name Blutaparon for the American (and African) taxa, he neither had solid morphological nor phylogenetic evidence for the distinctness of these taxa that could have supported his view. Consequently, Hernández-Ledesma & al. (2015) pointed out that Blutaparon should be treated as a synonym of Philoxerus. The phylogenetic results now presented in this paper solve this long-standing debate. Our phylogenetic data also show that the only partially fused androecium used by Robert Brown to discriminate species of Philoxerus from the then known New World species of Gomphrena with a complete staminal tube and lateral stamen appendages, represents a character state that is homoplastic within the Gomphrena clade (Fig. 6).