Rangeomorph Classification Schemes and Intra-Specific Variation: Are All

Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

Rangeomorph classiﬁcation schemes and intra-speciﬁc variation: are all characters created equal?

CHARLOTTE G. KENCHINGTON1,2* & PHILIP R. WILBY3 1Department of Earth Sciences, University of Cambridge, Downing Street, Cambridge CB2 3EQ, UK 2Present address: Department of Earth Sciences, Memorial University of Newfoundland, Prince Philip Drive, St John’s, NL, A1B 3X5, Canada 3British Geological Survey, Environmental Science Centre, Nicker Hill, Keyworth, Nottingham NG12 5GG, UK *Correspondence: [email protected]

Abstract: Rangeomorphs from the Ediacaran of Avalonia are among the oldest known complex macrofossils and our understanding of their ecology, ontogeny and phylogenetic relationships relies on accurate and consistent classification. There are a number of disparate classification schemes for this group, which dominantly rely on a combination of their branching characters and shape metrics. Using multivariate statistical analyses and the diverse stemmed, multifoliate rangeomorphs in Charnwood Forest (UK), we assess the taxonomic usefulness of the suite of characters currently in use. These techniques allow us to successfully discriminate taxonomic groupings without a priori assumptions or weighting of characters and to document a hitherto unrecognized level of variation within single taxonomic groups. Variation within the currently defined genus Pri- mocandelabrum is too great to be realistically assigned to different species and may instead reflect primary character diversity, ontogenetic changes in character state or ecophenotypic variability. Its recognition cautions against generic-level diagnoses based on single differences in character state and will be crucial in understanding the mode of growth of these enigmatic organisms.

Supplementary material: Data tables, deﬁnition of the characters used in the analyses, and detailed descriptions and breakdowns of methods and results are available at https://doi.org/10. 6084/m9.ﬁgshare.c.3726937

Gold Open Access: This article is published under the terms of the CC-BY 3.0 license.

Rangeomorphs were a major component of early phylogenetic position (e.g. Liu et al. 2015; Dufour Ediacaran macroscopic communities (c. 580– & McIlroy 2016). 557 Ma), even dominating many of the preserved The unique anatomy of rangeomorphs (Fig. 1) assemblages in Newfoundland, Canada and Charn- and their typical preservation as mere (albeit high wood Forest, UK (e.g. Liu et al. 2015). They are resolution) external moulds (Narbonne 2005) has characterized by a modular frond, the architecture thus far confounded attempts to develop a univer- of which appears self-similar over at least four sally accepted classification scheme. There is an orders of subdivision (Narbonne 2004; Brasier urgent need for such a scheme to support increas- et al. 2012). Despite this apparently simple organi- ingly sophisticated analyses of their ecological zation, rangeomorphs achieved considerable mor- interactions, environmental tolerance and develop- phological diversity. They exhibit variations in the mental biology (Darroch et al. 2013; Mitchell number of growth poles, in the degree of insertion, et al. 2015). Current taxonomic frameworks for ran- inflation and organization of their branches, and in geomorphs place the greatest weight on one of the the presence or absence of a stem/stems (Brasier two main sources of available characters: (1) cate- et al. 2012; Laflamme et al. 2012; Hoyal Cuthill gorical characters, such as a branching architecture & Conway Morris 2014). Of the major Ediacaran (see Fig. 2b; see Supplementary material table 1 clades recognized by Erwin et al. (2011), the Ran- for a full description of characters), the presence/ geomorpha have received perhaps the greatest absence of a holdfast and the number of growth attention. Nevertheless, almost every aspect of poles (Brasier & Antcliffe 2009; Narbonne et al. their biology remains contentious, including feed- 2009; Brasier et al. 2012; Laflamme et al. 2012; ing, growth, reproduction, internal anatomy and Liu et al. 2016); and (2) continuous characters

From:Brasier, A. T., McIlroy,D.&McLoughlin, N. (eds) 2017. Earth System Evolution and Early Life: A Celebration of the Work of Martin Brasier. Geological Society, London, Special Publications, 448, 221–250. First published online March 30, 2017, https://doi.org/10.1144/SP448.19 # 2017 NERC. The British Geological Survey and University of Cambridge, Department of Earth Sciences. Published by The Geological Society of London. Publishing disclaimer: www.geolsoc.org.uk/pub_ethics Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

222 C. G. KENCHINGTON & P. R. WILBY

Fig. 1. Examples of stemmed, multifoliate rangeomorphs from Bed B, Charnwood Forest. (a) Dumbbells. (b–i) Individuals referable to Primocandelabrum showing the variation present in terms of crown size (width and height), disc size, stem length and stem width, relative to total height: (b) BB (GSM 105872), (c) 6A2b (GSM 105969b), (d) 10C3c (GSM 106051c), (e) 10B8 (GSM 106049), (f) 19A4 (GSM 106040), (g) 3(1) (GSM 105945), (h) 6A2d (GSM 105969d), (i) 20Ba (GSM 106039). (j, k) Feather dusters 8C3 (GSM 106045) and 8C1 (GSM 106043), respectively. Scale bars are all 2 cm. All images # NERC 2017 British Geological Survey and University of Cambridge, Department of Earth Sciences, reprinted with permission. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 223

Fig. 2. (a) Continuous characters: 1, disc width; 2, disc length; 3, width of stem at base; 4, width of stem at inflection point; 5, width of stem at base of branches; 6, height of stem to inflection point; 7, height of stem to base branches; 8, crown width (widest point); 9, crown length; 10, length of left branch; 11, length of right branch; 12, total height (from centre of disc/base of stem). Dashed white line shows the outline of the fossil; blue line, first order branch outline; purple line, second order branch outline; pink line, third order branch outline. See online version for colour. (b) Categorical characters: displayed/rotated, both rows/one row of lower order branches are visible; furled/unfurled, branch edges are visible/tucked in giving a scalloped/smooth outer margin; inflation, width of the branch increases distally/medially/proximally; radiating/subparallel, branches emerge from central axis at increasing/similar angles; concealed/unconcealed, central axis is visible/concealed (dashed lines) by the branches arising from it. Terminology after Brasier et al. (2012), see Supplementary material, table 1. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

