Phylogenetic Relationships
Results of the current MrBayes and ML analysis (Fig. 1) of one nr (ITS) and three cp (matK, ndhF and trnL-trnF) markers for the most part show the same results for the evolutionary relationships among and within all genera included when compared to the ML and Maximum Parsimony analyses of other studies (e.g., [1], [22], [23], [51], [52]). Since the evolutionary relationships among clades and the placement of genera within Polygonaceae are congruent across these studies, these results will not be reiterated here and only differences are discussed briefly. For example, the placement of Gilmania luteola with respect to Pterostegia drymarioides differs from Kempton's recent study [53] on Eriogonoideae. In Kempton's analysis, G. luteola is placed as sister to all other Eriogoneae, and Pterostegieae (including P. drymarioides and Harfordia macroptera) is sister to that clade with both relationships well supported. In our study and in previous analyses G. luteola branches before P. drymarioides and Eriogoneae with good support (1.00 PP/100% BS) and Burke and Sanchez [52] include Pterostegia in Eriogoneae. Since Kempton's [53] taxon sampling for Eriogonoideae and in particular Eriogoneae is much denser, we defer to her results. Furthermore, in Polygonoideae, the addition of more data for Atraphaxis resolved its position as sister to Polygonum (0.92 PP/87% BS), which is a novel result.
While most subclades within Muehlenbeckia receive good to moderate support, the relationships among these clades are not clear from the MrBayes and ML analyses, and more data are necessary to clarify this. For the most part, relationships within Muehlenbeckia are consistent with results from previous studies (e.g., [22]). In Schuster et al.'s [22] study, most species of Muehlenbeckia that occur in Australia formed a clade (except M. axillaris and M. tuggeranong), albeit with weak bootstrap support. The current MrBayes analysis shows similar results, except that M. adpressa may be included in clade x along with M. tuggeranong, M. axillaris and other species from New Zealand. Clade y-a includes a species pair of the tropical M. arnhemica from northern Australia and M. zippelii from north eastern Australia and New Guinea, and another species pair formed by M. diclina from southern Australia and M. rhyticarya from the East coast. The relationship of these two sister pairs shows a pattern observed in other groups of Australian plants, such as Eucalyptus L′Hér and Jacksonia Rees [54], [55] as well as birds including Melithreptus honeyeaters [56] and fairy wrens [57]. There appears to be a deep split between a Monsoon group from the Northern Territory and an East/South Coast group, which may have once been separated by the Carpentarian barrier [54]–[57]. The second clade (y-b) formed by Australian species of Muehlenbeckia in the MrBayes analysis includes M. gunnii from southern Australia and Tasmania as well as M. costata and M. gracillima from the East Coast. Clade z is always recovered with good support and is composed of a well-supported subclade formed by all Central and South American species sampled, which is sister to M. australis from New Zealand.
Age Estimates and Fossil Calibrations
These are the first age estimates for clades of the buckwheat family Polygonaceae. Our findings are based on dating methods using a relaxed molecular clock model calibrated with one leaf fossil of Muehlenbeckia and six pollen fossils of this genus and other Polygonaceae as implemented in Beast. Using fossils for calibration is not an easy task (e.g., [58] and references therein), but it remains the best dating method currently available when care is taken with the calibration process [59], [60]. Our results will allow for further hypothesis testing in a historical biogeographic context, although there are relatively large time span errors (Fig. S1).
