Buried deep beyond the veil of extinction: Euphyllophyte relationships at the base of the spermatophyte clade

Taxon selection

This study includes 28 anatomically preserved taxa known from permineralized specimens (Appendix 1). Nine of these are early seed plants or putative seed plants: Calathopteris heterophylla Long, Elkinsia polymorpha Rothwell, Scheckler et Gillespie, Laceya hibernica May et Matten, Tetrastichia bupatides Gordon, Triradioxylon primaevum Barnard et Long, Tristichia longii Galtier, Tristichia ovensi Long, Tristichia tripos Galtier et Meyer-Berthaud, and Yiduxylon trilobum Wang et Liu. Six are placed in the Stenokoleales: Brabantophyton runcariense Momont, Gerrienne et Prestianni, Crossia virginiana Beck et Stein, Stenokoleos bifidus Matten et Banks, Stenokoleos holmesii Matten, Stenokoleos setchelli Hoskins et Cross, and Stenokoleos simplex Beck. Eight have been classified as aneurophytalean progymnosperms: Aneurophyton germanicum Kräusel et Weyland, Proteokalon petryi Scheckler et Banks, Reimannia aldenense Arnold, Rellimia thomsonii Leclercq et Bonamo, Tetraxylopteris schmidtii Beck, Triloboxylon ashlandicum Scheckler et Banks, and Triloboxylon arnoldii Matten. Aside from these groups, we included Actinoxylon banksii Matten, a species described initially as a pityalean progymnosperm (Matten, 1968) and discussed by Beck (1976) as a potential archaeopteridalean progymnosperm. We also included three euphyllophytes of unresolved taxonomic affinities: the Emsian plant described by Gensel (1984) from the Battery Point Formation of Gaspé (Canada), Gothanophyton zimmermanni Remy et Hass (Emsian), and Langoxylon asterochlaenoideum Scheckler, Skog et Banks (Givetian). The outgroup used to root the analyses is Psilophyton dawsonii Banks, Leclercq et Hueber. This plant is the best characterized early euphyllophyte, to date, in terms of anatomy and morphology, and predates younger and structurally more complex Devonian euphyllophytes (Banks et al., 1975).

Character definition and scoring

We used 49 characters, of which 40 are discrete characters (33 anatomical, seven morphological) and nine are continuous characters (five of them are ratios and four are absolute sizes) (Appendix 2). The matrix was assembled in Mesquite 3.2 (Maddison and Maddison, 2009). Characters were scored from the literature. Overall, the matrix has 10.23% missing data: 11.34% for the discrete characters and 5.56% for continuous characters (Appendices 3, 4).

Continuous characters have been shown to add phylogenetically useful information that may not be codified in discrete character states (Escapa and Pol, 2011). Here, continuous characters (Appendix 3) are each based on a single measurement. To avoid over-emphasizing small differences between taxa, these measurements were converted into ranges by adding and subtracting 10% from the measured value. The ranges were subsequently standardized to be equivalent to one step of a discrete character by dividing the end values of each range (maximum and minimum) by the highest overall maximum of the character across the entire set of taxa. Standardization was performed to avoid variation in character weighting resulting from the magnitude of absolute values.

Only one of the discrete anatomical-morphological characters (Appendix 4) is not vegetative, because reproductive structures are known in very few of the taxa included in this study: Psilophyton, Rellimia, Tetraxylopteris, Aneurophyton, and Elkinsia (Beck, Appendix 3 (continued); Banks et al., 1975; Scheckler, 1976; Bonamo, 1977; Serlin and Banks, 1978; Schweitzer and Matten, 1982; Rothwell et al., 1989; Serbet and Rothwell, 1992; Dannenhoffer et al., 2007).

In coding the morphology of the Devonian plants into discrete characters, we strived to avoid introducing a priori homology assumptions and to consider how morphology and anatomy may have related to development in these plants (based on current knowledge of development in living plant lineages). As a result, some characters introduce novel perspectives on basic determinants of sporophyte organization. One of these is a character (10) implying that sporophyte axes fall into two major types with distinctly different modes of development that lead to an internal organization exhibiting either radial symmetry (termed the radial organographic domain) or bilateral symmetry (bilateral organographic domain). These two modes of development are underpinned by different regulatory programs and, in plants that possess both types, the development of axes with bilateral symmetry on radially symmetrical subtending axes marks a switch from one regulatory program to the other, in a way similar to the onset of specific leaf developmental programs at the shoot apical meristem in derived euphyllophytes (Sanders et al., 2007; Lenhard, 2017).

