Material and methods

We collected samples of Astianthus viminalis for anatomical and phylogenetic studies in the field. To broaden our geographic sampling, we included additional samples of A. viminalis from the National Herbarium of Mexico and its wood collection (MEXU; see Table 1). To determine the phylogenetic placement of Astianthus, we generated sequences of A. viminalis, as well as sequences of Bignonia potosina (K. Schum. & Loes.) L.G. Lohmann, and Tecoma stans (L.) Juss. ex Kunth for this study (see Table 1 for collection or specimen information). All other sequences of ingroup and outgroup taxa were the same as those used in (^{Olmstead et al. 2009}; a complete list with voucher information and GenBank accession numbers can be found in this study). We also included 25 additional sequences from taxa available on GenBank (Supplementary material 1, Tabla S1) to fill gaps in rbcL sequences from the original sampling.

DNA extraction, amplification, and sequencing. We extracted DNA using the DNeasy Plant Mini Kit (Qiagen, Valencia, California) from either silica-dried plant material, fresh material, or herbarium specimens, following the manufacturer’s protocols.

For each accession, we amplified portions of the ndhF and rbcL genes and the trnL-F spacer. These three regions from the plastid genome have been useful to estimate phylogenetic relationships within the Bignoniaceae at the tribal (^{Olmstead et al. 2009}), generic (^{Lohmann 2006}), and species (e.g.,^{Fonseca & Lohmann 2015}, ^{Carvalho-Francisco & Lohmann 2020}) levels. We sequenced the ndhF marker in two pieces, using the PCR primer pairs 5F-1318R and 972F-3R described in an earlier study (^{Olmstead & Sweere 1994}). For the trnL-F region, we used primers C and F (^{Taberlet et al. 1991}), and for rbcL, we used F and R primers previously described (^{Hipkins et al. 1990}, Supplementary material 2, Figure S1). PCR reactions were prepared by adding each primer at 1mM to GoTaq^® Green Master Mix (Promega M1722), and adjusting with ddH₂0 for a final volume of 25µL. PCR conditions consisted of an initial denaturation of 90s at 96 ºC, followed by 35 cycles of 30s at 95 ºC, 60s at 55 ºC, 60s at 72 ºC, and a final extension of 4min at 72 ºC. Upon completion, PRC products were run in a 1% agarose gel and assessed for size and quality in a UV transilluminator using gel red (Biotium 41003, Hayward, California).

Sanger sequencing was carried out at the National Biodiversity Laboratory (LaNaBio) at the Institute of Biology, UNAM, Mexico, in both directions. Briefly, sequencing reactions were prepared using BigDye Terminator 3.1 (Applied Biosystems), and run for 30 cycles consisting of 10s at 96 °C, 5s at 50 ºC, and 4 min at 60 ºC. Samples were cleaned using Centri-Sep plates (Thermo Fisher Scientific, Waltham, Massachusetts) following the manufacturer’s directions. Samples were analyzed in an Applied Biosystems ABI 3730xl 96-capillary DNA analyzer (ThermoFisher Scientific, Waltham, Massachusetts).

Sequence editing and alignment. We examined and edited raw chromatograms for all newly generated sequences in Sequencher 5.4.6 (Gene Codes, Ann Arbor, Michigan). Contigs were assembled using default settings without lowering the threshold. We aligned each region individually and manually in Mesquite 3.5 (^{Maddison & Maddison 2011}), giving preference to transitions over transversions. All sequences generated for this study are available in GenBank with the following accession numbers: MT235272-MT235276 (rbcL), MT232737-MT232741(ndhF), and MW291155-MW291159 (trnL-F).

Phylogenetic analyses. We assessed evolutionary models for each region separately using jModelTest 2.1.7 (^{Darriba et al. 2012}), and the Akaike information criterion (AIC; ^{Akaike 1974}). We analyzed each dataset separately and in combination, following a total evidence approach (^{Kluge 1989}).

We conducted Maximum Likelihood analyses using RAxML-HPC (^{Stamatakis 2014}) as implemented in the XSEDE tool in CIPRES (^{Miller et al. 2010}), using a random seed (-p) of 12345, and the default (25) number of distinct rate categories (-c). For each analysis, matrices were run according to the model selected and the slow ML search algorithm. For the combined analyses including data from all three partitions (ndhF, rbcL, and trnL-F), we allowed for a mixed model (also slow search algorithm). Bootstrap analyses (100 replicates) were performed with the RAXML fast bootstrap algorithm implemented in CIPRES.