Results from this study indicate that Polygonaceae likely diverged much earlier than previously thought (55.8–70.6 Myr ago) with estimated mean ages of 110.9 Myr for the exponential/lognormal and 118.7 Myr for the uniform analyses and comparatively early with respect to other eudicots (125 Myr ago). Taking the 95% highest posterior density values into account, the age estimates range from 90.7–125 Myr (Fig. 2). Given that eudicots are thought to have emerged approximately 125 Myr ago, this is relatively old for a group in the superasterids (including Asteridae, Caryophyllales and Santalales [61]). Eudicot (tricolpate) pollen appears in the fossil record about 125 Myr ago [33] and this date is well accepted based on the presence of monocolpate pollen and spores in earlier stratigraphic layers [35]. It stands to reason though that eudicots could be older than 125 Myr, because they likely originated before the massive and abrupt appearance of tricolpate pollen in the fossil record. To our knowledge, no other studies discuss the age of Polygonaceae specifically, but several authors give estimates for the age of Caryophyllales. Ages of 99–102 Myr for crown Caryophyllales [35], 94.2–94.5 and 110.7–111.3 Myr for their crown and stem ages respectively [36], 104–111 Myr [62] and approximately 101 Myr [61] are given. Different data, analytic and fossil calibration approaches utilizing one to many fossils were used for these studies and they are therefore not necessarily comparable, but their results indicate a range of 94–111 Myr for the emergence of Caryophyllales.
The fossils used to calibrate the trees in these studies are for the most part macrofossils. Only one macrofossil, a Muehlenbeckia-like leaf has so far been reported for Polygonaceae [41], while several more calibration points are available when fossil pollen is taken into account. Thornhill et al. [47] argue that pollen fossils have several advantages over macrofossils due to the durability of sporopollenin and because they are stratigraphically and temporally vastly more abundant. Therefore, the probability of fossil pollen indicating a date closer to the actual origin of a group is higher. A weak point for pollen fossils is the limited availability of morphological characteristics in some groups [40 and references therein], [47]. In Polygonaceae however, pollen morphology is a character of potentially great phylogenetic value [11], [63]–[65]. For example, in Polygonum the ektexine clearly differentiates the four recognized sections in the genus. Persicaria, which had been included in Polygonum until recent molecular analyses showed that it is not closely related to this group [66]–[68], has a rather different pollen type as well [11], [63]–[65]. It is also important to note that one character that supports the segregation of Duma from Muehlenbeckia [22] is that they have a completely different pollen morphology as evidenced by Scanning Electron Microscopy data. While Muehlenbeckia has a punctate-striate pollen morphology, Duma has a faveolate pollen surface with micro-spinules [69], [70]. This supports the inclusion of fossil pollen data in our analyses. Using pollen fossils allows for more calibration points, which estimates rate heterogeneity among lineages better and should result in more accurate age estimates [47]. Thornhill et al.'s [47] results indicate that calibrations with additional fossil pollen dates yield older estimated ages compared to analyses dated with macrofossils alone, and this might explain our comparatively old age estimates for Polygonaceae with respect to previous analyses of Caryophyllales.
In addition, in our results, age estimates are consistently older for the uniform than for the exponential/lognormal analysis. Other authors [40] using a similar calibration scheme also found that exponential priors resulted in younger ages than analyses using uniform priors. This is not unexpected, because in our calibration scheme the exponential/lognormal priors gave a much smaller probability to the maximum age of 125 Myr than the uniform priors. In our exponential/lognormal prior calibration, the mean probability distribution was at the older age boundary of the fossil, whereas in the uniform calibration the probability for ages ranging from the maximum to the minimum age was the same. This was done to give a higher probability to ages older than the fossil find date for the exponential/lognormal analyses (see Material and Methods for more explanation). Setting the mean age to a date closer to the lower age boundary could have potentially resulted in slightly younger age estimates for the exponential/lognormal analyses. However, the time span between offset (younger fossil age) and mean (older date) overall only differed between 2.7 and 9.7 Myr (Table 3), so it is unlikely that this would have resulted in a considerably younger age of Polygonaceae. Interestingly, overall variation of ages was similar in the exponential/lognormal and uniform analyses (33.4 vs. 33.1 Myr respectively for clades shown in Table 4 and Fig. S1).