In another example, the presence of adaxial–abaxial polarity (character 44), whose determinants are not known in these extinct plants, was inferred based on presence of asymmetry between the adaxial and abaxial sides of primary xylem in the traces of laterals. Features such as adaxial concavity of the trace, asymmetry in the position of protoxylem strands within the trace, or asymmetry in the outline of the trace (e.g., abaxially but not adaxially lobed) were interpreted as marks of adaxial–abaxial polarization of tissue development.

Characters based on histology of the cortex are also relevant to how the sporophyte of these Devonian plants developed. For instance, presence of more than one cell type in the cortex (e.g., parenchyma and sclerenchyma) indicates differential gene expression in different regions of the cortex. Thus, differentiation of cortical regions that are consistently distinct in cell type must be a result of partitioning of the cortex volume into developmental domains specified by distinct regulatory programs, such as partitioning into concentric layers (e.g., bands of sclerenchyma nests in the inner cortex) or into radial sectors (e.g., alternating parenchymatous and sclerenchymatous areas around the periphery of axes, in the outer cortex, as seen in Dictyoxylon-type organization).

Similarly, protoxylem architecture (characters 17–19) must reflect patterns of polar auxin transport in the developing tip of axes. In living seed plants, basipetal auxin flow from leaf primordia causes development of the sympodia that characterize their eusteles (Benková et al., 2003). These correspond to protoxylem strands that do not converge into a central strand, which is also absent in the Devonian euphyllophytes with permanent protoxylem architecture (sensu Beck and Stein, 1993), such as cladoxylopsids, as well as in some early seed plants (e.g., Tristichia longii). Currently, we do not know what patterns of polar auxin transport were like at the tips of Devonian plant axes. However, if polar auxin flow from the apical meristem had been involved in procambial development of the Devonian euphyllophyte axes, a radiate protoxylem architecture (i.e., characterized by convergence of protoxylem strands from lateral into a central protoxylem strand of the main axis) would have necessitated convergence of auxin transport pathways from laterals into a single central stream.

Phylogenetic analyses

Phylogenetic searches were conducted in TNT 1.5 (Goloboff and Catalano, 2016), using equally weighted parsimony as the optimality criterion, and 50,000 trees were held in the memory using the command “hold 50,000”. The parsimony analyses were initiated using the command “xmult=hits10”. Using this command, the analysis departs from 50 random addition sequences (RAS), followed by tree bisection–reconnection. The resulting trees were submitted to a combination of Ratchet (default options), Tree Drifting (default options), and sectorial searches (default options). Bootstrap values were generated using the “bootstrap resampling” command with standard tree search parameters and 100 replicates. Consistency indices (CI) and retention indices (RI) were calculated using the “stats.run” script provided with the TNT installation package.

We used two character-sampling regimes in two different analyses. The first tree search (Analysis 1) was run using only discrete characters. The second analysis included discrete plus continuous characters (Analysis 2). All characters were equally weighted and unordered to avoid introducing bias from a priori assumptions. The time calibrated tree was produced with R software (R Core Team, 2017) utilizing the functions timePaleoPhy and geoscalePhylo in the paleotree and strap packages, respectively (Bapst, 2012; Bell and Lloyd, 2015).

Analysis 1 (discrete characters only)

This search resulted in 19 most parsimonious (MP) trees (tree length 84; CI = 0.548, RI = 0.683). In the strict consensus tree (Appendix 5), the ingroup forms a large polytomy that includes only two resolved clades. One of these consists of two aneurophytes (Tetraxylopteris and Proteokalon), whereas the other includes the seed plants (except for Yiduxylon), with Tristichia tripos sister to the remaining seed plants, which form a polytomy.

