Transitional changes in arborescent lignophytes at the Devonian–Carboniferous boundary

Abstract: It is usually considered that after the extinction of the Devonian tree Archaeopteris, no new arborescent lignophytes were established before the late Tournaisian. A reassessment of this pattern is presented here based on a three-fold approach: a re-evaluation of the taxic diversity of Tournaisian lignophyte trees based on descriptions of new plants from palaeotropical latitudes, a study of the patterns of phenotypic changes occurring among early lignophytes using a principal coordinate analysis and a phylogenetic analysis of the affinities of the arborescent taxa. The best supported results indicate that a substantial taxonomic and phenotypic diversity of arborescent lignophytes was already established in the first part of the Tournaisian, including some taxa that persisted until the Serpukhovian. Two genera may have originated in the Late Devonian and crossed the Devonian–Carboniferous boundary. Fewer originations and a decrease in phenotypic diversity occurred in the Viséan. The phenotypic distinctiveness of tree stems compared with those of other growth forms in the lignophytes is assessed. We propose a scenario in which the presence of lignophyte trees is continuous across the Devonian–Carboniferous boundary, with arborescent taxa distinct from Archaeopteris already present in the latest Devonian, possibly in upland floras, and diversifying significantly soon after the Devonian–Carboniferous boundary. Supplementary material: A list of taxa, characters and matrices used in the principal coordinate analysis and phylogenetic analysis is available at http://www.geolsoc.org.uk/SUP18447.

An important aspect of current research on the colonization of continental surfaces by plants (terrestrialization) addresses the evolution of the first forest ecosystems and the environmental conditions of their development. Two key steps need to be considered: (1) the appearance of the tree habit during the Middle Devonian, followed by the formation of the earliest forests in lowland areas during the Middle and Late Devonian (Mintz et al. 2010, and references therein); (2) the colonization of uplands by trees, first evidenced in tropical regions during the Pennsylvanian (Falcon-Lang & Scott 2000;Falcon-Lang & Bashforth 2005). Both steps probably had significant effects on the climate by modifying the carbon cycle through changes in soil structure, sedimentological processes, and amount of carbon buried in sediments (Beerling & Berner 2005;Davies & Gibling 2010). We are focusing here on the diversification of arborescent lignophytes (progymnosperms and seed plants) during the Mississippian to better understand the transition between the Devonian and Carboniferous forest ecosystems.
The fossil record indicates that the evolution of a variety of hydraulic and mechanical novelties during the Devonian (Mosbrugger 1990; Rowe & Speck 2004) was followed by the development of arborescent taxa in all the major groups of vascular plants (i.e. lycopsids, ferns sensu lato, sphenopsids and lignophytes). These early trees presented a wide range of morphologies and reproductive strategies, and it can be hypothesized that they all interacted differently with their environment. The spread of lignophyte trees with their vegetative body characterized by the production of a significant biomass, the possession of true leaves and deep root systems, is thought to have affected the Devonian ecosystems at a global scale (Algeo & Scheckler 1998;Beerling & Berner 2005;). In addition, the advent of the seed habit during the Late Devonian allowed lignophytes producing seeds to colonize dry or disturbed environments and act as major contributors to plant-environment interactions during the last phase of the terrestrialization.
The colonization of continental surfaces by arborescent lignophytes during Devonian and Carboniferous times is currently viewed as a four-stage process, as follows.
Stage 1 (Givetian and Late Devonian). This stage corresponds to the worldwide colonization of lowlands by the archaeopteridalean progymnosperms, which formed the first dense shady forests on Earth (Beck & Wight 1988). Archaeopterids tolerated a wide variety of soil conditions but prevailed in well-drained sites. They occurred in both distal (Algeo & Scheckler 1998) and proximal floodplain environments as recently evidenced by the report of in situ middle Givetian archaeopterid-like stump casts in the Appalachian Basin (Mintz et al. 2010). In contrast, the first seed plants, which did not become well established prior to the late Famennian, are reconstructed as small bushy colonizers inhabiting disturbed habitats (Scheckler 1986;Cressler 2006;. Stage 2 (Devonian-Carboniferous boundary and Hastarian). Archaeopteridalean progymnosperms became extinct around the Devonian-Carboniferous boundary, possibly in connection with a global crisis in terrestrial plants (Streel et al. 2000). The majority of seed plants reported from deposits at the transition between the Devonian and the Carboniferous (i.e. Strunian beds of Ireland; earliest Mississippian deposits yielding permineralized plants from the New Albany Shale) and in the Hastarian localities of Thuringia and Montagne Noire (Scott et al. 1984;Galtier 1988;Galtier et al. 1988) belong to non-arborescent pteridosperm orders such as the Calamopityales, Lyginopteridales and Buteoxylales. As a result, this transitional stage has been interpreted as characterized by an open vegetation composed of shrubby, r-selected species (DiMichele & Hook 1992). A few taxa, such as Archaeopitys from the New Albany Shale, or Araucarites and Aporoxylon from Thuringia, however, indicate that some lignophytes of larger stature might have been present.