224 C. G. KENCHINGTON & P. R. WILBY

(see Fig. 2a), such as shape metrics (length to width Rangeomorph terminology is reviewed in Supple- ratios) and their overall morphology (Laflamme mentary material table 1. et al. 2004; Laflamme & Narbonne 2008). Each method has particular merits and disadvantages. For example, the comparative resilience of categor- Multifoliate rangeomorphs from ical characters to tectonic overprinting (shearing Charnwood Forest and distortion) and their possible application to fragmentary fossils makes them attractive (Brasier Stemmed, multifoliate rangeomorphs (Fig. 1) are et al. 2012), but only a few taxa are sufficiently abundantly preserved on Bed B in Charnwood For- well preserved to determine the number of branches est (Wilby et al. 2011; n ¼ 64). They are all oriented present throughout ontogeny. Equally, although in the same direction (s ¼ 158), suggesting that they the overall morphology can differentiate forms were felled and covered by the same event bed and, above the species level (Hoyal Cuthill & Conway as such, will have experienced similar flow and bur- Morris 2014), shape metrics are not easily trans- ial conditions and the same tectonic processes. ferred between different tectonic settings. This substantially reduces the likelihood and magni- Liu et al. (2016) have argued that categorical tude of any artefacts associated with differential characters are more appropriate to genus-level dia- modes of preservation being introduced into the gnoses and that continuous characters should be statistical analysis. Each specimen consists of a disc, restricted to differentiating species. This was based a discrete stem and a multifoliate crown, and falls on the inference that the branching architecture into one of three morphotypes based on general reflects genetic controls and that shape metrics are appearance. The crown is here defined as that part more susceptible to environmental influences (and of a multifoliate rangeomorph that consists of the thus convergence), which relies on the consistency branches. Fifty-three specimens are referable to Pri- of branching between multiple specimens. Accord- mocandelabrum (Fig. 1b–i), a taxon first described ingly, forms that differ in even a single branching from the Bonavista Peninsula, Newfoundland (Hof- character (e.g. Vinlandia v. Beothukis) are classified mann et al. 2008); nine specimens conform to the as different genera, with the possibility that they ‘dumbbell-like taxon’ of Wilby et al. (2011; Fig. reflect different species within one genus, or even 1a); and two specimens (Fig. 1j, k) are referable to variation within a species, being rejected (Brasier the so-called ‘feather dusters’ from Mistaken Point, et al. 2012). Very few species level diagnoses Newfoundland (Clapham & Narbonne 2002). Vari- note variation in branching architecture within a ation in the overall morphology and branching taxon, although local variation has been noted and architecture within the Bed B Primocandelabrum attributed to either taphonomic overprint or ontog- sp. individuals led Wilby et al. (2011) to suggest that eny. Examples include: the presence of displayed several taxa or forms were represented on Bed B. branches in Avalofractus, where they are more Primocandelabrum is principally diagnosed by typically rotated, as well as the pivoting of individ- its coarse branching and the triangular outline of ual branches (Narbonne et al. 2009); local unfurling its crown, with P. hiemaloranum being separately in Bradgatia (Brasier et al. 2012); and the rever- distinguished on the basis of its Hiemalora-like sal of branch overlap in Charnia (Wilby et al. holdfast, which has numerous sinuous to linear 2015). Such variation may, potentially, be more rays (Hofmann et al. 2008). The type material widespread, but it has not been systematically from Bonavista consists of just eight specimens, investigated. all of which are poorly preserved. Only two main Phylogenetic relationships within the Rangeo- branches are generally visible in this material, morpha will remain opaque, and the metrics of spe- none of which reveals the taxon’s branching archi- cies richness or diversity will have uncertain value, tecture. As such, no formal diagnosis of the branch- until the taxonomic robustness of individual charac- ing pattern exists by which this genus or its ters is better known. Herein, we use multivariate constituent species may be definitively identified statistical analysis to test the extent to which branch- at other localities. The Bonavista specimens are ing characters correlate with continuous morpholog- often reconstructed as a two-dimensional fan or ical characters and to assess which characters are candelabrum in life (Hofmann et al. 2008), while most taxonomically useful. Our analyses use the the single specimen assigned to Primocandelabrum abundant and well-preserved assemblage of multi- from NW Canada suggests a more brush-like foliate, stemmed rangeomorphs present on Bed B appearance (Narbonne et al. 2014). In the specimens in Charnwood Forest (Wilby et al. 2011; Fig. 1). from Charnwood Forest that fit the broad diagnosis This assemblage includes variability in many of of Primocandelabrum, it is clear that the crown con- the character states used in rangeomorph taxonomy sists of three first order branches that split from the and therefore provides an opportunity to reassess main stem at a single point (Fig. 1b–i). Each of the species concept within the Rangeomorpha. these first order branches is referred to here as a Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 225 folium, plural folia. All the Primocandelabrum sp. et al. 2013). The approach adopted in the present from Charnwood Forest have rotated and unfurled study is to combine clustering methods with multi- first order branches, but differ in their other branch- variate analytical techniques. Various iterations of ing characters. We note that none of the Primocan- the analyses were run to determine the extent of delabrum specimens from Bed B are referable to influence of outliers and particular characters on Primocandelabrum hiemaloranum, although an iso- the resultant groupings. All tests were run in the R lated disc with rays is known from nearby Bradgate statistical package, version 3.1 (R Core Team Park (Wilby et al. 2011). 2014). The methods and iterations run are discussed In contrast with the Primocandelabrum sp. spec- in the Supplementary material. trum on Bed B, the dumbbell has a crown that is The main advantage of the statistical techniques circular in outline and consists of upwards of five outlined here is that they allow the characterization first order branches, which are curvilinear in outline of clusters of specimens based on all characters and are entirely concealed along their lengths; a and on subsets of characters, and that they weight proportionally longer stem and large holdfast disc all characters equally. They also provide a way of (Fig. 1a); and displayed first order branches. testing the correlation of continuous and categorical The Mistaken Point feather dusters are a poorly characters and of identifying statistically meaningful defined, bucket group for small, brush-like fronds, groups within a population. If categorical characters which have numerous (more than three) first order correlate with each other, and with the continuous branches emanating from a shared point at the top characters, then the groups can be considered to be of the stem and a small, bulb-shaped holdfast (Clap- morphologically distinct and are therefore more ham & Narbonne 2002; Mason & Narbonne 2016; likely to be taxonomically meaningful than ones Fig. 1j, k). Two of the .200 known specimens where the characters vary essentially randomly. of feather dusters from the Mistaken Point E surface have been described as Plumeropriscum hof- manni, but the remainder still lack systematic Materials and methods description (Mason & Narbonne 2016). The rela- tionship of feather dusters to Primocandelabrum Jesmonite resin casts were made of 64 specimens of remains uncertain, although the presence of more stemmed, multifoliate taxa from Bed B in Charn- than three folia in Plumeropriscum is considered wood Forest. The original moulds taken from the to be diagnostic at the generic level (Mason & Nar- bedding surface and the corresponding casts are bonne 2016). The gross morphology, number of held at the British Geological Survey, Keyworth, branches and overall appearance of Plumeropris- Nottingham, UK. cum is superficially similar to that of the two feather Tests were run on all specimens that fitted into dusters on Bed B. They are distinct from all speci- the broad ‘stemmed, multifoliate’ morphospace for mens of the dumbbells, even those of comparable which both categorical and continuous characters size. The feather dusters have smaller holdfasts, pro- could be recorded. Casts were studied under con- portionally shorter stems, a crown that is more del- trolled lighting conditions using a microscope toid in shape and branches that appear more tightly equipped with a camera lucida. Continuous charac- constrained. However, the branching architecture of ters were measured by hand to the nearest millimetre Plumeropriscum has not been described and cannot using a ruler (Fig. 2a). First and second order be determined with confidence from the published branching characters (Fig. 2b) can be discerned in photographs. The inclusion of Plumeropriscum in 48 specimens (40 Primocandelabrum sp., six dumb- this study would require additional data pre- bells and two feather dusters). These characters treatment (retro-deformation), which could poten- were diagnosed and recorded for each of these spec- tially impart bias to the analyses, and as such should imens, including any variation within a specimen be treated separately. (Supplementary material table 2). The identification of branching characters probably represents the largest source of primary error due to taphonomic Statistical techniques overprint and complexities arising from the compound nature of the fossils. To minimize these Taxonomy has sporadically made use of statistical biases, only those characters that were consistent, and computational approaches to deal with large or that varied consistently, across the specimen datasets. The techniques used most commonly were included in the statistical analyses. Third-order include dimensionality reduction analyses, such as and higher characters can only be discerned in a few principal components analysis and multiple corre- individuals and so were not included in the analyses. spondence analysis (Lajus et al. 2015), as well as Characters for which states could not confidently be clustering algorithms (Jardine & Sibson 1971; determined were left blank to minimize any errors Dunn & Everitt 2004; Bonder et al. 2012; Pagni introduced due to misidentification. All the Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

226 C. G. KENCHINGTON & P. R. WILBY descriptive characters proposed by Brasier et al. confidence, and which are therefore left blank. (2012) were used in these analyses, in conjunction This necessarily results in a dataset that has a rela- with the ‘constrained/unconstrained’ character of tively large number of missing values, which is Narbonne (2004). Although some of these charac- problematic for any statistical test. The missing val- ters may not be taxonomically meaningful, the ues were imputed following guidelines for best effect of the removal of any character from the anal- practice (Josse & Husson 2012; Josse et al. 2012). yses would be to impart a subjective bias to the analyses. As dumbbells can be confidently distinguished Character selection from Primocandelabrum sp. based on gross morphology, they are used to assess the validity of the Some of the continuous characters are strongly cor- statistical tests, although no assumption is made related with one another. For instance, the length of regarding the taxonomic rank of this difference. the left and right branches of Primocandelabrum are the same length in most specimens, and also corre- Data pre-treatment late with the height and/or width of the crown (depending on whether it is taller than it is wide). The data were not retro-deformed prior to analysis Similarly, the width of the stem measured at any due to a lack of independent strain markers on Bed point along its length will tend to be correlated B, meaning that holdfasts are the only structures with the width at any other point along the stem; available for strain analysis. For all multifoliate those specimens with narrower stems tend to be nar- specimens, the correlation between disc length and rower along their entire length. To avoid any bias disc width is even across the surface (R2 ¼ towards characters that are themselves strongly cor- 0.9915). Together with the fact that the holdfasts related due to the double-correlation of the analyses exhibit little variation in ellipticity (s ¼ 2.6%, (Dillon & Goldstein 1984), most analyses were run Wilby et al. 2015), this indicates that tectonic on a reduced character matrix using only those char- shear is even across the surface, that it operated par- acters whose proportions do not innately co-vary allel to the long axis of the fossils and that it affected with one another. all specimens evenly. If the holdfasts were originally circular (the common assumption in the Outlying individuals retro-deformation of Ediacaran fossils; Wood et al. 2003), then all the fossils have been shortened to The dataset includes organisms of a significantly 77% of their original height (i.e. a compaction of larger size than the main population, but a lack of 23%) parallel to their long axis. Although this will ontogenetic intermediates between the two size have affected the absolute proportions, it will have groups (cf. Wilby et al. 2015). Accordingly, any done so equally for all the studied fossils and so changes in morphology or mode of life attributable the relative proportions may be robustly compared. to ontogeny cannot be constrained by the present Several aspects of the data make it challenging to dataset. The presence of such outlying individuals work with, requiring various data pre-treatments. can influence the results of each of the described The variance of each continuous character was methods (cf. Dillon & Goldstein 1984) and so sev- investigated using the makeProfilePlot function eral iterations of the data were run to investigate (Coghlan 2014; Fig. 3). The total size of the individ- the extent of the influence of these factors. If any ual is a source of great variance in the data and nec- character differences are ontogenetic, the lack of essarily affects the absolute values of all other specimens recording intermediate ontogenetic continuous characters. Accordingly, all the continu- stages precludes confirmation of such patterns. ous characters were divided by total height (taken from the top of the crown to the base of the stem, Analyses Fig. 2a) to standardize them. As only well-preserved specimens with clear distal margins were used in Principal components analysis was run on continu- this study, we consider that this step does not bias ous characters; multiple factorial analysis was run the analyses. As categorical characters (here, on categorical characters and factorial analysis of branching architecture) are either binary or consist mixed data was run on the mixed datasets, i.e. of a small number of discrete categories, they have those with both categorical and continuous charac- a proportionately large variance compared with ters. Hierarchical clustering (HCPC) was performed the continuous characters. The method used here on the results of the principal components analysis, scales the data as a part of its algorithm, scaling multiple factorial analysis and factorial analysis all values to unit variance so that all characters influ- of mixed data analyses, producing hierarchical den- ence the results equally. The majority of the fossils drograms (Husson et al. 2010). The influence of the included in this study are incompletely preserved, or iteration on the results was examined by determining include characters that cannot be determined with the percentage of individuals that were assigned to Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 227