Biogeographic Hypotheses
With respect to the historical biogeography of Polygonaceae, Schuster et al. [51] noted that the family might have its origin in Africa, because the African Symmeria and Afrobrunnichia likely are sister to all other members of the family [23], [52]. The difficulty with testing this hypothesis is that the position of Afrobrunnichia is uncertain and strongly varies with taxon sampling and genetic markers used as does the position of Symmeria when Afrobrunnichia is excluded. Therefore, we decided to exclude these two species from our analyses. Until more data for these important African species are available, we can only develop hypotheses about the historical biogeography of Polygonaceae.
For Polygonaceae we here propose a working hypothesis, which involves either an African or a Gondwanan ancestor that gave rise to an American and Caribbean lineage (Eriogonoideae) and a second, widespread lineage that mostly occurs in the Northern Hemisphere (Polygonoideae). The question is whether diversification of the two main clades Eriogonoideae and Polygonoideae can be explained by vicariance or LDD. If the African Afrobrunnichia and Symmeria with a disjunct distribution in Africa and South America are indeed sister to the rest of the family vicariance seems somewhat plausible for Eriogonoideae, because the age estimates of 97.8/105.5 (78.2–122.5) Myr (Table 4) fit the time frame for the separation of South America from Africa 119–105 Myr ago [71]. Clades within Eriogonoideae indicate a complex pattern of dispersal events between Central and South America, the Caribbean as well as western and eastern North America. The disjunction of the South American and African Symmeria will require further testing to say more about vicariance or LDD patterns of Eriogonoideae.
Polygonoideae may have an even more complex history, because they include several large clades with a worldwide distribution (Persicaria, Polygonum and Rumiceae). Within Polygonoideae, Knorringia sibirica from Central Asia and Yunnan is always indicated as sister to all other members of Polygoneae (Fig. 1). Within Polygoneae, the split between the mainly Australasian Muehlenbeckia and its closest relative Fallopia is dated at 41.0/41.6 (39.6–47.8) Myr. Most extant species of Fallopia occur in temperate Asia (mainly China, Japan and Korea) although some species are widespread due to anthropogenic factors [72]. Reynoutria, another genus from temperate Asia, is sister to Fallopia + Muehlenbeckia. It is plausible that the ancestor of Muehlenbeckia could have spread to Australia and/or New Zealand from temperate Asia, because there is evidence for exchange of taxa between Asia and Australia in the Miocene [54], [73], [74]. The presence of Muehlenbeckia in Oceania may be explained by stepping-stone dispersal of its ancestor from Asia (maybe via New Guinea). It should be noted that the extant Fallopia and Reynoutria are not native to Australia and New Zealand and that Muehlenbeckia is not extant in temperate Asia.
Alternatively, one could argue that the origin of Muehlenbeckia was a vicariant event, in which its ancestor rafted on a Gondwanan fragment such as India or Australia, because India made contact with Asia approximately 43 Myr ago and Australia is thought to be isolated only since 35–28 Myr ago [71], [75]. However, by definition, a vicariance explanation for the diversification of Muehlenbeckia implies that the clades formed by species of this group which are found in New Zealand (clade x), Australia (clades y-a, y-b) and South America (subgroup of clade z) were present on all of these constituent land masses before the breakup of Gondwana 95–30 Myr ago [76]. This is unlikely, since diversification of the crown clade of Muehlenbeckia is estimated at 20.5/22.3 (14.2–33.5) Myr (Fig. 2, Table 4), which is too young to satisfy the vicariance hypothesis. At 14.4–33.5 Myr, the 95% highest posterior density values of the uniform prior analysis for Muehlenbeckia (Fig. 2) is close to the breakup age of Gondwana, but all clades within Muehlenbeckia with disjunct distributions across e.g. New Zealand and South America are too young for vicariance.