Stage 3 (Ivorian to Serpukhovian). This stage corresponds to a dramatic increase of the number of arborescent lignophyte taxa. One of them is Protopitys, a putative heterosporous progymnosperm (Beck & Wight 1988). The 10 other genera reported by  are suspected to represent seed plants based on the morphology of their leaves (Galtier et al. 1998a) and other traits of their primary body. Most of these taxa occur in well-drained environments, often in areas with volcanic activity, which suggests that they might have adapted more successfully to dry and/or disturbed conditions than other plant groups.
Stage 4 (Early to Middle Pennsylvanian). This stage sees the diversification of the earliest coniferophytes (sensu Hilton & Bateman 2006, i.e. cordaites and conifers). It corresponds to the first undisputable evidence of upland forests, although upland coniferophytes might have already been present in the end of the Mississippian (DiMichele et al. 2010;Dolby et al. 2011). These upland forests are suspected to have contributed to the Carboniferous-Permian glaciations in Gondwana through increased weathering in upland areas (Falcon-Lang & Scott 2000;Falcon-Lang & Bashforth 2005).
This sequence of events, however, is subject to a number of uncertainties that need to be taken into account. (1) There is a dearth of localities yielding plants at the Devonian-Carboniferous boundary and just above. (2) The biogeographical information is limited. The Hastarian plant localities investigated so far are concentrated in Western Europe and North America. (3) There is probably a preservational bias. Most of the plant deposits corresponding to stage 1 yield compressions, a type of fossil that is unfavourable to the recognition of arborescent lignophytes. The compressed seeds, frond remains and distal portions of branches suggesting a dominance of small-statured seed plants in the Late Devonian may not reflect the complete range of lignophyte growth forms actually present at this time.
(4) There is possibly an effect of taxonomic bias on the assessed diversity. The affinities of specimens that have not been revised recently may give an erroneous image of the actual diversity of the arborescent lignophytes. This problem is especially noteworthy for the Hastarian. For example, whether Archaeopitys from the New Albany Shale is congeneric with Archaeopteris, or with Pitus, or represents a distinct genus remains uncertain (Gordon 1935;Beck 1976). The status of genera such as Araucarites and Aporoxylon from Thuringia is likewise doubtful (Hörich 1915). (5) The affinities of the Mississippian trees are uncertain. Because the reproductive traits commonly used in phylogenetical analyses are unknown in most of these taxa, they are generally ignored in such studies and their systematic positions remain unresolved.
Therefore, many questions related to the patterns of changes affecting the arborescent lignophytes between the Late Devonian and the early Pennsylvanian are still unresolved. This paper focuses on the following problems: were arborescent lignophytes really a minor component of Hastarian (stage 2) floras? When did these Mississippian trees originate and what are their relationships with Archaeopteris and with the seed plants? Can we provide an alternative scenario for the evolution of the arborescent lignophytes around the Devonian-Carboniferous boundary? What biological and environmental factors might have contributed to the observed succession?
To address these questions a first step was to improve the information about the second stage of the succession (Hastarian), which is the least understood, and in the broadest palaeogeographical framework possible, so as to appreciate the local or global significance of the observed changes. We present here a synthesis of recent studies and work in progress that provides new elements on the diversity and distribution of arborescent lignophytes during this period. It is complemented by a statistical treatment and a phylogenetic analysis to better circumscribe the patterns of diversification between stage 1 and stage 4.

Famennian-Mississippian evolution of taxic diversity
Until the last decade, the number of arborescent lignophyte taxa described in the late Tournaisian (Tn3, Ivorian) greatly outnumbered that known from the preceeding substage of the Tournaisian, the Hastarian. Investigations recently conducted on localities of Hastarian age and situated in the palaeotropical belt have provided new records of the presence of arborescent lignophytes closer to the Devonian-Carboniferous boundary (see taxa in black in Fig. 1, in bold in Table 1). They allow a reevaluation of the patterns of changes affecting the taxic diversity of these plants between the Latest Devonian and the Early Mississippian. Information considered in the following review is summarized in Figure 1 and Table 1.