Analysis 2 (discrete + continuous characters)

Addition of continuous characters led to full resolution: we recovered a single MP tree (tree length 96.4330; CI = 0.528, RI = 0.656) (fig. 1 and Appendix 6). The putative aneurophyte Reimannia is recovered as sister to the rest of ingroup species. An aneurophyte clade (fig. 1) consists of Aneurophyton and Cairoa forming a grade basal to the divergence of two clades, one including Triloboxylon ashlandicum and Rellimia, while the other includes Proteokalon and Tetraxylopteris. The aneurophyte clade is sister to a larger clade, characterized by the presence of an organographic domain that exhibits bilateral symmetry and termed the “bilateral clade” (fig. 1), in which Yiduxylon (putative seed plant) and Actinoxylon (putative progymnosperm) form a basal grade. A major dichotomy separates this larger clade into two clades, one including seed plants, while the other includes the Stenokoleales. In the former, Triloboxylon arnoldii is sister to the seed plant clade (fig. 1); in the latter, Gensel's (1984) plant and Langoxylon form a grade basal to the Stenokoleales (fig. 1). Within the seed plant clade, Tristichia tripos is sister to the remaining seed plants, which are resolved in two clades: one in which Tetrastichia is sister to Calathopteris + Elkinsia, and one in which Triradioxylon + Tristichia ovensi is sister to Laceya + Tristichia longii. Within the Stenokoleales, Brabantophyton + Crossia is sister to a clade consisting of grade that includes Gothanophyton and Stenokoleos simplex basal to S. holmesii + S. bifidus.

Clades and synapomorphies

The results of Analysis 2 provide synapomorphies that support each clade. The seed plant clade (exclusive of Yiduxylon) (fig. 1) is supported by the presence of pulvinus-like branch bases (character 36). This character is also the synapomorphy that defines the Stenokoleos clade and is present in Crossia, which implies that pulvinus-like branch bases evolved independently in these groups. It is worth noting that Tristichia, a genus classified among the seed plants, is polyphyletic, with one species (T. tripos) sister to all other seed plants, whereas the other two (T. ovensi and T. longii) are each part of a different clade within the seed plants.

The synapomorphies that unite the Stenokoleales in a monophyletic group (including a Stenokoleos clade, as well as a Brabantophyton + Crossia clade, sister to the clade formed by Gothanophyton and Stenokoleos; fig. 1) are the traces supplying the bilateral organographic domain, which consist of more than one vascular bundle (character 42), with the bundles diverging tangentially from the tip of a xylem rib (character 35), and the bipartite architecture of the bilateral organographic domain (at its base, which is the only known portion of it) (character 45). Some of these features are present also in Yiduxylon, Triloboxylon arnoldii, and Tristichia longii, suggesting that traces to the bilateral domain consisting of multiple vascular bundles with tangential divergence may have evolved independently in other groups. The stenokolelalean clade is also characterized by the highest values of the ratio of maximum primary xylem diameter to maximum axis diameter (character 0); similar values (>0.6) are found outside of this clade only in Gensel's (1984) plant.

The larger clade formed by the two sister clades each of which includes the seed plant clade (with Triloboxylon arnoldii as sister group) and the stenokolealean clade (with the grade formed by Gensel's 1984 plant and Langoxylon at the base), respectively (fig. 1), is united by the architecture of the vascular supply to the bilateral domain, wherein traces that diverge from the primary xylem ribs do not exhibit further divergence as they enter the base of appendages with bilateral symmetry (character 43).

The aneurophyte clade (exclusive of Triloboxylon arnoldii and Reimannia; fig. 1) is supported by the presence of recurring appendages with terete xylem (character 37) and also by two continuous characters: an increase in the ratio of primary xylem to axis surface area (as seen in cross section) (character 1) and lowest metaxylem tracheid diameter values (in the radial organographic domain; character 4).

We recovered a large lignophyte clade (fig. 1), which includes all ingroup taxa, except for Reimannia (a putative aneurophyte in which secondary growth, if present, has yet to be discovered). The clade is supported by the presence of secondary xylem (character 26). However, in the current tree topology, this character shows a reversal to absence of secondary growth in the ancestor of the clade including Gensel's (1984) plant + Langoxylon + Stenokolelales, and a further reversal to presence in the Brabantophyton + Crossia clade, within the Stenokoleales.

Current understanding of relationships

A small number of studies have addressed questions of phylogeny with implications for the relationships of Devonian euphyllophytes traditionally associated with the origin of seed plants. Rothwell and Serbet (1994) and Hilton and Bateman (2006) were concerned primarily with the relationships among major seed plant lineages and included progymnosperms as outgroups. The focus of Matten (1992) and Galtier and Meyer-Berthaud (1996) was on the relationships among the earliest seed plant groups, and they also addressed relationships between seed plants, Stenokoleales, and progymnosperms. In the most recent analysis, Momont (2015) addressed relationships between major euphyllophyte groups, including progymnosperms, Stenokoleales, and seed plants.