Bayesian analyses were run in MrBayes 3.2 (^{Ronquist et al. 2012}), as implemented in CIPRES. An initial 5 million generation analysis was run using the selected model for each region to optimize parameters (including temperature) and ensure that the chains were running properly and reached stationarity. We checked chain swap information and parameter acceptance rates to ensure that parameters were acceptable (between 0.1 and 0.7), making sure all parameters had an ESS > 200, and examined appropriate chain behavior in Tracer 1.7.1 (^{Rambaut et al. 2018}). We then conducted a second run including 20 million generations for each of our analyses. For the analyses with the concatenated dataset, the parameters associated with the model of evolution (Revmat, Statefreq, and Shape) were unlinked, while the ratemultipler, the topology and the branch lengths were linked across partitions. For all analyses, we implemented a temperature of 0.1 for the cold chain to ensure appropriate mixing. We sampled every 1,000 generations, and eliminated 25 % of the trees as burn-in. Sampling of the parameter space by the MCMC chains was summarized using the .sump and .sumt commands, while trees were visualized in FigTree 1.4.3 (^{Rambaut 2010}).

We conducted all analyses on the Cyberinfrastructure for Phylogenetic Research cluster (CIPRES; ^{Miller et al. 2010}), which is housed at the San Diego Supercomputer Center (www.phylo.org/), and tree visualization and annotation was performed in R (^{R Core Team 2020}).

Anatomical sampling and methods. Woods pulled from the MEXU xylarium were rehydrated in boiling water and glycerin for two hours following ^{Pace (2019)}. All samples were softened in 4% ethylenediamine for two days within a paraffin oven (^{Carlquist 1982}). Anatomical sections of the transverse, longitudinal radial and longitudinal tangential planes were performed with the aid of a sliding microtome and permanent steel knives sharpened with sandpapers of different grids (^{Barbosa et al. 2018}). Wood sections were obtained from unembedded materials and stained in 1 % aqueous safranin. Samples with cambium and bark underwent a previous step, being gradually embedded in polyethylene glycol 1500 (^{Rupp 1964}), and subsequently sectioned with the aid of an anti-tearing coat of a polystyrene resin (^{Barbosa et al. 2010}). The latter were double-stained for 15 minutes in Safrablau (^{Bukatsch 1972}, modified by ^{Kraus & Arduin 1997}). All sections were dehydrated in an ethanolic series, with butyl acetate being used in the last step, and mounted in Canada Balsam to make permanent slides.

Wood descriptions followed the IAWA Committee for hardwood (^{IAWA Committee 1989}), IAWA Committee for bark features (^{Angyalossy et al. 2016}), and ^{Carlquist (2001)}, adjusting to the specificities of the family whenever needed. Measurements were performed using ImageJ 1.52a (National Institute of Health, USA, www.imagej.nih.gov/ij, ^{Rasband 2012}). Since approximately half of the Bignoniaceae family is composed of lianas, and it has been well-documented that lianas tend to converge to similar anatomies (^{Carlquist 1985}, ^{Angyalossy et al. 2012}, ²⁰¹⁵, ^{Chery et al. 2020}), we focused the comparison of Astianthus with shrub and tree members of the family.

Results

Phylogenetic placement of Astianthus. A summary of our individual and combined data matrices, including dimension, number of variable and parsimony informative characters are presented in Table 2. The GTR + gamma was recovered as the best model of DNA substitution in all analyses and implemented for all datasets.

The results of the Bayesian and Maximum Likelihood analyses of the combined datasets are largely congruent with those of ^{Olmstead et al. (2009)} and ^{Lohmann (2006)}, including strong support for the tribes within Bignoniaceae as well as the family (Figure 2).

Figure 2 Consensus tree derived from a 20 million generation Bayesian analysis of the concatenated dataset (ndhF, rbcL, trnL-F). Both Bayesian Inference and Maximum Likelihood analyses strongly support Astianthus as sister to Campsis, within the Tecomae s.s. Posterior probabilities are provided in bold above branches and maximum likelihood bootstrap values in regular font below branches. 