In Western Europe (Laurussian region), the youngest specimens assigned to Archaeopteris are found in latest Famennian (Strunian) localities of Ireland (Fairon-Demaret 1986, and references therein). The oldest woody axes distinct from Callixylon, the wood of archaeopteridalean progymnosperms, were described by Matten and his collaborators from Strunian deposits corresponding to the Retispora lepidophyta, probably LL, miospore zone in Ireland (Matten et al. 1976(Matten et al. , 1980Matten 1995). Wood from Kerry Head, near Ballyheigue, was first compared with the Mississippian genera Pitus and Eristophyton (Matten et al. 1976), then assigned to the 'Pityales' in following papers (Matten et al. 1980(Matten et al. , 1984. No detailed descriptions or illustrations were given. Wood from Hook Head was assigned to Pitus (Matten 1995). In one case (locality 3 of Matten 1995), compressions of Archaeopteris foliage co-occur with wood assigned to Pitus. This evidence remains doubtful, however, as the preservation of the wood did not allow us to observe the radial pitting in enough detail to rule out the presence of the grouped pits characteristic of archaeopterid wood (Matten 1995, fig. 4A, and re-observation by A.L.D. of Matten's slides). Tree trunks of Pitus primaeva recently reported from the Isle of Bute in Scotland, and suggested to be early Courceyan (Hastarian) in age (1) in Table 1), may represent the earliest Mississippian evidence of lignophyte trees in Western Europe.
Little is known about arborescent lignophytes of Mississippian age from Northern Europe. Fragmentary wood remains supposed to be Viséan from the Arkhangelsk Region in Russia are assigned to three genera, Eristophyton, Paleoxylon, and the new genus Tovoxylon (Antashchuk & Snigirevsky 2003;Orlova 2009Orlova , 2010 (2) in Table 1). Paleoxylon, a morphogenus consisting of Pitus-type wood, was first described from the Viséan of northern France (Coulon & Lemoigne 1969;Galtier et al. 1998b). The main distinctive characters of Tovoxylon consist of (1) the possession of wood tracheids showing one row of circular pits on the radial walls of tracheids; (2) radial pits not occupying the entire tracheid width and having a circular aperture; (3) rays small, one cell high only.
In North America, no new investigations were conducted on the arborescent lignophytes from the New Albany Shale since the historical works of Scott & Jeffrey (1914) and Read (1936), synthesized by Hoskins & Cross (1952). The oldest evidence of arborescent lignophytes other than Archaeopteris in the USA remains Archaeopitys, from the Falling Run Member of the  Table 1).
Thuringia (Germany) and Montagne Noire (southern France), which have long been studied for their assemblages of anatomically preserved remains, were part of European terranes whose palaeogeographical position, between Laurussia and Gondwana, is not fully resolved (Matte 2001). Apart from Archaeopitys, the oldest evidence of arborescent lignophytes  Heckel & Clayton (2006), Thorez et al. (2006), Buggisch et al. (2008) and Korn & Kaufmann (2009). Taxa represented by bold lines correspond to most recent data.  other than archaeopteridaleans has long been limited to two genera from the Hastarian deposits of Thuringia in Germany, Aporoxylon (Unger 1856) and Araucarites. Their conspecificity was suggested (Göppert & Stenzel 1888). However, differences in the size of the rays, which are all uniseriate and rarely exceed three cells in height in Araucarites but are multiseriate and 10 or more cells high in Aporoxylon, leave this question open. Our recent investigations of woody axes from the nearby locality of Kahlleite quarry confirmed the presence of Aporoxylon but also revealed the presence of two additional taxa, the putative seed plant Eristophyton and the progymnosperm Protopitys (Decombeix et al. 2005;(4) in Table 1). These two genera have a long stratigraphical extent in the Mississippian but had not been previously recorded from deposits older than the Ivorian. In addition to this occurrence, Protopitys and Eristophyton have subsequently been identified in the Hastarian Lydiennes Formation of Montagne Noire (Decombeix 2007;Decombeix et al. 2007Decombeix et al. , 2008. A new genus, Faironia, was also recognized in the Hastarian plant beds of Montagne Noire (Decombeix et al. 2006;(4) in Table 1), extending the systematic diversity of Hastarian trees in European terranes to a total of five genera.