Rothwell and Serbet (1994) and Hilton and Bateman (2006) did not include Stenokoleales in their analyses. Both these studies sampled extensively seed plant diversity including all living and extinct gymnosperm groups, as well as angiosperms, but included only three progymnsperms representing the aneurophytes, archaeopterids, and cecropsids. Both the study by Matten (1992) and the one by Galtier and Meyer-Berthaud (1996) included Stenokoleales and progymnosperms. Whereas Galtier and Meyer-Berthaud (1996) sampled six protostelic early seed plants, Matten (1992) included only two terminals for seed plants, each representing a composite concept of a “Lyginopteridales plant” and a “Calamopityales plant” based on seven and four genera, respectively. In both studies, the Stenokoleales and the progymnosperms (only aneurophytes in Galtier and Meyer-Berthaud's study; aneurophytes and archaeopterids in Matten's study) are also included as single terminals represented by composite plant concepts drawn from several species or genera. Momont's (2015) analysis included four aneurophytes (Rellimia, Aneurophyton, Triloboxylon ashlandicum, and Tetraxylopteris), one archaeopterid (Callixylon), three Stenokoleales (Stenokoleos holmesii, S. simplex, and Brabantophyton), and two seed plants (Elkinsia and Tristichia tripos). Aside from these, the analysis also included basal euphyllophytes (Psilophyton and Armoricaphyton) and cladoxylopsids (one pseudosporochnalean and one iridopterid).

In the analysis of Hilton and Bateman (2006), rooted with the aneurophyte Tetraxylopteris, progymnosperms (Tetraxylopteris and an Archaeopteris + Cecropsis group) and seed plants form a basal polytomy. Rothwell and Serbet (1994) used a theoretical set of ancestral characters states to root their analysis and recovered an archaeopterids + cecropsids group sister to the seed plants, in a clade that is sister to the aneurophytes. These relationships were recovered consistently and only collapsed in a polytomy when parsimony was relaxed to MP + 2 steps. Matten (1992) also recovered a polytomy between aneurophytes, archaeopterids, and seed plants, when Stenokoleales were excluded from analyses. When included, Stenokoleales resolved as sister to seed plants in a clade that was sister to a progymnosperm clade (aneurophytes + archaeopterids). In the study of Galtier and Meyer-Berthaud (1996), Stenokoleales are sister to seed plants. However, this configuration may be a result of using only aneurophytes and Stenokoleales as outgroups, each as a single terminal (composite plant concept), with the analysis rooted by the aneurophyte lineage. Finally, in Momont's (2015) phylogenetic analysis rooted with P. dawsonii, the Stenokoleales and seed plants form a large polytomy, with the archaeopterid as the sister group. Basal to this clade, Aneurophytales form a paraphyletic group. The cladoxylopsids form a clade that is sister to the progymnosperm + Stenokoleales + seed plants clade.

Among these studies, three included Stenokoleales in addition to Aneurophytales and seed plants (Matten, 1992; Galtier and Meyer-Berthaud, 1996; Momont, 2015). In all these studies, Stenokoleales and the seed plants form a clade. However, (1) in Galtier and Meyer-Berthaud's study, this relationship is constrained by the choice and number of taxa (see above); and (2) in Matten's and Momont's analyses, relationships within the Stenokoleales + seed plants clade are unresolved (polytomy of Momont 2015 and each of the two groups represented by a single terminal of Matten 1992). Whereas in Matten's study the progymnosperms (aneurophytes + archaeopterids) form a clade that is sister to Stenokoleales + seed plants, in Momont's analysis, the progymnosperms form a paraphyletic grade basal to Stenokoleales + seed plants.

Phylogeny and classical taxonomy

Our analysis (Analysis 2) recovers an aneurophyte clade (exclusive of Reimannia aldenense and Triloboxylon arnoldii, both classified at least tentatively as aneurophytes—see below), a Stenokoleales clade (including Gothanophyton zimmermanni, a plant of unresolved affinities), and a seed plant clade (exclusive of Yiduxylon trilobum, which was discussed as a putative seed plant; Wang and Liu, 2015). Recovery of an aneurophyte clade is in contrast to the results of the only previous analysis that included more than one aneurophyte progymnospmerm (Momont, 2015), which recovered aneurophytes as a paraphyletic group. However, if Reimannia is an aneurophyte, our results also support the paraphyletic status of aneurophytalean progymnosperms, and if Triloboxylon arnoldii is an aneurophyte, the implication is that aneurophytes are polyphyletic.

