Ten samples (accessions) were used for each of the 5 species of Adenocaulon (Table 1). Based on the results from previous phylogenetic analyses of Asteraceae (Panero and Funk, 2008; Ortiz et al., 2009), additional samples were used from eight genera that are closely related to Adenocaulon (Gerbera, Leibnitzia, Chaptalia, Trichocline, Brachyclados, Pachylaena, Mutisia, and Chaetanthera of the tribe Mutisieae of the subfamily Mutisioideae) and 15 distantly related genera of the subfamily (Trixis, Dolichlasium, Jungia, Perezia, Nassauvia, Acourtia, Leucheria, Proustia, and Lophopappus in the tribe Nassauvieae of the subfamily Mutisioideae; Onoseris, Urmenetea, Lycoseris, Aphyllocladus, Gypothamnium, and Plazia from the tribe Onoserideae of the subfamily Mutisioideae). We also included nine species from Wunderlichioideae and five species from Barnadesioideae. We used Acicarpha spathulata (Calyceraceae) to serve as the outgroup, following Panero and Funk (2008) (Table 1).

Total genomic DNA was isolated from silica-dried leaf materials and herbarium specimens using a Universal Genomic DNA Extraction Kit (Takara, Dalian, China). Nine gene regions, including eight plastid DNA regions (ndhF, matK, rpoB, rbcL, trnL-F, rpl32-trnL, rpl32-ndhF, and psbA-trnH) and the nuclear ITS, were employed. According to Shaw et al. (2007) and Timme et al. (2007), both rpl32-trnL and rpl32-ndhF have faster molecular evolution rate than other chloroplast regions and can provide adequate resolution in phylogenetic reconstructions of Asteraceae. The ITS data were generated from single amplifications using primers ITS4 and ITS5 (White et al., 1990). Primers used for amplification and sequencing were tabe and tabf for trnL-F region (Taberlet et al., 1991), Z1 and 1204R for rbcL (Zurawski et al., 1981), and psbA_F and trnH_R for psbA-trnH intergenic spacer (Sang et al., 1997). The primer sets for ndhF, matK, and rpoB regions were from Panero and Funk (2008), while the primers for rpl32-trnL and rpl32-ndhF were from Shaw et al. (2007). Amplified products were purified using a Qiaquick gel extraction kit (Qiagen, Inc., Valencia, California, USA) and sequenced in both directions by an ABI 3730 automated sequencer (Applied Biosystems, Foster City, California, USA). The resulting sequences were edited using Sequencher (version 4.1.4) and aligned with MUSCLE (version 3.6; Edgar, 2004), followed by manual adjustment in Se-Al (v2.0a11; Rambaut, 2002). All sequences generated for this study were deposited in GenBank under accession numbers shown in Table 1.

The nuclear and the plastid datasets were analyzed both separately and simultaneously using maximum parsimony (MP), Bayesian inference (BI), and maximum likelihood (ML). The topologies of the data sets were compared with each other to detect any incongruence. Because no significant incongruence was observed, we chose to combine the nrDNA and cpDNA data sets. For the maximum parsimony analysis we used PAUP/(version 4.0b10; Swofford, 2003). All characters were weighted equally and unordered. Heuristic searches were conducted using 100 random-taxon-addition replicates with tree bisection-reconnection (TBR) branch swapping, using MulTrees option in effect, and a maximum of 10,000 trees. Bootstrap analyses (1,000 pseudoreplicates) were conducted to examine the relative level of support for individual clades on the cladograms of each search (Felsenstein, 1985).

Models of nucleotide substitution were selected based on the Akaike Information Criterion (AIC) as determined by MrModelTest (version 2.3; Nylander, 2004). Maximum-likelihood searches and bootstrap analyses were performed on the XSEDE online computing cluster accessed via the CIPRES Science Gateway (Miller et al., 2010) using RAxML-HPC2 (version 7.4.2; Stamatakis et al., 2008). Data were partitioned into gene regions, allowing for independent parameter estimates on each partition, with branch length estimates optimized across all gene regions. The GTR + Γ model was employed for all analyses.

Bayesian inference was conducted using MrBayes (version 3.2.2, Ronquist and Huelsenbeck, 2003). The Markov chain Monte Carlo (MCMC) algorithm was run for 10,000,000 generations with one cold and three heated chains, starting from random trees and sampling one out of every 1,000 generations. The burn-in and convergence diagnostics were graphically assessed using AWTY (Nylander et al., 2008). The burn-in trees were excluded, and the remaining trees were assumed to be representative of the posterior probability (PP) distribution.