Our phylogenetic combined analyses based on Maximum Likelihood (ln = -38901.785924) and Bayesian Inference frameworks led almost identical topologies with minor differences not related to the placement of Astianthus, therefore only the consensus Bayesian tree is shown (Figure 2; the ML tree is available in Supplementary material 2, Figure S1). In all analyses, Astianthus is strongly supported as monophyletic (1.0 PP, 100 % ML BS). Astianthus falls within the Core Bignoniaceae clade (1.0 PP, 100 % ML BS; sensu^{Olmstead et al. 2009}), within tribe Tecomeae s.s. (1.0 PP, 98 % ML BS), and sister to Campsis radicans (L.) Bureau (1.0 PP, 100 % ML BS).

Wood anatomy of Astianthus. Growth rings distinct, delimited by narrower vessels and radially narrow fibers (Figure 3A). Wood semi-ring porous (Figure 3A). Vessels without a specific arrangement, solitary or in radial multiples of 2-3 (Figure 3A-B), clusters of 3-4 vessels common, perforation plates simple, some wide vessel elements with foraminate perforation plate on horizontal end walls (Figure 3C). Intervessel pits alternate, minute (6 µm), vessel-ray pitting with distinct borders, similar to intervessel pits in size and shape throughout the ray cell, helical thickening absent. Vessel diameter 117 ± 32 µm, frequency 14 ± 3 vessels/mm², two vessels per group, vessel length 258 ± 38 µm. Tyloses and deposits absent both in sapwood and hardwood. Fibers thin to thick walled (Figure 3A-C, E), with simple to minute bordered pits, septate fibers present (Figure 3E). Axial parenchyma vasicentric to aliform with short confluences (Figure 3A-B), and confluences marginating the rays (Figure 3B), with 2-4 cells per parenchyma strand (Figure 3D). Rays 3-4-seriate (Figure 3D-E), longitudinal merging of two rays common, rays lower than 1 mm. Rays either homocellular with procumbent cells only (Figure 3F) or heterocellular, with body composed of procumbent cells and one row of square marginal cells (Figure 3G). Sheath cells common (Figure 3E). Axial parenchyma cells storied (Figure 3D), and in certain areas narrow vessels also storied, but not conspicuously (storied fusiform cambial initials). Crystals absent.

Figure 3 Wood anatomy of Astianthus. A-B. Transverse section. Semi-ring porous wood, growth rings delimited by narrower vessels, and radially narrow fibers (asterisks). Vessels solitary to multiples of 2-3. Clusters sometimes present. Fibers thin to thick walled. Axial parenchyma aliform with short confluences, some confluences also marginating the rays (arrows). C. Foraminate perforation plate in wide vessel, as seen in transverse section. D. Longitudinal tangential section. Parenchyma cells storied, with 2-(3-)cells per parenchyma strand. Rays 3-4 cells in width. E. Longitudinal tangential section. Note sheath cells present (arrows) and the septate fibers. F. Longitudinal radial section. Homocellular rays, with procumbent cells only. G. Heterocellular rays, with body composed of procumbent cells and a row of marginal square to upright cells. Scale bars: A-B, D = 400 µm, C = 100 µm, E-F = 300 µm, G = 150 µm. 

Bark anatomy of Astianthus. Secondary phloem. Non-stratified phloem (Figure 4A). Conducting phloem with sieve tubes solitary or in radial multiples of 2-4 (Figure 4C). All sieve plates simple, on a transverse wall (Figure 4E). Sieve tube area 436 ± 136 µm², diameter 24 ± 13 µm, and sieve element length of 259 ± 24 µm. One companion cell lying on the corner of the sieve tube (Figure 4C) or sometimes with two companion cells lying on opposite sides of the sieve tube (Figure 4C), companion cells in strands of more than two cells. Parenchyma constituting the ground tissue (Figure 4A, C), parenchyma strands with 2-4 cells. Course of rays straight (Figure 4A). Ray width, height and composition equal to that of the wood (Figure 4E). Ray dilatation seemingly absent (Figure 4A). Sclerenchyma composed of fibers only, diffuse, either solitary or in multiples of two (Figure 4A, C-D), with a polygonal shape (Figure 4C), differentiating close to the cambium (Figure 4C). Axial parenchyma and sieve tube elements storied. Non-conducting phloem marked by sieve tubes and companion cells empty, collapsed. Dilatation phenomena practically restricted to cell enlargement, with not much cell division in both axial and ray parenchyma. No further sclerification.