Although not as well documented as that from Laurussia, the record of archaeopterids from the Late Devonian of Gondwana is not insignificant, with reports of compressed leafy branches and permineralized wood from Africa (Anderson et al. 1995;Meyer-Berthaud et al. 2000, 2004, South America (Berry et al. 2000) and Australia. In Australia, the presence of Archaeopteris has been recorded from several localities of Late Devonian age from Victoria (Talent et al. 2000). One anatomically preserved Callixylon wood fragment from the Famennian locality of Barraba (New South Wales) is currently under investigation . Until recently, the record of Mississippian lignophytes was almost non-existent in Gondwana before the Serpukhovian, which marks the onset of a Nothorhacopteris flora in South America (Balseiro et al. 2009, and references therein) and Australia (Hill et al. 1999). The arborescent lignophytes of Mississippian age from Australia were represented by logs of Viséan age assigned to the genus Pitus (Walkom 1928), a taxonomic assignation severely criticized by Gordon (1935). Early Mississippian environments were described as uniformly dominated by small to medium-sized lycophytes in this part of the world. Our recent research in two localities of upper Hastarian age in northeastern Queensland revealed the presence of abundant petrified fragments of trunks and branches representing three taxa of arborescent lignophytes (Decombeix et al. 2011) (5) in Table 1). One is the progymnosperm genus Protopitys, represented at Mont Saint Michael (Burdekin Basin) by the species P. buchiana. This is the earliest record for this species, which until now was known from Viséan horizons of Europe. The other taxon, as yet unnamed, has Pitus-type wood with a multiseriate pitting of the radial walls of tracheids, and high and wide rays. It differs from Pitus by the production of complex traces to lateral appendages. The third genus, Dameria (Decombeix et al. 2011), is a new taxon based on wood fragments from Montgomery Dam in the Broken River Region. Its main distinctive characters consist of: (1) the possession of wood tracheids showing one, or rarely two, rows of circular pits on the radial walls of tracheids; (2) radial pits not occupying the entire tracheid width and having a circular aperture; (3) rays mostly uniseriate and low; (4) ray-cells containing resin-like contents. This type of wood, which is common in early conifers but rare in older arborescent lignophytes, departs from the general structure of the Mississippian woods known thus far. It is surprisingly similar to a charcoalified wood fragment reported from the late Famennian (Fa2d, VH palynozone) locality of Trooz in Belgium (6) in Table 1) and shares many traits with Tovoxylon from northern Russia (Orlova 2009). We have hypothesized that the shared traits of Dameria, the Trooz charcoalified piece of wood, the two specimens of Tovoxylon and the Pennsylvanian conifer woods may represent a convergent response to comparable environmental constraints related to their possible marginal to extrabasinal habitats (Decombeix et al. 2011). The Hastarian specimens from Queensland represent the oldest occurrence of arborescent lignophytes in the Mississippian of Gondwana. Two other taxa were previously recognized, but in younger deposits, Cuyoxylon in the Serpukhovian of Argentina (Pujana & Cesari 2008) and a possible Eristophyton in the Serpukhovian of Morrocco (Chalot-Prat & Galtier 1989).
It will be clear from this review that the arborescent lignophytes were significant components of Mississippian floras in the palaeotropical belt. Recent advances highlight the following points. (1) A significant diversity of arborescent lignophytes is recorded since the Hastarian, less than 6 Ma after the Devonian-Carboniferous boundary (Hance & Poty 2006) and prior to the Ivorian. (2) Australia experienced the same broad patterns of changes, around the Devonian-Carboniferous boundary and in the Hastarian, as Montagne Noire and Thuringia. (3) Protopitys and Eristophyton, two genera characterized by an extended stratigraphical distribution in the Mississippian, evolved in the Hastarian. The two species recorded for Protopitys in the late Hastarian, and the large west-east distribution of this genus along the northern edge of Gondwana, indicate that it originated earlier, at a date and a place yet to be determined. (4) Based on the record from the Isle of Bute, the earliest evidence of Pitus could be Hastarian as well, but it is possible that Pitus evolved earlier, before the extinction of Archaeopteris. Another taxon that may have crossed the Devonian-Carboniferous boundary successfully is Dameria, if the charcoalified remains from Belgium and the Australian specimens that have a similar wood represent the same plant. (5) The rarity of arborescent lignophytes other than Archaeopteris in the late Famennian may indicate that they were present but in a marginal or extrabasinal habitat at a distance from the main depositional areas.