The combined nuclear and plastid sequences were used to estimate the divergence time of Adenocaulon using BEAST (version 1.8.0; Drummond and Rambaut, 2007). BEAST employs a Bayesian MCMC approach to co-estimate topology, substitution rates and node ages (Drummond et al., 2002). BEAUti was used to set criteria for the analysis, in which we applied a general time reversible (GTR) nucleotide-substitution model with Gamma + Invariant sites, gamma shape distribution (with four categories) and proportion of invariant sites. A Yule tree prior model was implemented in the analysis, with rate variation across branches assumed to be uncorrelated and lognormally distributed (Drummond et al., 2006). Posterior distributions of parameters were approximated using two independent MCMC analyses of 50,000,000 generations (sampling once for every 5,000 generations) after a 10% burn-in for each. Convergence of the chains was checked using Tracer (version 1.5; Rambaut and Drummond, 2007).

For the fossil calibrations, we followed the strategy adopted by Nie et al. (2016). The first calibration point was the Raiguenrayun cura Barreda, Katinas, Passalia & Palazzesi capitulescence from the Patagonia described as belonging to the crown Asteraceae and dated by radiometric methods at 47.5 Ma (von Nickisch-Rosenegk et al., 1999). This fossil was used as a minimum age constraint for the split between Barnadesioideae and the rest of family and with an applied lognormal prior distribution with an offset of 47.5 Ma, a mean of 1.0 and a standard deviation of 0.7. As a minimum age constraint for the crown group of the subfamily Barnadesioideae, we used the fossil pollen Quilembaypollis sp., from the latest Oligocene/earliest Miocene, dated at 23 Ma (Chiang et al., 1998). Following Bergh and Linder (2009), the 95% confidence interval for this prior lies between 16.9 and 44.1 Ma with the mean at 22.3 Ma.

We compiled distribution data for the Adenocaulon species and assigned the included taxa to respective ranges. Considering the areas of endemism in Adenocaulon and tectonic history of continents, we used four areas for the Adenocaulon taxa: (A) South America, (B) Central America, (C) North America, and (D) East Asia. Each sample was assigned to its respective area according to its contemporary distribution range.

Biogeographic inferences were obtained by applying both statistical dispersal–vicariance analysis (S-DIVA) and dispersal-extinction-cladogenesis (DEC) model calculated by Lagrange (Ree and Smith, 2008) implemented in RASP (version 3.0) using the default settings (Yu et al., 2010). A subset of 1,000 randomly selected trees from the posterior distribution output of BEAST was used and the maximum number of individual unit areas was set to two. The probability of dispersal between areas was maintained as equal.

The incongruence length difference and Shimodaira and Hasegawa (SH) tests failed to reveal significant incongruence, allowing the individual datasets to be combined. There were 12,416 characters of combined data set, of which 1,571 were variable, and 1,266 parsimony informative sites. In general, the combined data matrix provided greater resolution and stronger support for phylogenetic relationships than did the individual datasets (Figure 2).

The results from MP and BI analysis, as well as ML analysis, strongly support Adenocaulon being a monophyletic clade (PP = 100, BP = 100, PL = 100). Within Adenocaulon, the Central American species, A. lyratum, diverged first, followed by A. chilense being sister to a clade including the remaining species from North America and East Asia albeit with low support (PP = 64, BP = 54, PL = 85). The two species found in East Asia, A. himalaicum and A. nepalense, form a clade with high support values (PP = 100, BP = 93, PL = 98), which is sister to A. bicolor (North America) (Figure 2).

The chronogram and results of divergence-time estimation based on the nrDNA and cpDNA data showed that the crown age of Adenocaulon was estimated at 9.05 Ma with a 95% highest posterior density (HPD) of 4.73–14.99 Ma. The stem age, that is the divergence time between Adenocaulon and its close relatives, was estimated at 14.74 Ma (with 95% HPD: 9.6–23.7 Ma). The split between the South American species A. chilense from the remaining species of Adenocaulon was at 6.92 Ma (95%HPD: 3.55–12.05 Ma), whilst the disjunction between North American A. bicolor and East Asian species at 4.79 Ma (95% HPD: 2.11–8.68) (Figure 3). S-DIVA analysis demonstrated that Adenocaulon may have originated in the tropical America and then dispersed to North America and East Asia (Figure 3).

Our study showed that Adenocaulon is monophyletic within the Mutisieae (PP = 100, BP = 100, PL = 100) and is sister to a clade including Gerbera, Leibnitzia, Chaptalia, Trichocline, and Brachyclados (Figure 2). The genus is characterized by a distinct set of synapomorphy including basal constriction of anther appendages, small anthers, disciform capitula, and typical glandular achenes without pappus (Bittmann, 1990). The placement of Adenocaulon has long been disputed among numerous taxonomists, going as far as calling it an anomalous genus, and being unplaced in the tribe (Bremer, 1994). Based on our broad analysis of the early diverging clades of Asteraceae, Adenocaulon is confirmed to belong to the tribe Mutisieae (Mutisioideae) and to be sister to the Gerbera-complex (PP = 100, BP = 79, PL = 76) (Figure 2).