Figure 4 Secondary phloem of Astianthus and Campsis. A, C-E. Astianthus. B. Campsis. A. Secondary phloem non-stratified, diffuse fibers scattered across the entire tissue. Course of rays straight. Transverse section (TS). B. Secondary phloem non-stratified, with diffuse fibers scattered across the entire tissue. Ray course slightly undulated. TS. C. Sieve tubes in radial multiples of 2-4 common, diffuse fibers, differentiating close to the cambial region. Either one companion cell laying on one side of the sieve tube (lower arrow), or two companion cells, lying on opposite sides of the sieve tube (upper arrow). TS. D-E. Longitudinal tangential section. Fibers isolated, tapering (arrows). E. Sieve tube elements with simple, slightly inclined sieve plates (arrows). Scale bars: A-B = 300 µm, C-D = 200 µm, E = 100 µm. 

Periderm. Rhytidome present, with many reticulate periderms. New periderms forming inside the secondary phloem, and enclosing large amounts of nonconducting phloem. Phellem cells evenly thin walled, non-stratified. Phelloderm cells are parenchymatous and thin walled (1-3 cell layer). No mineral inclusions recorded.

Discussion

Phylogenetic placement of Astianthus. The phylogeny of the Bignoniaceae reconstructed here indicates that Astianthus falls within Tecomeae s.s. With originally 12 genera, ca. 70 species and Pantropical distribution, Tecomeae s.s. is one of the most diverse tribes of the Bignoniaceae both in terms of morphology and distribution, with members ranging from latitudes 40^o N to 40^o S (^{Olmstead 2013}) in Africa, Asia, the New World, and Oceania (Figure 5). Our analyses support Astianthus as sister to Campsis Lour., a lianescent genus with two species, one in eastern North America and one in China (^{Fischer et al. 2004}). As currently circumscribed, tribe Tecomeae s.s. includes three main clades (Figure 5): the first clade includes the Andean herb Argylia D.Don, which is sister to the rest of the Tecomeae s.s; the second clade includes predominantly Neotropical species, with Astianthus and Campsis (except for the Chinese Campsis grandiflora (Thunb.) Schumann); the third clade consists of the rest of Tecomeae s.s., with members across the Neotropics, Africa, and Asia-Pacific (^{Fischer et al. 2004}, ^{Olmstead et al. 2009}).

Figure 5 Phylogeny of Tecomeae s.s. with Astianthus highlighted in red. Feature comparisons. Geographical Occurrence (Africa, Asia, NW=New World, and Ocea = Oceania); Plant Habit (herb, liana, shrub or tree); Leaf Type (S = simple, PC = pinnately compound, PaC = palmately compound); Storied Structure (present or absent); Ray Composition (heterocellular and/or homocellular); Axial Parenchyma Type (PC = paratracheal confluent, SP = scanty paratracheal). NA = Not Applicable, plant without secondary growth. ? = unknown. 

The placement of Astianthus within the same tribe as Tecoma corroborates ^{Gentry’s (1992)} initial proposal that the leaf similarities between Astianthus and the Catalpeae genus Chilopsis represented a convergence to their riparian habit rather than an evidence of relatedness. On the other hand, the floral and fruit similarities shared between Astianthus and Tecoma were shown to corroborate phylogenetic findinds and earlier hypotheses of ^{Gentry (1992)}. Both genera share many species with yellow flowers, a cupular, 5-dentate calyx, and linear capsular fruits (^{Fischer et al. 2004}). Tecoma is composed of 14 species of shrubs to small trees distributed in tropical America from the Andes to Arizona (^{Gentry 1992}, ^{Fischer et al. 2004}). Most Tecoma have pinnately compound leaves, but the genus contains also some species with simple leaves, such as Tecoma castaneifolia (D. Don) Melch. from Ecuador, Tecoma tanaeciiflora (Kranzlin) Sandwith from Bolivia and Peru, and some specimens of T. weberbaueriana from Peru and Ecuador. Furthermore, nearly all species with pinnately compound leaves of Tecoma usually have simple leaves at the base of all branches (^{Gentry 1992}).