In previous analyses, the seed plant clade was placed as sister to a clade including the Stenokoleales. Broadly consistent with this hypothesis of relationships, we recover seed plants and Stenokoleales in two sister clades in which each of them is accompanied by “outlier” species: Triloboxylon arnoldii, a putative aneurophyte, forms a clade with the seed plants; Langoxylon asterochlaenoideum and Gensel's (1984) plant form a grade at the base of the Stenokoleales.

The clades representing the three major taxonomic groups are supported by minimal numbers of discrete synapomorphies (one each for the aneurophyte and seed plant clades, and three for the Stenokoleales) and bootstrap support values are low (Appendix 6). This situation is likely due to the constraints of our data set, including the relatively simple morphology of the plants, which limits the number of characters that can be defined; a general lack of knowledge of reproductive structures (with very few exceptions: Psilophyton, Rellimia, Tetraxylopteris, Aneurophyton, and Elkinsia), which further limited the number of characters; the fragmentary nature of the species included, which resulted in uneven numbers of characters that could be scored across species; the broad taxonomic sampling, which, due to the fragmentary nature of species, further limited the number of characters that could be scored across most taxa in the data set and, therefore, used in the analysis; and relatively high levels of homoplasy of anatomical characters. The lack of resolution in the results of Analysis 1 (discrete characters only) is, thus, not very surprising. Nevertheless, the majority rule consensus tree resulting from the analysis based exclusively on discrete characters (Appendix 7) shows that the same seed plant clade and aneurophyte clade recovered by the analysis including both discrete and continuous characters, as well as a Stenokoleos clade, are recovered in at least 68% of MP trees (68% for the aneurophyte clade, 73% for the Stenokoleos clade, and 100% for the seed plant clade). However, the same majority rule consensus tree also suggests that Brabantophyton, Crossia, and Gothanophyton are not recovered as part of a clade with Stenokoleos in a high number of MP trees, and shows that aneurophytes (and not Stenokoleales) are recovered as more closely related to seed plants in only slightly more than half of the MP trees (52%).

The outlier species in our results reflect problems in the taxonomy of Devonian euphyllophytes characterized by lobed protosteles. These are mostly due to discrepancies between theoretical concepts of higher taxa (e.g., progymnosperms, Aneurophytales, seed plants) and the realities of diagnostic characters preserved in specimens. For instance, not all species classified as progymnosperms (therefore, lignophytes) have been demonstrated to possess secondary growth, let alone from a bifacial cambium; additionally, reproductive structures are not known for all the species placed among progymnosperms (Bonamo, 1975; Beck, 1976). Likewise, most of the earliest species classified as seed plants preserve no evidence of reproductive structures to ascertain their seed plant identity (e.g., Galtier and Meyer-Berthaud, 1996; Dunn and Rothwell, 2012; Wang and Liu, 2015). Furthermore, for the Stenokoleales, little is known about the architecture and anatomy of their branching systems and no reproductive structures have been documented (Beck, 1960b; Matten and Banks, 1969; Matten, 1992; Momont et al., 2016b). Another major issue is that different species are known at different levels of anatomical and morphological detail, making for uneven coverage in terms of comparisons and character scoring. As a result of all these, previous taxonomic assignments of several of the species considered here have been based on comparisons and characters other than those that are diagnostic for the respective higher taxa. Such taxonomic decisions could explain at least in part why our phylogeny (Analysis 2) does not fully match the traditional taxonomic placements of some of the species (outlier taxa).