Our results indicate that the Central American A. lyratum diverged early. This species differs from the other four species in a stem winged by decurrently leaf bases, lyrate-pinnatifid sessile leaves and distinct club-shaped achene. Representatives of A. chilense from southern Chile and Argentina show petiolate entire leaves in a rosulate order. The pistillate flowers have a distinct bilabiate corolla and the scabrate pollen exhibit two layers of columellae with a larger inner one. The other three species from the Northern Hemisphere form a clade with high support. They share common features, e.g., leaves concentrated at the base, leaves triangular with cordate leaf base, heart- or kidney-shaped, with dentate or sinuate margin. Within this clade, the North American A. bicolor is sister to a clade with the two Asian species A. bicolor and A. himalaicum; these two species can be easily distinguished from each other by the glandular hairs: the former being pin-shaped, occurring on both vegetative and synflorescence branches, whereas the latter is nail-shaped but restricted to the synflorescence branches (Bittmann, 1990). A. nepalense is a recently described new species (Bittmann, 1990) endemic to Bhutan, Nepal, and India, and having distinct features such as winged by decurrently leaf-based stems and the bearing whip hairs achenes.

Our biogeographic reconstruction suggests that Adenocaulon may have originated in Central America and then dispersed to North America and South America during the Miocene (14.74 Ma, 95% HPD: 9.6–23.7) (Figure 3). The dispersal of biota between Central America, North America and South America (commonly called the Great American Biotic Interchange) is one of the most significant events in neotropical biogeography. The Great American Biotic Interchange was caused by the closure of the Isthmus of Panama ca. 3 Ma (Burnham and Graham, 1999; Cody et al., 2010), resulting in the exchange of plants and animals between the two once separated continents. However, some authors suggest that an earlier land connection made possible the exchange of biota within the Americas. A land bridge may have existed between North and South Americas 3–7 Ma (Bermingham and Martin, 1998). Moreover, some plants could have successfully crossed the Isthmus of Panama before it closed (Cody et al., 2010; Bacon et al., 2015). Here, we propose that Adenocaulon may have dispersed into South America from Central America during the Miocene via the land bridge link. Hypotheses of long distance dispersal events have gained popularity in the last decade as a way of explaining many disjunct distributions of plant lineages (Givnish et al., 2004; Popp et al., 2011; Nie et al., 2013).

Dispersal of Adenocaulon may have been enhanced by birds due to its special fruit structures. The prominent glandular hairs on the cypselae are very sticky and cling readily to fabrics, fur, and feathers. The intercontinental disjunction within the range of Adenocaulon is possibly a result of dispersal of cypselae on the feathers of birds. The bird dispersal pattern may also account for the intracontinental disjunction of A. bicolor (in the Pacific Northwest, the Black Hills, and the northern Great Lakes region). Thus, we do not postulate vicariance as a factor that shaped the disjunct pattern of Adenocaulon.

The intercontinental dispersal of plants in the Northern Hemisphere has been a focus of biogeography for a long time (Donoghue et al., 2001; Donoghue and Smith, 2004). During the Cenozoic, two routes for plant dispersal between the New World and the Old World were the Bering Land Bridge and the North Atlantic Land Bridge (Milne and Abbott, 2002; Wen et al., 2010). These two links have played significant roles in the formation of the modern flora of the Northern Hemisphere but happened in different geologic times (Tiffney, 1985a,b). The Bering Land Bridge is assumed to have existed from the Paleocene to Miocene (Tiffney and Manchester, 2001) but later this dispersal route was blocked due to global cooling (Milne, 2006). We suggest that Adenocaulon most likely migrated from North America to East Asia via the Bering Land Bridge. The estimated divergence time between the North American and East Asian groups is ca. 11.91 Ma, which coincided the cooling of Northern Hemisphere in which the continuous distribution of the Boreotropical flora was disrupted (Tiffney and Manchester, 2001). Our results suggest Adenocaulon may have originated in Central America, dispersed into North America and further into East Asia via the Bering Land Bridge, and finally reached the Himalayas.

Our results suggest that the biogeographic history of Adenocaulon species have been influenced by dispersal events, which could be facilitated by their unusual floral features. The pappus and the fruit trichome can prevent desiccation, and other characteristics of the family lack in Adenocaulon. Instead, the fruits of Adenocaulon are covered by glands with a sticky secretion. An umbrella-like “infructescence” emerges from the otherwise short plants usually embedded in other under-story vascular plants or mosses, and exposes the sticky fruits. Pervious researches indicated that viscid fruits can adhere by bird feathers (Carlquist, 1967, 1983). Also, the possibility that birds could have been the dispersal vectors for Adenocaulon fruits was proposed by Bittmann (1990) to explain some of the extant distributions of the genus. Alternatively, successful over-water dispersal events could be postulated for Adenocaulon; the oily fruit cover is suggested to aid in flotation, as was the case for other families. Thus, the absence of pappus in Adenocaulon does not mean that this genus lacks mechanisms for dispersal. The sticky fruits, with the “infructescence,” can act as an alternative dispersal mechanism involving several potential dispersal vectors.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.