Late Devonian-Mississippian evolution of phenotypic disparity
Assessing disparity between forms in the early lignophytes is a difficult task because of the limited amount of comparable information between taxa. More than a century of work on anatomically preserved specimens has provided a significant amount of data on the stem anatomy of these plants. This information is synthesized here for a large selection of Devonian and Mississippian lignophytes and analysed to investigate (1) the disparity between arborescent and non-arborescent taxa and (2) the patterns of phenotypic diversification in the evolution of arborescent taxa from the Late Devonian to the Mississippian.
Sixty-five taxa were coded for 15 characters related to the primary and secondary vascular anatomy of stems and to the mode of production of ultimate appendages. Characters have binary states or non-ordered multistates and we chose to restrict the list of characters to those that are known for all the taxa.
Principal coordinate analyses were conducted using PAST software (Hammer et al. 2001). They are based on a dissimilarity matrix, here the number of character state differences for each pair of taxa. The first two principal coordinate axes were used to represent the occupied morphospaces in a plane.
The results of the analysis using the complete dataset are illustrated in Figure 2. The two first axes represent 48% of the total variation, with the first axis representing most of the variation. The morphospaces of arborescent and non-arborescent taxa show little overlap and are roughly separated along the first axis. Protopitys is placed in the arborescent group along axis 1 but close to the aneurophytalean progymnosperms along axis 2. This isolated position reflects the unusual combination of characters of the genus. The oldest seed plant in our analysis, Elkinsia, has a somewhat central position, the Lyginopteridales and Buteoxylales being relatively close. The other Mississippian lignophytes form two distinct groups. The first one, characterized by small steles, large metaxylem tracheids, very large rays and relatively long internodes, encompasses taxa assigned to the Calamopityales and Medullosales. These taxa are reconstructed as semi self-supporting shrubs or lianas, with arborescence probably being a derived state in some Medullosales (Rowe et al. 1993;Dunn 2006). The second group, characterized by a large stele, metaxylem tracheids with a diameter equal or inferior to secondary xylem tracheids, short internodes and smaller rays, corresponds to the putative arborescent seed plants. Figure 3 illustrates the results when only arborescent taxa are considered, with emphasis on their age. The two first axes represent 62% of the total variation, with the first axis representing 32%. The morphospace occupied by the Mississippian taxa is larger than that corresponding to the Late Devonian species of Archaeopteris, indicating a diversification of the vegetative body in the Early Carboniferous. When the age of the arborescent taxa is studied in more detail, it is clear that the Tournaisian taxa occupy a larger morphospace than either the Late Devonian or the Viséan taxa. The Tournaisian period is not only transitional between the Late Devonian and the Viséan, including morphologies present in these two periods, but it also contains forms no longer represented in the Viséan. This result suggests that the Tournaisian was a time of diversification for the vegetative structures of the lignophyte trees.
Working on characters instead of taxa offers a way to assess changes without the possible bias caused by doubtful taxonomic identification. The analysis of morphological disparity presented here shows that, for the group of vegetative characters considered, the diversification of lignophyte trees in the Mississippian paralleled that of smaller growth forms represented by the calamopityalean, lyginopteridalean and buteoxylalean seed ferns (Galtier 1988). It supports a major diversification of the lignophyte trees during the Tournaisian. Such analysis also shows that lignophyte trees have a distinctive stem morpho-anatomy, with little phenotypic similarity to that of the other growth forms evolved among the early lignophytes.