Crown width Crown width (a) Crown height (b) Crown height i Length left Disc width Length right Stem height (c-i) Disc width Stem width (at i) Disc height Stem height (c-i) Stem height (c-bb) Stem width (base) Stem width (at i) Stem width (top flare) vectori vectori 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5

0 102030405060 0 102030405060

Index Index

Crown width Crown width (c) Crown height i (d) Crown height Disc width Length left Stem height (c-i) Length right Stem width (at i) Disc width Disc height Stem height (c-i) Stem height (c-bb) Stem width (base) Stem width (at i) Stem width (top flare) vectori vectori -2-101234 0.0 0.5 1.0 1.5

0 102030405060 0 1020304050

Index Index

Fig. 3. Proﬁle plots: graphic representations of the variance within each continuous character. Vectori is a measure of the variance and Index refers to the number of the specimen (in random order). (a) All specimens, all continuous characters, data not scaled to unit variance (see Supplementary material 1.2). The two outsize specimens are at the right-hand edge of the plot. The greatest variance is within crown width and disc width, length left and length right. (b) As (a), but on the reduced character matrix, the difference in variance between crown width and disc width compared with the other characters is more pronounced. (c) As (b), but scaled to unit variance. (d) All Primocandelabrum and feather dusters, including outsize specimen BB (right-most data points). the same group for each pair of iterations. The num- character state used to describe that cluster (where ber of clusters was chosen based on an analysis of 100% indicates that all individuals in the cluster dis- inertia gain, a measure of the within-group variance play that character); and (3) the average value for a (plotted as a histogram of variance v. number of clus- continuous character within a cluster compared ters; see insets, Figs 4–6). The greatest jump in iner- with the average for all specimens (Tables 1 & 2). tia gain (i.e. the greatest decrease in within-group variance) is taken as the best node at which to cut the dendrogram into clusters. For each cluster, three Results factors describe the success of the cluster discrimination: (1) the percentage assignment of individuals The results of the analyses including dumbbells are displaying a particular character state to a cluster presented in Figure 4, whereas the analyses run on characterized by that character state (where 100% Primocandelabrum only are presented in Figure 5 indicates that all individuals which display that char- and on Primocandelarum and feather dusters in Fig- acter are placed in the cluster); (2) the percentage ure 6. The characteristics for each cluster resulting of individuals within a cluster that display a given from all analyses run are listed in Tables 1 and 2. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

228 C. G. KENCHINGTON & P. R. WILBY

(a) continuous and categorical characters (b) continuous and categorical characters cluster 1 cluster 2 8C2 cluster 3

01234 3(1) inertia gain 10C3c 10C3a BB 19B1cluster 2 19C4 3(2) 6 13A8 6A1 6B2b 13A4 19A1 6a2 l 10A23D8 3D7cluster 3 4A38B4 5C6 10C3b2B1 8B3b Dim 2 (18.36%) 10B81B15E3 5F5 cluster 1 5A4 3E1(1) 5C57A5 10A75C1 19A37A4 3C3 10C8 5E2 10B3 8C3 8C1 5G34B319A4 4A1b10b10 (right) 5F3 -20246 20Ba 01234 6 BB 5F3 5F5 3(1) 3(2) 4A3 8B4 7A5 2B1 4B3 1B1 6A1 7A4 5A4 5E3 5C6 5C1 3C3 8C1 8C3 8C2 3D8 3D7 5C5 5E2 5G3 6a2 l 10B8 19A3 19B1 13A8 19A4 19A1 6B2b 10A2 10A7 13A4 10B3 8B3b 20Ba 10C8 19C4 10b10 10C3c 10C3b 10C3a 3E1(1) -4 -2 0 2 4 6 8 4A1b (right) Dim 1 (41.89%)

(c) categorical characters only (d) categorical characters only cluster 1 cluster 1 cluster 2 cluster 3 10C3a cluster 4 cluster 2 cluster 5 8C2 cluster 6 cluster 3

0.00 0.10inertia 0.20 0.30 gain 8B4 cluster 1 19C4 10C3b7A54A310A213A8BB 19B13(1)2B13(2) 19A1 10C3c 6a21B1 l 6 6B2b5C6 5G35C55C1 13A4 3E1(1) 6A1 cluster 6 19A3 5E25F5 3D75A4 10C83D810A77A48B3b 10B819A45E33C34B3 cluster 2 Dim 2 (12.08%) cluster 4 10B3 4A1b (right) 10b10 8C1 cluster 5 5F3 8C320Ba 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 6 -1.0 -0.5 0.0 0.5 1.0 BB 5F3 5F5 3(1) 3(2) 5A4 7A4 2B1 1B1 8B4 7A5 4A3 4B3 6A1 5E2 5E3 3D8 5C1 8C3 3C3 5C6 3D7 5C5 8C2 8C1 5G3 6a2 l 13A4 10B8 19B1 20Ba 8B3b 10A2 19A4 10B3 19A1 6B2b 13A8 10A7 19A3 10C8 19C4 10b10 10C3c 10C3b 10C3a 3E1(1) -1.0 -0.5 0.0 0.5 1.0 1.5 2.0

4A1b (right) Dim 1 (43.92%)

(e) continuous characters only (f) continuous characters only cluster 1 cluster 2 8C2 cluster 3 BB

1.5 3(1)19B1 0.0 0.5inertia 1.0 1.5 gain 10C3c

19C4 3D8 cluster 2 3(2) 6 6B2b 8C1 10C3a 19A1 13A4 13A8 6A1 1.0 6a2 l 3D7 8B3b5E3 10B8 cluster 1 10A2 10A7 19A3 8C3 4A3 2B1 5A44A1b (right) 5F5 5C6 10b105C1 10C3b 5C5 Dim 2 (26.63%) 3C3 cluster 3 7A4 5E2 8B4 0.5 3E1(1) 20Ba 1B1 10C8 19A45F3 10B3 7A55G3

4B3 -2-10123 0.0 6 BB 5F3 5F5 5E3 5E2 3(1) 3(2) 8B4 2B1 1B1 5A4 4A3 6A1 4B3 7A5 7A4 3D7 5C6 5C5 8C3 8C2 8C1 5C1 3D8 3C3 5G3 6a2 l 19A1 10B3 20Ba 19A4 6B2b 10A2 13A4 10B8 13A8 19A3 19B1 10A7 8B3b 10C8 19C4 10b10 10C3c 10C3a 10C3b 3E1(1) -4 -2 0 2 4A1b (right) Dim 1 (55.39%)

Fig. 4. Results of the cluster analysis (HCPC) on the dataset, including all individuals for which categorical and continuous characters could be determined. All values were standardized to total height. (a) Cluster dendrogram and (b) factor map for analysis on continuous characters (reduced character matrix) combined with categorical characters; (c) cluster dendrogram and (d) factor map for categorical characters only; (e) cluster dendrogram and (f) factor map for morphological characters (reduced character matrix) only. In (a), inertia gain supports division into two or three clusters; in (c) it supports division into two or six clusters and in (e) it supports two or three clusters. Dumbbells plot separately in (a), (b), (c) and (d), but with other specimens in (e) and (f). Schematic diagrams describe the clusters that match their colour (see online version for colour): where a categorical character trait was not signiﬁcant in describing the cluster, the most common state for the cluster was used; where more than one state is common in the group, both states are depicted. Where a continuous character did not signiﬁcantly describe the cluster, average values for the population were used. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 229