Outlier taxa

In the case of Triloboxylon arnoldii, considered an aneurophytalean progymnosperm, Stein and Beck (1983) have pointed out that (1) emission of traces to laterals that consist of paired vascular bundles, along with (2) the presence of sclerenchyma in the inner cortex, set this species apart from the concept of a typical aneurophyte. Indeed, in our analysis T. arnoldii is recovered nested well within the bilateral clade, as sister to a seed plant clade, with which it shares numerous features; the clade formed by seed plants + Triloboxylon arnoldii has scattered sclerenchyma in the inner cortex (character 29) as a synapomorphy. Its sister group relationship with the seed plant clade raises the possibility that the species reconstructed as T. arnoldii may represent seed plant remains. However, as pointed out by Stein and Beck (1983), the significant size difference between the main axes of T. arnoldii and the much smaller size of the traces to lateral appendages documented on these axes suggests alternative interpretations. It is possible that these traces supplied fertile appendages, as suggested by their resemblance to the anatomy of reproductive organs described in aneurophytes. In turn, this suggests we may still be missing some of the vegetative parts of this plant, i.e., that T. arnoldii may have also had larger vegetative branching systems of intermediate size between the main axes and the putative fertile appendages (Stein and Beck, 1983), in which case those branching systems could have possessed a vascular supply of radial symmetry.

The position of Reimannia aldenense outside of the aneurophyte clade and as sister to all other ingroup species, is supported by several synapomorphies: regular branch taxis (character 12), lobed primary xylem (i.e., actinostele) (character 14), and metaxylem tracheids with bordered pits (character 24). The placement of Reimannia implies that either it is not an aneurophyhte or, if it is, then aneurophytes may be a bigger group that includes additional, yet to be discovered, diversity and forms a grade at the base of the bilateral clade. While Reimannia aldenense has been assigned to the aneurophytes (Matten, 1973; Stein, 1982), the species shows significant anatomical differences from the typical aneurophyte, which could explain its outlying position. These include the absence of alternating bands of sclerenchyma and parenchyma in the outer cortex, tangentially produced traces (Stein, 1982) and the absence of secondary xylem (Even though the possibility that Reimannia did produce secondary tissues cannot be ruled out, for the time being, the lack of secondary growth represents a difference from Aneurophytales.). The fragmentary nature of some members of Aneurophytales (including Reimannia itself) as currently characterized, as well as the broad anatomical diversity present in the group (Stein, 1982), may also have contributed to the placement of Reimannia recovered here.

Another outlier, Actinoxylon banksii, was first described by Matten (1968) as a protostelic progymnosperm and was placed in the order Pityales. Subsequently, Beck (1976) dissolved the order Pityales and placed Actinoxylon among archaeopteridalean progymnosperms, implying a eustelic, rather than protostelic architecture. We question this assignment on two grounds: (1) Beck himself was unsure of the exact nature of tissues at the center of Actinoxylon axes, which he indicated with a question mark (Beck, 1976, fig. 2); (2) in the original description, Matten (1968, figs. 1, 2, 4, 5) mentioned and showed tracheids in the incompletely preserved region at the center of the stele. Consequently, here we treat Actinoxylon as having a mesarch protostele. In the MP tree of Analysis 2, Actinoxylon is recovered as sister to the clade including Stenokoleales and seed plants, supported by the presence of protoxylem strands along the primary xylem ribs of the stele (character 18) as synapomorphy. One continuous character, the ratio of maximum primary xylem diameter to maximum axis diameter (character 0), also supports the clade formed by Actinoxylon and its sister group, reaching higher values at the base of this clade and throughout most of it. The position of Actinoxylon suggests that it is more closely related to seed plants than the aneurophytalean progymnosperms.

Yiduxylon trilobum, an early Devonian euphyllophyte, was putatively assigned to the seed plants by Wang and Liu (2015). These authors interpreted Yiduxylon as a transitional form between aneurophytes and early seed plants based on (1) presence of protoxylem strands only at the tip of the primary xylem ribs (and not along rib midplanes); (2) size of tracheids and rays in the secondary xylem, though to represent an intermediate between pycnoxylic wood (as typically attributed to aneurophytes) and manoxylic wood (typically attributed to early seed plants); (3) presence of bordered pits on both tangential and radial walls of secondary xylem tracheids, as seen in aneurophytes but not in seed plants, which have pits only on the radial walls; (4) tangential divergence of traces to laterals (seen in seed plants; e.g., Tristichia longii), instead of radial divergence (as in aneurophytes). In contrast to early seed plants, in Yiduxylon the traces to laterals that diverge as paired vascular bundles divide further, to form four bundles per trace. Additionally, Yiduxylon is distinguished from the seed plants in our analysis by the absence of sclerenchyma in the inner cortex. This combination of characters is responsible for the position of Yiduxylon as sister to the clade formed by the other species possessing a bilateral domain.

