Plant Material and Pollen Characterization
Inflorescences of C. amboinicus, C. barbatus and C. neochilus were collected at the Medicinal Plant Garden of the Federal University of Lavras - Brazil. The material was fixed and maintained in ethyl alcohol:propionic acid (3:1) and stored at -20°C. Voucher specimens were deposited at the Research Center for Chemistry, Biology, and Agriculture (CPQBA) and the State University of Campinas Herbarium (UEC) in São Paulo, Brazil, under the following records: C. amboinicus (CPQBA 364), C. barbatus (UEC 121403), and C. neochilus (CPQBA 1388).
Slides were prepared according to the acetolysis method of Erdtman (1952). A total of fifty pollen grains were examined using an optical microscope (Leica DMLS) equipped with a Nikon Digital Sight DS-Fi1 micro-camera. The variables evaluated were: polar diameter (P); equatorial diameter (E); colpus diameter (D); apocolpium side (d); colpus width (CW); and colpus length (CL) and thickness of nexin and sexin. The pollen shape was defined by the P/E ratio. The measurements were carried out with the aid of Image Tool v. 3.0 software.
In scanning electron microscopy analyses, anthers were subjected to three consecutive ten-minute washes in cacodylate buffer, followed by postfixation in 1% osmium tetroxide for a minimum of one hour at room temperature. Subsequently, the material underwent three rinses in distilled water, followed by dehydration in a series of acetone concentrations (25%, 50%, 75%, 90%, and 100%), and was then transferred to the critical point apparatus (CDP 030). The pollen grains were extracted from the anthers on stubs, which were submitted to the sputter coater (SCD 050), and finally analysed on a LEO EVO 40 scanning electron microscope.
Cluster Analysis
A palynological survey was conducted, covering 14 species of Coleus available in the literature (Halbritter 2016; Khalik 2016; Smitha et al. 2018) and three from this study (Coleus amboinicus, C. barbatus and C. neochilus). A dendrogram was constructed based on morphometric data (see Table 2). The analysis utilised the Euclidean distance method implemented in R software (R Core Team 2023), along with the factoextra (Kassambara and Mundt 2020), dendextend (Galili 2015), and igraph (Csardi and Nepusz 2006) packages.
Phylogenetic Inference
Phylogenetic reconstruction was performed using the ITS, matK, rbcL, rps16 and trnL-F markers. Sequences were obtained from the GenBank platform (Table S1). Of the 17 Coleus species previously sampled in the dendrogram, only C. bishopianus (Gamble) Smitha & A.J. Paton, C. intraterraneus (S.T. Blake) P.I. Forst. & T.C. Wilson, C. neochilus (Schltr.) Codd, and C. fruticosus Wight ex Benth. lacked available sequences and could not be included in the phylogeny. The outgroup was represented by Alvesia clerodendroides (T.C.E.Fr.) B.Mathew, Alvesia rosmarinifolia Welw., Tetradenia fruticosa Benth., and Tetradenia riparia (Hochst.) Codd.
Alignment was performed using the MEGA program v. 2 (Stecher et al. 2020). Evolutionary model selection was determined by the Akaike criterion using the MrModelTest v. 2.3 program (Nylander 2004). The concatenated matrix was generated using the Mesquite v. 3.61 program (Maddison and Maddison 2021). Phylogenetic reconstruction was conducted through Bayesian inference using the MrBayes v. 3.2.2 program (Huelsenbeck and Ronquist 2001) available on the CIPRES platform (www.phylo.org). The following settings were applied: nchains = 4; nruns = 2; ngen = 107 and burnin = 0.25; samplefreq = 1,000.
Ancestral state reconstruction for categorical data
We assessed pollen grain shape from the perspective of marginal reconstruction of ancestral states using the maximum likelihood method. We performed this analysis in the R software (R Core Team 2023) with the packages corHMM (Beaulieu et al. 2013), phytools (Revell 2022), geiger (Pennell et al. 2014), and evobiR (Heath and Adams 2015). For the selection of the evolutionary models, we used the MK model for discrete data (Pagel 1999a,b; Lewis 2001). The Mk model can include the ‘equal rates’ (ER), in which the transitions between states occurred at equal rates, ‘All-rates-different’ (ARD), in which the transition rates (total change of state per branch) are variables, and the Symmetric Model (SYM) which considers that the transition rates are symmetrical, being used specifically for multistate data (Harmon 2019). The selection of the model for each trait considered the corrected Akaike criterion (AICc) and Akaike weight (AICw).
Phylogenetic mapping of continuous data
For continuous data, evolutionary models may follow a process of “Brownian motion”, in which the character's value changes randomly, both in direction and magnitude, over any time interval; “Early-Burst”, in which character change tends to concentrate at the base of the tree; and “Ornstein-Uhlenbeck”, which exhibits a tendency toward a central value under stabilizing selection (Harmon 2019). The evolutionary model testing for pollen size (polar diameter) was conducted using the R software (R Core Team 2023) with the geiger (Revell et al. 2023) and phytools (Revell 2022) packages. The model with the lowest AICc and highest AICw was selected for analysis. For character state mapping in phylogenetics, we utilized the ape package (Paradis et al. 2023).