(a) continuous and categorical characters (b) continuous and categorical characters

1.5 cluster 1 cluster 2 cluster 2 cluster 3 8C2

4 cluster 4 13A4

0.0 0.4inertia 0.8 1.2 gain

4A1b (right) 3(1) 1.0 2 10b10 5C1 19C4 cluster 4 cluster 3 13A8 8B3b 6A1 10C3c 5E3 19B1 20Ba 5F5 19A3 3C3 10A7 Dim 2 (24.84%) 5E2 3D8

0.5 19A4 10C3a 10B8 3(2) cluster 2 1B1 10A2 cluster 1 4A3 6a2 l 10B3 7A4 2B1 5F3 10C8 5C5

-2 0 7A5 4B3 5G3 10C3b 8B4 0.0 5F3 5F5 5E2 5E3 3(1) 3(2) 1B1 6A1 7A4 7A5 8B4 4A3 2B1 4B3 3C3 8C2 5C1 5C5 3D8 5G3 6a2 l 13A8 10B8 10B3 20Ba 19A4 10A7 13A4 10A2 19A3 8B3b 19B1 10C8 19C4 10b10 10C3c 10C3b 10C3a -6 -4 -2 0 2 4 4A1b (right) Dim 1 (29.85%)

cluster 1 cluster 1 cluster 2 cluster 3 cluster 2 5F3 cluster 4 0.15 cluster 5 cluster 3 cluster 4 0.00inertia 0.10 gain cluster 6 20Ba 0.10

cluster 1 cluster 6 8B4 7A5 Dim 2 (15.45%) 10A2 4A3 10C8 19B1 3D85E2 10C3a 10C3b10C3c 10A7

0.05 3(1) 5G3 7A4 5F5 10B3 13A8 2B1 10B8 4A1b (right) cluster 2 5C55C1 19C4 3(2) 3C3 19A4 1B1 6a2 l 4B3 5E3 6A1 8C2 19A3 8B3b cluster 5 cluster 3 10b10 -0.5 0.0 0.5 1.0 1.5 13A4 0.00 5F3 5F5 3(1) 3(2) 8B4 1B1 7A4 7A5 4A3 2B1 4B3 6A1 5E2 5E3 5C1 8C2 5C5 3D8 3C3 5G3 6a2 l 10B8 13A8 13A4 19A4 19B1 10A2 10A7 8B3b 19A3 10B3 20Ba 19C4 10C8 10b10 10C3c 10C3a 10C3b

4A1b (right) -1012 Dim 1 (38.13%)

(e) continuous characters only (f) continuous characters only cluster 1 cluster 2 8C2 cluster 3 3(1)

0.0 0.4inertia 0.8 1.2 gain 19B1

cluster 3 10C3c 3D8 3(2) 10C3a 19C4 13A8 10B8 6a2 l 13A4 5E3 6A1 8B3b 4A1b (right) 10A2 2B1 19A3 5F5 10A7 4A3 10C3b

Dim 2 (35.80%) 5C1 cluster 1 cluster 2 5C5 10b10 7A4 8B4 5E2 1B1 20Ba 3C3 19A4 10B3 10C8 5F3 7A5 5G3

4B3 0.0 0.2 0.4 0.6 0.8 1.0 1.2 5F3 5F5 3(2) 3(1) 4B3 4A3 1B1 2B1 7A5 6A1 8B4 7A4 5E3 5E2 3D8 5C5 3C3 8C2 5C1 5G3 6a2 l -3 -2 -1 0 1 2 3 4 8B3b 10B3 19B1 19A4 10A7 19A3 10A2 13A4 10B8 13A8 20Ba 10C8 19C4 10b10 10C3c 10C3b 10C3a

4A1b (right) -4 -2 0 2 Dim 1 (40.03%)

Fig. 5. Results of the cluster analysis (HCPC) on the dataset including all individuals for which categorical and continuous characters could be determined, excluding dumbbells, feather dusters and outlying specimen BB. All values were standardized to total height. (a) Cluster dendrogram and (b) factor map for analysis on continuous and categorical characters; (c) cluster dendrogram and (d) factor map for categorical characters only; (e) cluster dendrogram and (f) factor map for continuous characters only. Inertia gain plots support division into four clusters in (a), two, three or six clusters in (c) and two or three clusters in (e). Schematic diagrams describe the clusters that match their colour (see online version for colour): where a categorical character trait was not significant in describing the cluster, the most common state for the cluster was used, and where more than one state is common in the group, both states are depicted. Where a continuous character did not significantly describe the cluster, average values for the population were used. otnoscaatr ol edtrie,ecuigdmblsaduigterdcdcaatrmti.Alvalues All matrix. character reduced (a the height. using total and to dumbbells standardized excluding were determined, be could characters continuous 6. Fig. aeoia hrces (c characters; categorical ntegop ohsae r eitd hr otnoscaatrddntsgicnl ecietecluster, common the is describe state significantly one not than did more used. character where were continuous and significant the population a used, not describe the Where was was diagrams for depicted. cluster Schematic trait values are the character (e). average states for categorical in both state clusters a group, common where five the most colour): or in the for three cluster, version two, the online and describing (see (c) in colour in their clusters match six that or clusters three two, (a), in clusters (f and dendrogram 230 (e) (c) (a)

0.0 0.5 1.0 1.5 0.00.20.40.60.81.0 0.00 0.05 0.10 0.15 Downloaded from eut ftecutraayi HP)o h aae nldn l niiul o hc aeoia and categorical which for individuals all including dataset the on (HCPC) analysis cluster the of Results continuous andcategoricalcharacters 13A4 continuous charactersonly 8C1 categorical charactersonly 6A1 10C3a 10C3c 4A1b (right) 8C2 13A4 19C4 19C4 10b10 13A8 3(2) 5C1 10C3a 6a2 l 10B3 19B1 1B1 5G3 3(1) 5G3 4B3 BB 5C5 8C2 7A5 5C1 19A4 6a2 l atrmpfrcniuu hrcesol.Ietagi lt upr iiinit w rthree or two into division support plots gain Inertia only. characters continuous for map factor ) BB 3(2) 5E2 2B1 8B4 20Ba 13A8 2B1 1B1 3(1) 10C3b 5F3 10C3c 10A2 3C3 8B4 BB 4A3 lse edormad(d and dendrogram cluster ) 10C3b 7A5 8C2 19B1 19A3 19B1 7A5 1B1 3(1) 10A2 5G3 6a2 l 4A3 10C3a 4B3 13A4 10B3 10C3b 6A1 2B1 10B8 http://sp.lyellcollection.org/ 5E2 10C8 3D8 3D8 5C5 8B4 10C8 10C8 5C5 10A7 7A4 7A4 5F5 lse edormad( and dendrogram Cluster ) 19C4 7A4 4A1b (right) 10C3c 10B8 10b10 3(2) 10B3 5C1 .G ECIGO .R WILBY R. P. & KENCHINGTON G. C. 3D8 5E3 10A7 5F5 4B3 19A4 10B8 3C3 5E2 8B3b 19A4 3C3 5E3 0.0 0.4 0.8 8B3b 5F5 0.00 0.10 8B3b 0.0 0.5 1.0 1.5 13A8 5F3

inertia gain 5E3 inertia gain 6A1 8C3 inertia gain 8C1 4A3 20Ba 20Ba 10A2 8C1 10A7 5F3 10b10 8C3 19A3 19A3 8C3 4A1b (right) atrmpfrctgrclcaatr ny (e only; characters categorical for map factor ) 5F3 (d) (b) (f)