Phylogenetic analysis of the affinities of the arborescent taxa
According to the most commonly accepted hypothesis on early lignophyte evolution, both the Archaeopteridales and the seed plants originated from the progymnosperm Aneurophytales   Figure 2, showing the disparity of the vegetative body anatomy in Devonian and Mississippian arborescent lignophytes with emphasis on age (Protopitys is not represented). The morphospace of the Tournaisian + Viséan taxa is larger than that of the Devonian Archaeopteris, indicating a diversification of the vegetative body of lignophyte trees during the Mississippian. The Tournaisian corresponds to the period of maximum disparity. (Rothwell 1982) (however, see Beck (1981 and Meyen (1984) for alternative interpretations). The most recent phylogenetic analyses either place the Aneurophytales at the base of the lignophyte tree and the Archaeopteridales sister-group to the seed plants (Rothwell & Serbet 1994) or place Aneurophytales and Archaeopteridales in an unresolved polytomy at the base of the lignophytes (Hilton & Bateman 2006). The relations of the Protopityales are unresolved because of the small amount of data on this group. Beck (1976) hypothesized that they evolved independently from the Aneurophytales, an idea still plausible in the current state of our knowledge. Thus, according to currently available data, the tree habit (1) evolved independently in the progymnosperms and in the seed plants, (2) possibly evolved twice among the progymnosperms, in the Archaeopteridales (Devonian) and in the Protopityales (Mississippian), and (3) evolved numerous times among the seed plants from the Mississippian to the present. Because only a small number of early lignophyte taxa are fully reconstructed, phylogenetic analyses cannot reflect the diversity suggested by the fossil record, and this is especially true when we consider the arborescent taxa of the Mississippian for which reproductive organs are unknown. With the exception of Protopitys, all of them have been assigned a priori to the seed plants . The genera Pitus, Bilignea and Eristophyton have been tentatively interpreted as representing a monophyletic group (Galtier & Scott 1990). Two of them, Pitus and Bilignea, have been included in a phylogenetic analysis (Hilton & Bateman 2006), using reconstructions from dispersed vegetative and fertile organs. They are indeed placed within the basal seed plants but their position within this group was not resolved by the analysis.
To investigate the affinities of the arborescent taxa of the Mississippian with the data currently available, we tested the relevance of a cladistic analysis based mostly on vegetative characters. Such characters are usually regarded as highly adaptive and more plastic than reproductive characters, and thus receive less consideration in macroevolutionary analyses. We established a list composed of characters used in previous analyses and of well-documented characters previously used in systematic discussions. It includes characters of reproduction (2), general morphology (3), primary vascular system (8), secondary xylem (6), phloem and cortex (4), and leaves and their mode of production (9). The matrix contains 19 Devonian and Carboniferous taxa and 32 characters, and has 18% of ambiguous cells. A heuristic analysis was conducted using PAUP 4 (Swofford 2003) with the options TBR branch swapping, MULPARS. All characters are unordered and have equal weight. Psilophyton was used as an outgroup. Four analyses were conducted, experimenting with the addition or omission of Archaeopteris and Protopitys. A bootstrap analysis was conducted for 100 replications using the options TBR and MULTREES of PAUP. Figure 4a shows the strict consensus tree without Protopitys. The result is not robust but is none the less globally consistent with results of previous analyses of lignophyte phylogeny that contain numerous reproductive characters (Rothwell & Serbet 1994;Hilton & Bateman 2006). Similarities include the paraphyly of the progymnosperms, the position of Archaeopteris as sister-group to the seed plants, seed plants monophyly, the relationship of the Medullosales with the Callistophytales, as well as those of Thucydia with the Cordaitales (Cordaixylon, Mesoxylon). Close relationships between the Calamopityales and the Medullosales and between Laceya and Elkinsia have also been suggested by several researchers (e.g. Stein & Beck 1992;Galtier & Meyer-Berthaud 1996). The Mississippian trees form a polyphyletic group. Pitus is sister group to the non-arborescent taxa plus Faironia; Endoxylon is closer to Thucydia and the two cordaitales included in the analysis. This analysis indicates that the Mississippian trees show more affinities with the seed plants than with the progymnosperms, even when no assumptions are made on their reproductive structures. In this analysis, however, early conifers are basal and Elkinsia represents one of the most derived taxa, which contradicts both its stratigraphical age (late Famennian) and apparent degree of morphological complexity. We suspect that Archaeopteris introduces a bias and attracts the arborescent taxa in a basal position within the seed plants. With the removal of Archaeopteris from the dataset (Fig. 4b), Laceya and Elkinsia occupy a basal position within the seed plants, which is more compatible with the currently available fossil record. Apart from Laceya, Elkinsia and Faironia, early spermatophytes are organized in two clades, one composed of trees and its sister-group consisting of the non-self-supporting pteridosperms showing a 'manoxylic' (i.e. not dense) wood. The introduction of Protopitys in this analysis (Fig. 4c) does not change the topology of the consensus tree. Protopitys is placed within the seed plants, in the 'tree' group. In a last experiment, we added both Archaeopteris and Protopitys to the matrix (Fig. 4d). This results in a consensus tree in which Archaeopteris is sister to a group that includes all the other arborescent taxa and the seed plants. The arborescent taxa form a large basal polytomy and some may not be spermatophytic. Interestingly, in a previous (unpublished) attempt to include Protopitys in a lignophyte phylogeny (Hammond 2004), this taxa was resolved as sister-group to the seed plants. Although we consider that information on Protopitys is too incomplete at the moment to conclude on its affinities, it is interesting to keep in mind that some workers considered the alternative hypothesis that Protopitys might in fact be a seed plant for which only male reproductive structures have yet been found (e.g. Rothwell & Serbet 1994, p. 479). Further data on this genus are needed to determine whether all the arborescent taxa of the Mississippian were indeed seed plants.