The placement of Yiduxylon trilobum in our analysis indicates that this species may not be a seed plant as proposed by Wang and Liu (2015), but is consistent with those authors’ interpretation of the species as an intermediate between aneurophytes and seed plants. However, current knowledge of Yiduxylon is relatively limited, with anatomical features such as the presence of a central protoxylem strand and metaxylem tracheid pitting type (let alone reproductive structures) still undocumented. Therefore, further in-depth characterization of this plant may lead to changes in its phylogenetic placement.

Gothanophyton zimmermanni is another plant of uncertain taxonomic placement that combines stenokolealean features (tangential divergence of paired bundles forming the trace to a lateral appendage) with P-type tracheid pitting, plesiomorphic among euphyllophytes. Compared by Remy and Hass (1986) with the aneurophytes, Gothanophyton was discussed as a putative iridopteridalean cladoxylopsid by Scheckler et al. (2006). In our analysis, Gothanophyton is nested among the Stenokoleales and sister to the Stenokoleos clade, with which it is united by the number of primary xylem ribs: four (or more) (character 15).

Langoxylon asterochlaenoideum, a Middle Devonian euphyllophyte, was not assigned to any specific taxonomic group by Scheckler et al. (2006). Langoxylon combines features of several major Devonian taxa, including the Aneurophytales (e.g., similar length of actinostele ribs with several protoxylem strands along the midplanes), Archaeopteridales (e.g., pith-like zone at the center of the stele), Stenokoleales (e.g., protoxylem parenchyma), as well as Gothanophyton and the Iridopteridales (H-shaped bilaterally symmetrical traces) (Scheckler et al., 2006). Our analysis recovered Langoxylon as sister to the Stenokoleales (including Gothanophyton), with which it forms a clade defined by the presence of protoxylem parenchyma (character 21) and by higher ratios between primary xylem rib basal width and maximum xylem diameter (in cross section) (character 3). The placement of Langoxylon suggests that, if it is not a stenokolealean, it represents a lineage closely related to the Stenokoleales. The same inference applies to Gensel's (1984) plant, recovered as sister to the Langoxylon + Stenokoleales clade, in a clade supported by two continuous characters – the ratio between the primary xylem size and axis size (characters 0 and 1).

Broader phylogenetic patterns

In traditional taxonomic treatments, the three major taxonomic groups considered here, the aneurophytes, Stenokoleales, and seed plants, are not as distinctly different from each other, in terms of salient anatomical features, as their evolutionary relationships would predict. These blurry boundaries are due to (1) the small number of characters used to define these groups (particularly in the absence of known reproductive structures); (2) the broad (and sometimes overlapping) ranges of variation of these characters within each of the major taxonomic groups, and (3) outlier taxa (discussed above) that introduce “exotic” combinations of characters in each group. As a result, the sets of features that define these groups are significantly overlapping. In fact, differences within each major group can be bigger than those between the groups. Thus, it is not surprising that our analysis including only discrete characters resulted in a large polytomy, especially since we included only one character on reproductive biology (scored for <20% of the species) and very few based on morphology (see also discussion under “Phylogeny and classical taxonomy”).

In this context, it is worth noticing that addition of continuous characters related to anatomy significantly improved phylogenetic resolution, as compared to the analysis based only on discrete characters. Importantly, the continuous characters we used seem to carry phylogenetic signal for the set of taxa analyzed here, i.e., actinostelic euphyllophytes associated with the origin of seed plants. This added resolution suggests that continuous characters based on anatomy are useful for understanding relationships between these particular Devonian species and among major groups of Devonian plants. Characters that are particularly consequential include the ratio between the primary xylem size and axis size as seen in cross sections, both in terms of diameter (character 0) and surface area (character 1), as well as a measure of the slenderness of primary xylem ribs (i.e., the ratio between basal rib width and overall xylem diameter in cross section; character 3) and the diameter of metaxylem tracheids (character 4). The conclusion that the continuous characters used here carry phylogenetic signal is supported by the fact that (1) we recovered an aneurophyte clade, a stenokolealean clade, and a seed plant clade, despite the lack of resolution for these groups in the analyses that used exclusively discrete characters, and (2) we recovered placements of most of the species that are largely congruent with the recommendations of previous taxonomic studies. Therefore, despite the low levels of support, our results provide reasonable hypotheses of relationships between the earliest seed plants and similar protostelic Devonian euphyllophytes.