Dim 2 (14.93%) Dim 2 (23.77%) Dim 2 (33.61%)

b -0.5 0.0 0.5 1.0 1.5 -2024 -2 -1 0 1 2 3 atrmpfraayi ncniuu and continuous on analysis for map factor ) cluster 3 cluster 2 cluster 1 byguestonOctober1,2021 10C3a cluster 6 cluster 3 cluster 2 cluster 1 cluster 6 cluster 5 cluster 4 cluster 3 cluster 2 cluster 1 4A1b (right) -1 20Ba 4- 4 2 0 -2 -4 BB 8C1 6a2 l 8B4 8B4 10C3b7A5 10C3b continuous andcategoricalcharacters 10b10 19B1 cluster 5 19C4 cluster 1 2B1 13A8 2- 4 3 2 1 0 -1 -2 8C2 cluster 3 cluster 2 cluster 1 10C8 4A3 5C5 4B3 3(2) 5F3 10C3a 3(1) 7A4 5G3 continuous charactersonly 4A3 cluster 2 10A2 10C3c 7A5 6A1 3(2) 2B1 5C1 3C3 5E2 3D8 categorical charactersonly 19A3 5G3 10B3 10B8 13A4 19A3 5C1 10A7 1B1 19B1 5C5 8C3 10B3 19A4 6a2 l 10A2 10C3c Dim 1(27.79%) 3D8 5G3 4B3 5F5 8C3 19A4 5E3 0 5F5 1B1 10A7 8C1 4A1b (right) cluster 1 1B1 8B3b 19A3 Dim 1(42.91%) 3(1) 7A5 Dim 1(36.48%) 10C8 3C3 10B3 cluster 3 8B3b cluster 3 7A4 10B8 5F3 7A4 10C8 10B8 5E2 cluster 2 5E2 5F3 13A8 10A7 10b10 4B3 5F5 5E3 BB 19A4 cluster 2 5C5 5E3 6A1 13A8 3D8 cluster 3 3C3 8B3b 19C4 20Ba 19C4 5C1 8C3 8C2 10A2 4A3 cluster 1 2B1 10b10 10C3b 20Ba 8B4 6a2 l 19B1 cluster ) 1 10C3c 13A4 3(2) 3(1) 8C2 4A1b (right) BB 10C3a 8C1 13A4 cluster 4 6A1 2 Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 231

Summary of results percentage of variance described by each dimension was comparable to that for the full character matrix, All tests discriminated dumbbells to some degree. but slightly higher when the continuous characters In fewer than 10% of the iterations run, one or two only were considered (55% for the reduced and Primocandelabrum or feather duster specimens are 27% for the full matrix, compared with around included in the dumbbell cluster, or the smallest 40% for the reduced and 20% for the full matrix two dumbbells are placed in a different cluster to using only continuous characters). The influence of the majority of that group. Even then, the dumbbells iteration on these aspects is presented in the follow- all plot close together in the principal components ing sections, focusing on the assignment of individu- space. The separation of dumbbells from Primocan- als within the group assigned to Primocandelabrum. delabrum is further supported by profile plots of var- This is quantified for the hierarchical cluster analyses iance (Fig. 7a, b) and bivariate plots (Fig. 7c–f). as the percentage of individuals that are assigned The morphological and branching characters of the to the same group for each pair of iterations. Signifi- feather dusters place one specimen (8C1, Fig. 1k) cance refers to the 95% confidence level (i.e. most frequently with the dumbbells and one (8C3, p , 0.05) throughout. The utility of particular char- Fig. 1j) within the Primocandelabrum clusters. acters in defining clusters is summarized in Table 5. The clustering consistently supports the placement of Primocandelabrum individuals into two or Hierarchical clustering three clusters by analysis of inertia gain (inset, Figs 4–6) that are distinct from dumbbells. The pro- In the following descriptions, all descriptors of con- file plots of the variance within the data show a tinuous characters are proportional compared with smoother, more consistent correlation of continuous the mean values of all individuals. Individual char- traits when sorted into three groups as determined acters that are not mentioned in the descriptions by cluster analysis than when the individuals are do not significantly describe the cluster – that is, unsorted (Fig. 7a, b). However, within Primocande- the continuous characters for the cluster are not sig- labrum, there is considerable variation in the com- nificantly different from the mean values for the position of each cluster (both individuals placed group – and the categorical characters are not sig- into the cluster and the character descriptions of nificantly different between groups. the cluster) depending on the iteration. The lowest All iterations of the analyses using dumbbells percentage match between iterations resulted from show the greatest increase in inertia gain in moving analyses using only categorical characters. Primo- from two to three clusters (Fig. 4a, c, e). In most iter- candelabrum clusters are defined by at most once ations, the next greatest increase in inertia gain is in character state with 100% inclusion or exclusion moving from three to four clusters (Fig. 4a, e), but of the individuals that display a particular character for branching characters only, the next greatest state (coded blue or green in Table 2b). The other increase was between six and seven clusters. character states defining the cluster also describe For most analyses where dumbbells were other clusters in the iteration – that is, individuals included, the clusters can be characterized as in multiple clusters share characters. Within Primo- follows: candelabrum, no iteration produces a set of clus- Cluster 1 (e.g. Figs 4b, d, f & Table 1) consists of ters within which all individuals are identical in Primocandelabrum sp. typified by: a short, nar- terms of their branching and distinct from other row stem, small disc and tall crown; first order clusters (without dividing the individuals into 15 branches that are rotated, furled, concealed and clusters). Importantly for this study, continuous and have proximal inflation; and second order categorical characters do not correlate perfectly branches that are rotated, unfurled, concealed, within Primocandelabrum as they do for dumb- radiating and have median inflation. bells – evident by the lower percentage of group Cluster 2 (e.g. Figs 4b, d, f & Table 1) consists of matches (Tables 3 & 4). It is not clear at this point Primocandelabrum sp. characterized by: a wide, whether this reflects the greater number of speci- short stem and a large, wide crown (Fig. 2a); first mens within Primocandelabrum or that the dumb- order branches that are unconcealed; and second bells innately show less variability. The outsize order branches that are concealed and have specimen BB plots as an outlier to the bulk popula- median inflation. tion in the principal components space, especially Cluster 3 (Figs 1a, 4b, d, f & Table 1) corresponds to for morphological characters only (Fig. 6e, f). the dumbbells (Wilby et al. 2011) and is charac- The continuous characters that significantly terized by: a long stem, large disc and small contribute to the construction of the dimensions are crown; first order branches are displayed, furled, similar in each iteration, but the categorical charac- concealed and with distal inflation; second order ters significantly contributing to the dimensions are branches are displayed, furled and have proximal more variable. For the reduced character matrix, the inflation. Table 1. Variables categorizing the clusters as determined by hierarchical clustering analysis. All analyses including dumbbells and only including specimens for 232 which both continuous and categorical characters could be determined. (a) Continuous: first values are the average for the cluster (number in bracket of row name); values in brackets are the average for all specimens Downloaded from http://sp.lyellcollection.org/ .G ECIGO .R WILBY R. P. & KENCHINGTON G. C. byguestonOctober1,2021

Table 1. (b) Categorical: ﬁrst number is percentage of individuals in cluster with speciﬁed character (not including imputed, i.e. originally missing, values) and the second number is percentage of individuals with that character included in the cluster http://sp.lyellcollection.org/ AGOOP LSIIAINSCHEMES CLASSIFICATION RANGEOMORPH byguestonOctober1,2021

Blue, .75% for all metrics of cluster assignment; green, ,75% for one metric, but either 100 or 0% of individuals with character state are included in cluster (i.e. total inclusion/exclusion); yellow, .75% for all bar one metric, but total inclusion/exclusion of individuals with a character state within/from cluster; orange, all metrics 50% , x . 75% (see online version for colour). br, categorical characters only; mo, continuous characters only analysed; nbb, outsize specimens removed; nd nbb, outsize specimens and feather dusters removed; red, reduced character matrix. 233 234 Table 2. Variables categorizing the clusters as determined by hierarchical clustering analysis. All analyses excluding dumbbells and only including specimens for which both continuous and categorical characters could be determined.

(a) Continuous: ﬁrst values are the average for the cluster (number in bracket of row name); values in brackets are the Downloaded from average for all specimens http://sp.lyellcollection.org/ .G ECIGO .R WILBY R. P. & KENCHINGTON G. C. byguestonOctober1,2021

Pink, average value for cluster larger than average value for all clusters; orange, average value for cluster larger than average value for all clusters, but smaller than the average value for another cluster; blue, average value for cluster smaller than average value for all clusters (see online version for colour). Table 2. (b) Categorical: ﬁrst number is percentage of individuals in cluster with speciﬁed character (not included imputed, i.e. originally missing, values) and the second number is percentage of individuals with that character included in the cluster Downloaded from http://sp.lyellcollection.org/ AGOOP LSIIAINSCHEMES CLASSIFICATION RANGEOMORPH byguestonOctober1,2021

br, categorical characters only; mo, continuous characters only analysed; nbb, outsize specimens removed; nd nbb, outsize specimens and feather dusters removed; red, reduced 235 character matrix. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

236 C. G. KENCHINGTON & P. R. WILBY

(a) Crown width (b) Crown width Crown height Crown height Length left Length left Length right Length right Disc width Disc width Disc height Disc height Stem height (c-i) Stem height (c-i) Stem height (c-bb) Stem height (c-bb) Stem width (base) Stem width (base) Stem width (at i) Stem width (at i) Stem width (top flare) Stem width (top flare) Primocandelabrum cluster 2