Of the three types of analyses conducted in this paper, those aimed at exploring the relationships of the lignophyte trees have provided the least conclusive results. First, affinities of the non-archeaopteridalean arborescent taxa with the seed plants, although suggested by three out of four of the experiments conducted, are not strongly supported. However, the topologies that are the most consistent with the stratigraphical appearance of the taxa and with current evolutionary ideas were obtained when Archaeopteris was removed, and these topologies include the arborescent taxa in the seed plants. Second, the four experiments undertaken in this paper generate tree topologies in which the non-archeaopteridalean arborescent taxa, Faironia excepted, are consistently grouped, whether as a grade or as a clade. It is possible that these topologies result from convergent features constrained by the tree habit rather than by real affinities. If not, it is interesting to note that according to the experiments conducted without Archaeopteris, the arborescent taxa (Faironia excepted) form a monophyletic group that may have undergone its own diversification within the seed plants.

Discussion
This paper examines the conditions in which a particular set of lignophytes evolved the tree habit around the Devonian-Carboniferous boundary and replaced the Late Devonian genus Archaeopteris.The results of the taxic diversity analysis indicate that the Mississippian arborescent lignophytes of the palaeotropical belt were diversified in the Tournaisian. This was followed by a long phase characterized by the persistence of some, but not all, of the Tournaisian genera, and a low number of originations until the Namurian. The first range of taxa evolved in the Hastarian, or possibly before, predating the Ivorian diversification that was previously documented in the literature.
Analysis of phenotypic changes across the Devonian-Carboniferous boundary supports the occurrence of a Tournaisian diversification. The arborescent taxa of Tournaisian age occupy a wider morphospace than that of the Devonian Archaeopteris, which highlights a phenotypic diversification of the lignophyte trees in the Early Mississippian. The Viséan morphospace appears reduced compared with that occupied by the Tournaisian taxa and represents only part of the latter. This indicates that, based on stem features, the range of vegetative morphologies exhibited by the lignophyte trees narrowed after the Tournaisian phase of diversification. Another outcome of the disparity analysis is that it highlights the phenotypic distinctiveness of trees compared with other growth forms within the lignophytes. A large number of these differences are related to the structure of the primary body , which is supposedly less adaptive than other characters such as wood anatomy. This might be an indication that the Tournaisian diversification of lignophyte trees corresponds at least in part to a non-adaptive radiation; that is, the evolution of phenotypic novelties in an empty or undersaturated landscape characterized by a low level of competition (e.g. DiMichele & Bateman 1996.
We are aware of the limitations of our phylogenetic analysis and consider its results as tentative. Most of the tests presented in this paper support the idea that the Mississippian trees have seed plant affinities. Except Faironia, which shares characters with both arborescent and non-arborescent taxa, all the Mississippian taxa showing the tree habit are grouped. They occur either as a basal grade or as a clade within the seed plants. In the latter topology, which is the most consistent with the stratigraphical distribution of the fossils, the early spermatophytes apart from Laceya, Elkinsia and Faironia are organized in two clades. These represent increased specialization compared with the basal taxa and correspond to distinctive growth forms, one composed of trees and its sister-group consisting of the non-self-supporting 'manoxylic' pteridosperms.