Three clades including aneurophytes, seed plants, and stenokolealeans, respectively, are recovered not only in the analysis using discrete and continuous characters, but also in a high percentage of the MP trees obtained using only discrete characters, as demonstrated by the majority rule consensus tree (with the difference that in the latter tree the stenokolealean clade consists exclusively of Stenokoleos). The broad support for the three clades, even based primarily on vegetative anatomical characters, suggests that they may, indeed, represent natural taxa. However, relationships between the three groups are not fully and unequivocally resolved: whereas inclusion of continuous characters lends support to a closer relationship between Stenokoleales and seed plants, discrete characters alone seem to support (albeit only marginally: 52% of MP trees; Appendix 7) a closer relationship between aneurophytes and seed plants. Together, these suggest that fuller resolution of relationships with improved levels of support will require inclusion of additional characters coding for morphology and, especially, reproductive biology.

At this stage, the topology of the MP tree obtained using both discrete and continuous characters shows that continuous characters, phylogenetically significant at the level of our data set, tip the resolution of relationships in support of a closer relationship between Stenokoleales and seed plants, by recovering them as part of the same clade. This clade is part of a larger clade (including also Actinoxylon and Yiduxylon) that is defined by organography featuring axes characterized by bilateral symmetry (as indicated at a minimum by the symmetry of their vascular tissues)—the bilateral clade (fig. 1). Grouping of the seed plants and Stenokoleales in the same (more inclusive) clade is in agreement with Momont's (2015) results and indicates that seed plants may share a closer ancestor with the Stenokoleales than with aneurophytalean progymnosperms.

Although the Stenokoleales continue to be a poorly understood group of euphyllophytes, their placement in a clade that includes seed plants and is sister to aneurophytes implies that they are lignophytes (fig. 1), a possibility discussed previously on numerous occasions (Beck and Stein, 1993; Kenrick and Crane, 1997; Scheckler et al., 2006; Momont et al., 2016a). In turn, this implies that Stenokoleos, a genus in which secondary growth has not been demonstrated to date, may have at least harbored developmental potential for secondary growth. This inference is consistent with evidence for secondary growth in two other stenokolealeans, Brabantophyton and Crossia (Beck and Stein, 1993; Momont et al., 2016a). The same inference applies to Langoxylon and Gensel's (1984) plant, two plants nested deeply within the lignophytes, according to our results, but for which secondary growth has not been documented. As a general observation, the current absence of evidence for secondary growth in these species that are resolved in our phylogeny as lignophytes cannot be taken, on principle, as evidence for absence; and even less so if we consider that at least some regulatory mechanisms for secondary growth have been proven to be shared across a taxonomically much broader sampling than that considered here (Rothwell et al., 2008).

The placement of Actinoxylon as sister to the clade including seed plants and Stenokoleales implies that if Actinoxylon was confirmed as an archaeopteridalean progymnosperm, as proposed by Beck (1976), then progymnosperms as a whole would be polyphyletic. However, considering that Yiduxylon, a plant of equivocal taxonomic placement that shares several aneurophyte features, is recovered as sister to the clade including Actinoxylon, it is possible that Actinoxylon, Yiduxylon, and the aneurophyte clade represent a progymnosperm grade at the base of the clade that includes the seed plants and Stenokoleales (fig. 1). This paraphyletic progymnosperm group could also include Reimannia, another putative aneurophyte.

The age of Gensel's (1984) plant, which was dated based on spores, constrains the minimum age of the bilateral clade to the early Emsian (fig. 1). Although the oldest lignophyte fossil with demonstrated secondary growth, the aneurophyte Rellimia thomsonii, may be as old as the late Eifelian (Dannenhofer and Bonamo, 2003), the position of Gensel's (1984) plant as deeply nested within the lignophytes, constrains the minimum age of this clade, as well as that of the aneurophyte ancestor, to the early Emsian. This supports the late Emsian age proposed for what would be the oldest known Rellimia (and, by extension, progymnosperm) specimens, reported from the Devonian of Morocco by Gerrienne et al. (2010). The age of Gensel's (1984) plant also supports an early Emsian minimum age for the bilateral clade. The late Emsian Gothanophyton constrains the minimum age of Stenokoleales to the Emsian and the position of Triloboxylon arnoldii as sister to the seed plants constrains the minimum age of the seed plant ancestor to the Givetian.