Primocandelabrum “dumbbells” “dumbbells” Primocandelabrum cluster 2

vectori cluster 1 vectori 0 102030405060 0 102030405060 “dusters” “dusters”

0 1020304050 0 1020304050 Index Index

(c) “dumbbells” (d) “dumbbells” Primocandelabrum cluster 1 Primocandelabrum cluster 1 Primocandelabrum cluster 2 Primocandelabrum cluster 2 Primocandelabrum cluster 3 Primocandelabrum cluster 3 “dusters” “dusters” crown width total height 10 20 30 40 0 102030405060

0 10203040 10 20 30 40 disc width total height

(e) “dumbbells” (f) “dumbbells” Primocandelabrum cluster 1 Primocandelabrum cluster 1 Primocandelabrum cluster 2 Primocandelabrum cluster 2 Primocandelabrum cluster 3 Primocandelabrum cluster 3 “dusters” “dusters” stem width stem length 1234 0 5 10 15 20

10 20 30 40 0 102030405060 total height crown width

Fig. 7. (a) Profile plot of all individuals ordered by height and broad groups; (b) profile plot with individuals ordered by group determined through cluster analyses and then by height. (c–f) Simple biplots of all individuals within the analysed dataset (those for whom both branching and morphological characters could be determined). Primocandelabrum cluster 1 (corresponding to P. aethelflaedia) are characterized by long stems, Primocandelabrum cluster 2 (corresponding to P. boyntoni) are characterized by short stems and unconcealed folia and Primocandelabrum cluster 3 (corresponding to P. aelfwynnia) is characterized by wide crowns and large discs. Downloaded from

Table 3. Comparing percentage of individuals placed in the same group (percentage group match) between pairs of iterations

Iteration all nbo br br mo mo po po po po mo po mo po po br po br po br red mo red mo po po po po mo po po mo nbo nbo nbb nbbnd nbb mo nbb nbbnd red nbo red red mo red red red red nbbnd nbo red nbbnd nbbnd nbb nbb http://sp.lyellcollection.org/

all 80.85 63.27 91.84 47.73 78.00 SCHEMES CLASSIFICATION RANGEOMORPH nbo 80.85 74.47 85.11 74.42 92.31 85.42 br 63.27 61.70 57.14 59.09 br nbo 74.47 61.70 59.57 95.35 58.97 mo 91.84 57.14 100.00 45.45 82.00 mo nbo 85.11 59.57 100.00 46.51 66.67 83.33 po 47.73 92.68 95.24 69.05 50.00 po nbb 74.42 92.68 87.80 75.61 61.90 po nbbnd 92.31 73.81 po mo 45.45 95.24 100.00 69.05 90.48 po mo nbb 46.51 87.80 100.00 68.29 90.48 po mo nbbnd 66.67 88.10 po br 59.09 69.05 69.05 97.56 po br nbb 95.35 75.61 68.29 97.56

po br nbbnd 58.97 byguestonOctober1,2021 red 78.00 93.18 mo red 82.00 43.18 red nbo 85.42 84.62 81.40 mo red nbo 83.33 51.28 46.51 po red 50.00 93.18 po mo red 90.48 43.18 po red nbbnd 73.81 88.10 84.62 po mo red 51.28 nbbnd po red nbb 61.90 81.40 po mo red 90.48 46.51 nbb

Only individuals for which continuous and categorical characters could be determined were included in each iteration. Analyses excluding dumbbells (po) were divided into two groups to allow comparison with those analyses including dumbbells (which typically separates into its own group, with the other two groups consisting of Primocandelabrum and feather dusters). br, categorical characters only; mo, continuous characters only analysed; nbb, outsize specimen BB excluded; nbbnd, BB and ‘feather dusters’ excluded; nbo, outsize specimens removed; po, ‘dumbbells’ excluded; red, reduced character matrix. 237 238 Downloaded from

Table 4. Comparing percentage of individuals placed in the same group (percentage group match) between pairs of iterations excluding dumbbells

Iteration po po nbb po po mo po mo po mo po br po br po br po red po mo po red po mo po red po mo http://sp.lyellcollection.org/ nbbnd nbb nbbnd nbb nbbnd red nbbnd red nbb red nbb

nbbnd WILBY R. P. & KENCHINGTON G. C. po 90.24 87.18 69.05 57.14 88.10 po nbb 90.24 97.44 85.37 56.10 90.24 po nbbnd 87.18 97.44 87.18 51.28 92.86 po mo 69.05 90.24 79.49 42.86 45.24 po mo nbb 85.37 90.24 89.74 46.34 53.66 po mo nbbnd 87.18 79.49 89.74 48.72 83.33 po br 57.14 42.86 100.00 54.76 57.14 po br nbb 56.10 46.34 100.00 84.62 84.62 59.52 po br nbbnd 51.28 48.72 84.62 84.62 48.72 43.59 byguestonOctober1,2021 po red 88.10 54.76 88.10 89.74 92.68 po mo red 45.24 57.14 88.10 82.05 100.00 po red nbbnd 92.86 48.72 89.74 82.05 92.31 87.18 po mo red nbbnd 83.33 87.18 po red nbb 59.52 92.68 92.31 87.18 80.95 po mo red nbb 90.24 53.66 43.59 100.00 87.18 80.95

Only individuals for which continuous and categorical characters could be determined were included in each iteration. All iterations were divided into three clusters, as this was supported by inertia gain and by cluster descriptions for most iterations. br, categorical characters only; mo, continuous characters only analysed;nbb, outsize specimen BB excluded; nbbnd, BB and feather dusters excluded; po, dumbbells excluded; red, reduced character matrix. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 239

Table 5. Summary of usefulness for categorical and continuous characters in deﬁning the multifoliate rangeomorph species used in this study. (a) Continuous characters

‘Average value signiﬁcantly different?’ refers to whether the averages of the taxon are statistically signiﬁcantly different ( p , 0.05) from the average values of the other taxa. The other columns refer to the presence and extent of overlap in the ranges of the continuous characters expressed by individuals assigned to each taxon. Green, no overlap; pale yellow, 0 , x . 20% overlap; orange, 20 , x . 50% overlap; red, .50% overlap (see online version for colour).

Table 5.(b) Categorical characters

The rankings for the numbers of clusters deﬁned and for how well clusters are deﬁned are in descending order, where 1 ¼ most useful character, and are based on the characters colour coded in Tables 1 and 2. Green, ranked 1; yellow, ranked 2 or 3; red, ranked bottom two; blue, character state unique to a taxon (see online version for colour).

For most analyses where dumbbells were excluded, for each variable); first order branches show three clusters can be characterized as follows: proximal inflation. Cluster 3 (Figs 5–7, Table 2) is characterized by: a Cluster 1 (Figs 5–7, Table 2) is typified by: a long large disc, a wide crown and a wide stem; uncon- and narrow stem, large disc and small crown; first cealed first and second order branches with either order branches concealed with median inflation; distal or proximal inflation; and furled second and second order branches with median inflation. order branches. Cluster 2 (Figs 5–7, Table 2) specimens have a large crown, slightly small disc and short, slightly nar- Discrimination of dumbbells. For all tests where row stem (all approximating the average values they are included, dumbbells plot away from the main Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