Assessing the environmental conditions that would have triggered the evolution of the arborescent lignophytes succeeding Archaeopteris is more conjectural. Evidence for mass extinction in the plant fossil record appears controversial (McElwain & Punyasena 2007), especially around the Devonian-Carboniferous boundary, for which no global quantitative analysis of diversity changes has been attempted. A number of workers, however, have emphasized the extinction of some critical taxa and a drastic reorganization of plant communities, suggesting that a crisis affected land plants at the Devonian-Carboniferous boundary (Fairon-Demaret 1986;DiMichele & Hook 1992;Streel et al. 2000). The extinction of plants that, during the Late Devonian, inhabited lowlands would have impoverished the plant cover in such habitats. The resulting lowered competition between organisms would have favoured the settlement and radiation of plants with new body plans (seed plants, lignophyte trees) in this type of environment. On the other hand, indirect evidence provided by charcoalified remains suggests that by the Latest Famennian, diverse communities of plants with 'advanced morphologies' may have thrived in marginal to extrabasinal settings, possibly uplands, and at a distance from the lowlands preferentially occupied by their pteridophytic relatives (Fairon-Demaret & Hartkopf-Fröder 2004;. These new habitats would have provided an ideal environmental context for the evolution of the lignophyte trees succeeding Archaeopteris.
Based on this information, we propose in Figure 5 a scenario for the changes affecting the arborescent lignophytes around the Devonian-Carboniferous boundary. One alternative is related to the spermatophytic versus progymnospermous affinities of the taxa; that is, can the success of the Mississippian trees be explain by the seed-habit or are some of them free-sporing heterosporous plants like Archaeopteris? A second alternative is represented by the sole contribution of lowlands versus the additional contribution of upland floras to the evolution of the Mississippian taxa; that is, were there already upland lignophyte trees in the latest Devonian? In Figure 5, we postulate the occurrence of arborescent taxa different from Archaeopteris as early as the late Famennian, some inhabiting lowlands, others colonizing marginal to upland settings. This scenario proposes that both types of trees persisted at the Devonian-Carboniferous boundary, with lowlands possibly providing a refuge to some of the upland trees. The community of lignophytes inhabiting the lowlands after the demise of Archaeopteris at the Devonian-Carboniferous boundary was more diverse than previously thought and consisted not only of the shrubby r-selected species mentioned above but also of some arborescent taxa. The next stages of the scenario illustrate the diversification of the arborescent lignophytes originating from this community that spread through lowlands and eventually uplands during the late Hastarian and the Ivorian.
Additional data are now necessary to test the validity of this Fig. 5. Proposed scenario of the transitional changes in arborescent lignophytes from the latest Devonian to the Viséan. We postulate the presence of lignophyte trees distinct from Archaeopteris in the latest Devonian both in lowlands and in uplands. Unlike Archaeopteris, these taxa crossed the Devonian-Carboniferous boundary, possibly using lowlands as a refugium in the case of significant environmental changes. They then diversified during the Late Hastarian and Ivorian. Protopitys is represented separately as it is supposed to have a free-sporing heterosporous reproduction and thus to have environmental constraints more similar to those of Archaeopteris than to those of the other Mississippian trees. The dotted rectangles represent the two components of this scenario that need further investigation: the composition of upland floras during the Late Devonian and Mississippian, and the precise stratigraphical and spatial distribution of arborescent taxa at the Devonian-Carboniferous boundary.
scenario. Future work should focus on providing (1) a better understanding of the biology of both Archaeopteris and the 'Mississippian' lignophyte trees, to assess the existence of possible environmental constraints on their distribution, and (2) a more precise idea of the stratigraphical and spatial distribution of the earliest lignophyte trees distinct from Archaeopteris. Although limited at the moment, the information available on Late Devonian and Early Mississippian floras from outside Europe and North America (e.g. Decombeix et al. 2011) and on floras representing upland ecosystems (e.g. Fairon-Demaret & Hartkopf-Fröder 2004) is very promising. Future investigations of similar floras are thus expected to improve significantly our understanding of the transition between the Devonian and Carboniferous forests.
We thank C. Cleal and J. Hilton for their invitation to participate in the Lyell Meeting 2009 'Late Palaeozoic terrestrial habitats and biotas: the effect of changing climate'. We also thank M. Streel (Université de Liège) for his advice on biostratigraphy and palaeogeography, and R. Serbet, E. Taylor and T. Taylor (University of Kansas) for access to the Matten collection. Thanks are also due to two anonymous reviewers whose suggestions helped improve the paper. Our research was supported by two projects: INSU-Sciences de la Terre ('The Terrestrialization Process: Modelling Complex Interactions at the Biosphere-Geosphere Interface') and Agence Nationale de la Recherche 'ACCRO-Earth'. AMAP (Botany and Computational Plant Architecture) is a joint research unit that associates CIRAD (UMR51), CNRS (UMR5120), INRA (UMR931), IRD (R123), and Montpellier 2 University (UM27).