A PHYLOGENETIC ANALYSIS OF DARRIWILIAN GRAPTOLITES, SUBORDER AXONOPHORA
MetadataShow full item record
Stratigraphic identification and correlation as well as fine timescale resolution for the Ordovician and Silurian periods are highly dependent upon the identification of graptolite fossils found in the strata. Taxonomic identification is greatly aided by an understanding of the evolutionary history of the taxa. The Axonophora, a suborder of the Graptoloidea, arose in the Darriwilian Stage of the Middle Ordovician and became the dominant planktic graptolites during the Late Ordovician. Chronostratigraphy of the Middle to Late Ordovician worldwide can be resolved in time units of less than a million years with the aid of axonophoran graptolites and other fossils. Although evolutionary relationships within the Late Ordovician Axonophora are well-established, their early phylogenetic history has been unresolved. Recent phylogenetic studies (Fortey et al. 2005, Mitchell et al. 2007) incorporated twenty-eight of the ninety-three Darriwilian Axonophora (not including stem Diplograptoidea) catalogued by Sadler et al. (2009). In this study, I have used morphological analysis of thirty-nine previously uncoded axonophoran taxa to develop a phylogenetic matrix of discrete codes, melded with pre-existing character sets from other researchers, based on the 117-character Mitchell et al. (2009) data set derived from the Mitchell et al. (2007) study of the Diplograptoidea (all of which are included in the Axonophora), but with additions and modifications to the character set. No characters have been deleted, to keep the data matrix consistent with previous studies. Several qualitative characters have been now quantified using continuous measurements. This study, thus, expands on previous analyses, thereby providing a more detailed and highly supported resolution of axonophoran clade phylogeny and evolution.The majority of added data has come from published material from which high-quality specimen photographs and camera lucida drawings were used to perform the character coding. I photographed and analyzed additional unpublished material from several museum repositories. Thirty-four distal characters were down-weighted due to strong correlation with early astogenetic characters. I used TNT 1.5 phylogenetic software to do the analysis, resulting in over 1 trillion trees being generated and assessed. My analysis consistently produced a single most parsimonious tree (MPT) that was highly resolved. Bootstrap analysis, however, showed that this tree is not robust, with most branches weakly supported. Nevertheless, an analysis of the next 1314 suboptimal trees found far more support for the MPT in the form of strong sub-tree agreement. Results of several previous studies are confirmed—for example, the levisograptids are found to be a paraphyletic stem group leading to the undulograptids and all other Axonophora (Maletz 2011a). One taxon—Levisograptus dicellograptoides—that was provisionally assigned to Levisograptus with some question by Maletz (1998) is found to be securely assignable to this genus. Two large clades are found to share a common ancestor with the stem lineage taxon Pseudoclimacograptus scharenbergi. Both also have stem lineage archiclimacograptids at their base. These include a well-supported Dicranograptacea including both the Dicranograptinae and the Dicaulograptidae. The normalograptids are found to be the most derived clade in the Darriwilian Axonophora. Support is found for assignment of Ar. ambiguus to the Archiclimacograptus genus (Maletz 2011b) and the separation of the Haddingograptus genus from Archiclimacograptus (Maletz 2011b). Several other recently published hypotheses are analyzed and found to be not supported by this phylogeny.I defined and evaluated a set of thirteen continuous characters to determine whether improvements in the phylogenetic models could be achieved by replacing discrete qualitative characters with quantitative ones. Due to TNT treating continuous characters as ordered, which they largely are not, I found it necessary to convert measured characters into discrete characters by using Finite Mixture Coding techniques to bin the thirteen measurements. I demonstrate that five of these thirteen characters are statistically improved over the existing discrete characters.