240 C. G. KENCHINGTON & P. R. WILBY group of Primocandelabrum specimens and com- and composition of clusters within Primocandela- prise the second group when specimens are forced brum is more variable. The variation in the defini- into two clusters (Fig. 4). Iterations including only tion of each cluster is summarized in Tables 1 and 2. branching characters place one Primocandelabrum Between iterations including and excluding specimen (6A1) into the cluster with dumbbells dumbbells, either including outsize specimens or (Fig. 4), whereas iterations using only continuous using only morphological characters, results in a characters place 4A1b (right), 13A4 and BB (when very low match in the cluster assignment of Primo- included) into the dumbbells group (Fig. 4e, f). candelabrum specimens (around 50%; Tables 3 & Two dumbbells (3D7 and 5A4) do not cluster with 4). When either all characters or only branching the others in the iterations including only branching characters are considered, and when outsize speci- characters, placing in cluster 1 when outsize specimens are excluded, the match increases to 74 and mens are included and cluster 2 when outsize speci- 81%, respectively (Tables 3 & 4). When individuals mens are excluded. This is not due to the quantity of are divided into three clusters, the outsize specimen missing data, as other individuals with the same BB is excluded and there is a 92% match between number of missing values consistently plot with iterations including and excluding feather dusters the rest of the dumbbell specimens, but is rather when all characters are considered, 82% between due to the fact that 5A4 has unfurled second order iterations including and excluding feather dusters branches and is the smallest specimen and 3D7 is when only continuous characters are considered missing the furled second order character, which is and 84% when only categorical characters are con- a significant variable in defining the clusters in sidered (Table 4). this analysis (Table 2). 5A4 and 3D7 also have tall When the Primocandelabrum specimens are crowns and short stems compared with the rest of divided into three or more clusters, disc size the dumbbells, which place them closer to the Primo- becomes a significant contributing factor to cluster candelabrum proportions for these characters. definition (Table 2a). When the data are split into two clusters, disc size is only significant if only con- Assignment of feather dusters. When dumbbells tinuous characters are used, or if the analyses are were included in the tests, and when only continuous run on the reduced character matrix (Table 2). characters are used, 8C1 clusters most frequently Three clusters are most strongly supported by inertia with dumbbells and 8C3 clusters most frequently gain for most analyses (Figs 5 & 6). In addition, with Primocandelabrum cluster 2. However, when when individuals are divided into three rather than categorical characters and outsize specimens are two clusters, a higher percentage of individuals included, both 8C1 and 8C3 place in Primocandela- that displayed a given character state are placed in brum cluster 2. When dumbbells were excluded, the the cluster described by that character state, and feather dusters place in very different parts of the each cluster contains a higher percentage of indi- factor map far from each other on the trees (Fig. 6). viduals displaying the character states by which that cluster is described (Table 2). When divided Removing outsize specimens. For the iterations into more than three clusters, there is a similar or including dumbbells, removing the outsize speci- greater percentage of cluster and character match, men BB only affected the number of clusters for but only one or two characters significantly describe one iteration, i.e. when only categorical charac- each cluster. Only when the specimens are divided ters were used. For this iteration, six clusters were into 15 clusters do the individual clusters have a supported when outsize specimens were included, suite of categorical characters that are unique to but five clusters were supported when outsize spec- that group. imens were excluded. There was a 90–100% (Table 3) match with the group assignment of specimens All characters v. continuous characters only v. cat- for all tests except those iterations including dumb- egorical characters only. There is a match between bells and including all characters (88%; Table 3) and all characters and categorical characters of only those using categorical characters only (62%; 55% when outsize specimens are included, 60% Table 3). The latter test also showed an appreciable when outsize specimens are excluded and 49% difference between characters significantly corre- when outsize specimens and feather dusters are lated to cluster 1 in particular, with more variables excluded. Comparing all characters and continuous significantly correlated to this cluster when outsize characters only, there is a match of 88% when out- specimens are included (Table 3). size specimens are included, 81% when outsize specimens are excluded and 87% when outsize Assignment of Primocandelabrum specimens. Pri- specimens and feather dusters are excluded. Com- mocandelabrum specimens + feather dusters are paring continuous and categorical characters only, consistently distinguished from dumbbells in almost there is a match of 57% when outsize specimens all iterations of the data, but determining the number are included, of 60% when outsize specimens are Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 241 excluded and of 42% when outsize specimens and (Laflamme et al. 2009; Liu et al. 2015). How- feather dusters are excluded (Table 4). ever, there are several conceivable advantages that might have been bestowed on a rangeomorph by Assessment of cluster assignment. When dumb- the possession of furled branches, particularly for bells are included, the within-cluster averages for higher order branches. First, it allows for a tighter the dumbbell group are significantly different to packing of branches and minimizing of the overlap the within-cluster averages for the two Primocande- of rangeomorph elements. Second, it may have labrum (+feather duster) clusters. There is also a served to protect the tips (a likely site of growth; greater discrimination of the dumbbells from Pri- Antcliffe & Brasier 2007) from abrasion by neigh- mocandelabrum individuals than for groups within bouring branches and from environmental damage. Primocandelabrum based on categorical characters, Third, the flow of water within a furl might be evidenced by more characters with 100% inclusion slowed, increasing its contact time with the rangeo- or exclusion (i.e. 0% inclusion) of individuals into morph surface. On a larger scale, the surface texture the dumbbell cluster than into either of the Primo- of a row of adjacent furled branches may help to candelabrum clusters (Tables 1 & 2). break up the boundary layer, resulting in a faster When individuals are sorted into the clusters to flow over the surface (as for the riblets on shark which they are assigned by the majority of itera- skin; Dean & Bhushan 2010). By contrast, unfurled tions, the total variance within clusters is easily rangeomorph elements would increase the surface visualized through profile plots (Fig. 7a, b). Vari- roughness of the rangeomorph exchange surface, ance within clusters determined through the hierar- reducing the thickness of the diffusive boundary chical cluster analysis is lower than it is for all layer over the surface and enhancing the uptake of individuals treated as a whole (Fig. 3, evidenced nutrients and/or the removal of waste products. by smoother trend lines in Fig. 7). When bivariate Therefore if these characters are functionally im- character plots are constructed with individuals col- portant, then it might be expected that a trade-off oured according to their cluster number (Fig. 7c–f), between increased surface roughness and damage the clusters follow distinct trend lines for those char- prevention would determine whether a rangeo- acters shown to provide the strongest discrimination morph adopted a furled or unfurled branching of each cluster (Tables 1 & 2), but do not form clear architecture. trend lines for those characters that provide weak The functional advantage of certain aspects of discrimination of the clusters. These plots therefore branching, such as proximal v. median v. distal support the conclusions drawn from the results of inflation (Brasier et al. 2012), is not readily appar- the cluster analyses. ent. However, maintaining a consistent mode of inflation across different levels of branching may have served to reduce self-shading (cf. Enrı´quez & Discussion Pantoja-Reyes 2005) and could therefore have been favoured in slower flow or lower nutrient environ- The argument that generic level diagnoses should be ments. Other aspects of branching, such as rotated based on categorical characters, such as branching v. displayed branches and concealed v. unconcealed architecture, and that continuous characters should axes, may be implicated in the efficiency of branch be used for species level diagnoses rests on the infer- packing and consequently have other functional con- ence that categorical characters dominantly reflect a straints. Traits that provide an advantage through genetic control, whereas continuous characters, their functionality are more likely to be converged such as stem length, are more susceptible to envi- upon and therefore have the potential to skew phy- ronmental influences and, potentially, to ecopheno- logenies where taxa lack a large number of apomor- typic variability and convergence (Liu et al. 2016). phies. Given the current uncertainties regarding the If aspects of branching architecture can be influ- functional constraints on branching architecture, enced or modified by the environment to the same and the consequent possibilities of convergence, it extent as continuous characters, then this inference is perhaps premature to place greater phylogenetic would break down. relevance on some categorical characters over others, and over continuous characters. Potential functional significance of rangeomorph branching Influence of taphonomy The furling of branches is observed in many rangeo- Any analysis of fossil morphology must take into morph taxa (Brasier et al. 2012) and is seemingly account the biases that may be imposed by tapho- at odds with the presumed function of rangeo- nomic and post-fossilization processes (Matthews morph elements as exchange surfaces, as it decreases et al. 2017). Taphonomic interference may include the surface area exposed to the water column the local rotation of branches (Narbonne et al. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