For example, the chronogram (Fig. 2) shows evidence of LDD from New Zealand to Australia in Muehlenbeckia's clade x. Muehlenbeckia tuggeranong, which only occurs in Australia, is nested within clade x among species from New Zealand. Vicariance is unlikely, because crown clade x is estimated to be 13.3/14.6 (6.4–24.1) Myr old and Australia has been isolated since 35–28 Myr while New Zealand has been cut off since approximately 80–56 Myr. Long distance dispersal is likely also the most parsimonious explanation for the diversification of Muehlenbeckia, because the South American species of Muehlenbeckia are sister to M. australis, which is native to New Zealand and Norfolk Island in clade z. The age estimates for the split of M. australis and the South American clade at 12.5/13.1 (6.0–22.2) Myr is younger than the isolation ages of South America and New Zealand. South America has been isolated since 30 Myr [71], [76] and New Zealand is thought to be isolated since 55.8 Myr [40], [77], though dates around 80 Myr are more commonly found in the literature (e.g., [71], [75]).
The date for the diversification of the Central and South American subclade in clade z (M. tamnifolia, M. tiliifolia, M. urubambensis, M. volcanica) at 7.9/8.4 (3.2–15.3) Myr correlates well with age estimates for a second uplift of the Eastern Cordilleras of the Northern and Central Andes [78], [79]. Mountain building may also have influenced climatic and edaphic factors, since high mountain ranges create a barrier to precipitation [78]. Climatic and edaphic factors, landslides and erosion could have created a mosaic of microhabitats that afforded new possibilities for diversification [80], [81] as is thought to have happened in the species-rich Cape Floristic Province in South Africa [82]. Radiation events during the uplift of the Andes have also been reported in other groups such as Chloranthaceae [83], Ericaceae [84], Fabaceae [85], Rubiaceae [78], Lepidoptera [86] and hummingbirds [87].
Most diversification events in Muehlenbeckia occurred after 20.5/22.3 (14.2–33.5) Myr ago, which correlates with the aridification and cooling of Australia in the Miocene [88]–[91]. Aridification may have resulted in an increase in the frequency of bushfires. Several of the Australian species of Muehlenbeckia (in clades y-a and y-b) are adapted to fire [22]. For example, the fire-ephemeral M. diclina grows in scleromorphic mallee, a habitat characterized by stands of Eucalyptus L'Hér., Acacia Mill. and Triodia R.Br., which are adapted to burns. The fire-adapted species of Muehlenbeckia are difficult to classify as facultative or obligate seeders, because they usually senesce before another burn. However, they do respond strongly to fire cues for germination (Peter Clarke, personal communication). In contrast, the flora of New Zealand has few fire-adapted species and the Australian species that belong to the mixed Australian/New Zealand clade (x), such as M. tuggeranong and M. axillaris are also not adapted to fire. Radiation of Australian groups in the Miocene is observed frequently, and the crown clade of Duma also diversified within this time frame around 21.0/24.2 (8.8–38.8) Myr ago. Allocasuarina L.A.S. Johnson, Banksia L.f. [75], some Elaeocarpaceae Juss. [74], Eucalyptus [92] and some scleromorphic groups of Fabaceae Lindl. [93] also radiated during the Miocene in Australia. Similar adaptations to disturbance that may be caused by fire have occurred in other ecosystems, such as the South African fynbos, the chaparral in California, the Chilean matorral and the South American cerrado [37], [38], [94]–[96].
To summarize, because the sister genera of Muehlenbeckia mainly occur in temperate Asia, which has never been considered part of Gondwana, and because clades of Muehlenbeckia with disjunct distributions across e.g. New Zealand and South America are younger than when these landmasses broke apart, LDD rather than vicariance is likely the main driver for diversification within this group. The crown clade of Muehlenbeckia diversified 20.5/22.3 (14.2–33.5) Myr ago, and this is younger than the isolation dates of Australia (35–28 Myr), Antarctica (32–30 Myr), New Zealand (80–55 Myr) and South America (32–30 Myr). Our age estimates for Polygonaceae and clades such as Muehlenbeckia are a starting point for further testing of its phylogeny in a biogeographic context. This will give more insights about the origin of Polygonaceae and the diversification of specific clades within this diverse and widespread family.