Wood and bark anatomy of Astianthus in relation to other Bignoniaceae. Astianthus shares many wood anatomical features with other tree members of the Bignoniaceae, such as the paratracheal parenchyma with a tendency to confluences, radially thick-walled fibers delimiting growth rings, short rays, a straight grain, and rare crystals (Table 3, Figure 5; ^{Pace & Angyalossy 2013}, ^{Pace et al. 2015a}, ^{Gerolamo & Angyalossy 2017}). The presence of foraminate perforation plates in wide vessels is not found in all Bignoniaceae but is scattered in at least eight different distantly related lineages across the entire family. This feature seems to be related to species growing under strongly seasonal rain regimes (^{Pace & Angyalossy 2013}), a hypothesis that remains to be tested.

Considering less common anatomical attributes, Astianthus would still be a good fit in at least three different Bignoniaceae major clades: the Paleotropical clade, the Tabebuia alliance, and Tecomeae s.s. (Table 3). However, based on the Neotropical distribution, Astianthus is best placed in the Tabebuia alliance or Tecomeae s.s.. Members of both tribes can have a storied structure (although this feature is more common in the Tabebuia alliance) and homo to heterocellular rays (^{Pace et al. 2015a}). However, the combination of these two features in addition to the presence of septate fibers is found exclusively in Tecomeae s.s. (Table 3), supporting the phylogenetic placement suggested by the molecular data. The most notable differences between Astianthus and some members of Tecomeae s.s. (Figure 5) are likely associated to the difference of habits. Many of the genera of Tecomeae s.s. include lianas that seem to converge in a reduction in the wood axial parenchyma (^{Pace & Angyalossy 2013}), contrary to what is found in lianas from other plant families (^{Angyalossy et al. 2015}). In addition, in lianas in general the rays tend to become more heterocellular, similarly to what was seen in other Bignoniaceae, especially in the lianescent tribe Bignonieae (^{Pace & Angyalossy 2013}).

The morphological similarity between Astianthus and members of the North American tribe Catalpeae is not mirrored by the wood anatomy. Members of Catalpeae are marked by a heartwood with abundant tyloses, a non-storied structure, and vessel to ray pits simple to slightly bordered (^{Pace et al. 2015a}). On the other hand, Astianthus lacks tyloses, has storied axial parenchyma, and distinctly bordered vessel to ray pits.

The bark anatomy provides further support for the inclusion of Astianthus in Tecomeae s.s. Virtually all Bignoniaceae species studied thus far show a stratified bark, with clear fiber bands alternating with axial parenchyma and sieve tubes, regardless of the habit, ecological factors, or distribution (^{Roth 1981}, ^{Pace et al. 2011}, ^2015b). The single exception to this rule is Campsis, which emerged as sister to Astianthus, with whom it shares scattered single fibers across the entire phloem (^{Evert 2006}, Figure 4B), a potential synapomorphy of this clade. This finding corroborates previous assumptions that the bark anatomy carries a strong phylogenetic signal in the family, independently of the habit, aiding the delimitation of major clades within the family (^{Pace et al. 2015b}). Other bark features of Astianthus such as the presence of sieve tubes in radial multiples, axial parenchyma as a background tissue, a seemingly absent ray dilatation by cell divisions, and a reticulate rhytidome are more widespread in the family (^{Roth 1981}, ^{Pace et al. 2015b}).

In conclusion, our phylogeny reconstruction based on three plastid markers (ndhF, rbcL, and trnL-F) indicates that Astianthus is nested within Tecomeae s.s.. This placement is further supported by the non-stratified bark, scattered bark fibers, storied axial parenchyma, homo and heterocellular rays co-occurring, and septate wood fibers. These results show the importance of combining in-depth studies of morphology and anatomy with molecular phylogenetic data for an improved understanding on plant diversification, especially in the tropics. The placement of Astianthus within Tecomeae s.s. further supports a neotropical origin for the tribe.

Supplementary material

Supplemental data for this article can be accessed here: https://doi.org/10.17129/botsci.2779

Supplementary material 1

Supplementary material 2