242 C. G. KENCHINGTON & P. R. WILBY

2009), the deflation of individual branches (Brasier clusters. This is strong support for these individuals et al. 2013), as well as the displacement of branch representing a distinct and separate group and indi- order of overlap for unconstrained branches. cates that the statistical approach described here The fact that only branching characters consistent is useful in discriminating taxonomic groups in across a given specimen were included in the anal- mixed populations. It also suggests that both cate- yses presented here should minimize the influence gorical and continuous characters are, at some of taphonomy on the branching characters in level, taxonomically useful. this study. The overall morphology of a rangeomorph (i.e. Categorical or continuous characters, or both? If continuous characters) is potentially susceptible the clusters identified by these analyses represent to several sources of taphonomic overprint. For discrete taxa, which iteration should be used for example, the preserved size of the holdfast and the taxonomic discrimination: those including all char- number of concentric rings may be influenced by acters, only categorical characters, or only continu- burial depth in the sediment. Stem width may be ous characters? If categorical characters do indeed obscured by sediment settling beneath the stem dur- represent generic level diagnoses, then the division ing felling (cf. Laflamme et al. 2007). The shape of of clusters should first be based on these characters the crowns of Primocandelabrum and the dumb- and only subsequently subdivided into species by bells are inferred to have been three-dimensional the variation in continuous characters. However, in life, but are preserved as compound, two- unless taxa with the same branching characters dimensional impressions. The preserved morphol- have converged on forms that are indistinguishable ogy was also likely to have been influenced during based on the sum of their continuous characters, felling, depending on the stiffness and rheology then the combination of continuous and categorical of the branches. The preservation of only two characters should provide the best discrimination of main branches in the Primocandelabrum specimens meaningful taxonomic groupings. from Newfoundland is potentially due to the failure For continuous characters, the disc width, stem of a third branch to be expressed in the fossil, per- width, stem length, crown width and crown length haps held above the plane of preservation. By all served to distinguish pairs of clusters from each that token, it is possible that Primocandelabrum other, although no single character discriminated specimens from elsewhere, and indeed multifoliate each of the three clusters from each other and taxa in general, may have had more branches than from the average values of the group, both when are preserved. dumbbells were and were not included (Tables 1 & 2). This means that the combination of characters Validity of the statistical approach is required to identify each of the clusters identified within the population. For categorical characters, The group of individuals assigned to dumbbells can the nature of the first order branches (folia) proved readily be distinguished from Primocandelabrum most successful in discriminating dumbbells from individuals using a number of traditional taxonomic Primocandelabrum, especially displayed v. rotated, methods and so provide a good means of validating the sense of inflation and furled v. unfurled (Table the approach used here. The dumbbell cluster is 1). For second order branching characters, only consistently discriminated from Primocandelabrum furled v. unfurled and the sense of inflation were based on categorical and continuous characters in consistently significant in the division of the clus- isolation and in tandem. That the dumbbell cluster ters. Within Primocandelabrum, the only relevant is discriminated for all of these iterations indicates first order characters were the sense of inflation that categorical character states correlate consis- and concealed or unconcealed axes, as all individu- tently with particular values for the continuous char- als had rotated and unfurled branches (where this acters, at least at this taxonomic level. Crucially, for character could be determined). Median inflation the categorical characters that statistically signifi- discriminated single clusters with 100% inclusion cantly describe the dumbbell cluster, a high propor- or exclusion, and the absence of either concealed tion of the individuals within that cluster display or unconcealed axes was also useful in discriminat- a given character state and a high percentage of ing pairs of clusters from each other. For the second- the individuals that display that character state are order branch characters, only the sense of inflation assigned to the cluster (Table 3). In addition, the consistently discriminated pairs of clusters across mean values of the continuous characters for dumb- all iterations, although furling and concealed/ bells are statistically significantly different from the unconcealed characters were useful for some itera- means of the continuous characters describing the tions (Table 2). Primocandelabrum clusters and are more different In the analyses of Primocandelabrum, the clus- from the means of the Primocandelabrum clusters ter assignment for iterations using only continu- than are the means within the Primocandelabrum ous characters most consistently agrees with the Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 243 assignment from iterations using all characters, hav- characters place one (8C3, Fig. 1j) within Primo- ing a higher cluster assignment match than for cate- candelabrum clusters and the other (8C1, Fig. 1k) gorical characters only with all characters. Although most frequently with dumbbells. There are three variation in one continuous character (such as stem potential causes for the failure of feather dusters to length) could reflect ecological drivers, variation in fall consistently into the same cluster and into a all continuous characters as documented here is cluster separate from the Primocandelabrum speci- much harder to explain in this manner, especially mens: (1) the limited size of the dataset; (2) they rep- as all specimens are found on the same bedding resent two different taxa; and (3) the exclusion of a plane. Equally, the potential functional drivers for key taxonomic character (the number of folia) branching architecture as well as for continuous from the analyses due to its uncertainty in the vast characters means that one set of characters is not majority of specimens of both Primocandelabrum inherently more robust than the others. Accordingly, and feather dusters – as with dumbbells, it is only the clusters defined by all characters are interpreted possible to determine a minimum number of here to most likely represent the fundamental, bio- folia. The inclusion of the number of folia may logical clusters within the data. The greater mis- change the placement of feather dusters relative to match in cluster assignment between iterations Primocandelabrum and would further distinguish when only categorical characters are used – com- dumbbells from Primocandelabrum. Based on mor- pared with iterations using both categorical and phological characters, 8C1 is more similar to dumb- continuous characters, or only continuous charac- bells than it is to 8C3 or Primocandelabrum and its ters – has two likely causes: (1) the clusters defined branching structure (as far as can be determined) do not incorporate all of the individuals that share a only differs from that of dumbbells in having character state; and (2) the particular set of categor- median v. distally inflated first order branches, ical characters that are used to describe the clusters median v. proximally inflated second order branches varies widely between the different clusters in a and possibly also in the rotated/displayed nature of given iteration and between iterations. By contrast, its folia. the same continuous characters are used to define Even within the dumbbells – which these analy- the clusters, both between clusters in a given itera- ses demonstrate to be a discrete taxon – there is var- tion and between iterations (Tables 1 & 2). This iation in both continuous and categorical characters. could be taken to infer that categorical charac- Accordingly, although the stem of specimen 8C1 is ters are less taxonomically robust than continuous proportionally thicker and its holdfast smaller than characters. other specimens of dumbbells, even of similar Inclusion of outgroups is an important consider- size, its dominant placement in the dumbbell cluster ation. Although the inclusion of outgroups can suggests that it is a different morph of dumbbells, help to root clusters by providing information although whether it is a different species or simply on ancestral or shared characters, they can also a variant cannot be determined on the evidence mask variation within the group under scrutiny, par- from this solitary individual. All Primocandela- ticularly if there is character convergence. When brum specimens appear to have three rotated folia, dumbbells are included, the characteristics of indi- but because the number of folia and their rotated/ viduals belonging to dumbbells contribute to the displayed nature in 8C3 is uncertain, it is not clear definitions of the principal component space and whether this individual represents a separate taxon of all the clusters, both masking variation within or a different taphonomic expression of Primocan- Primocandelabrum and influencing the way that delabrum. The importance of folia number and group is divided (Tables 1–4). When dumbbells branching characters on the placement of the two are excluded, only the characters of Primocandela- feather dusters in particular indicates that these are brum contribute to the construction of the compo- key taxonomic characters. However, the consistent nent space and, consequently, these iterations placement of 8C3 within the Primocandelabrum should most reliably and accurately identify natural spectrum based on those characters that are avail- clusters within Primocandelabrum. The inclusion/ able, and its close position to Primocandelabrum exclusion of the two feather dusters has little effect in the principal component space even when its on the analyses, but does influence the clustering folia are set as ‘displayed’, does not preclude its to some degree (Tables 1–4), particularly when assignment to this genus. 8C3 is included because this individual is assigned to dumbbells. Taxonomic subdivision of Primocandelabrum Placement of feather dusters Determining the number and rank of taxa within the Primocandelabrum specimens is challenging given Separation of the two feather dusters is more prob- the variability of cluster composition depending lematic, as their morphological and branching on data treatment and the fact that many characters Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

244 C. G. KENCHINGTON & P. R. WILBY that define clusters do not do so exclusively (Table Primocandelabrum based on almost all categorical 2). Following the guidelines of Brasier et al. characters. Although the mean values of many con- (2012), where the branching characters of each tinuous characters for each cluster within Primocan- genus are unique, would result in 15 genera within delabrum are statistically significant from the the 40 well-preserved Primocandelabrum speci- means of the characters for Primocandelabrum as mens analysed here. There is also variation in the a whole (Table 2), the ranges in values for each continuous characters within each of these 15 group overlap (Table 5). In addition, morphological groups, and this is greater than the variation between and branching characters do not correlate as closely the groups. Although it is possible that the speci- as they do for dumbbells. An individual Primocan- mens here ascribed to Primocandelabrum represent delabrum could be assigned to a taxon based on a supra-generic group, the presence of so many mor- either its continuous or its categorical characters, phologically similar genera (let alone species) is but the resulting assignments are only likely to be considered to be unlikely and difficult to objec- the same around two-thirds of the time (Figs 4 & tively apply to other, less well-preserved material. 5; Tables 3 & 4). This morphological variation Accordingly, this interpretation is rejected here. does not appear to be the result of ontogenetic vari- The clustering analyses consistently support two ation because the size ranges of specimens ascrib- or three clusters within Primocandelabrum, with the able to each form overlap. best discrimination of clusters being achieved when Taken together, this suggests that the taxonomic the specimens are divided into three clusters. Ana- subdivision within Primocandelabrum is at a lower lyses based on only the branching characters show rank than the division between dumbbells and Pri- additional increases in inertia gain at higher levels mocandelabrum. The shared character states and of cluster division (Figs 5a, c & 6c), supporting overlap in morphology is inconsistent with division more than three clusters, but even then the best clus- at a generic level. It could be used as a basis to treat ter discrimination is generally achieved when the the groups as morphs of one species. However, the group is divided into three rather than more than three natural groupings are consistently identified three clusters. When Primocandelabrum is divided and can be distinguished on the basis of the sum into the three clusters determined by the majority of their characters. Accordingly, we consider divi- of cluster analyses, profile plots of the data show a sion of the Charnwood Forest Primocandelabrum smoother, more consistent correlation of morpho- specimens into three species (Figs 8–10) is justified. logical traits with total size than when the individu- They are designated as P. aethelflaedia (correspond- als are unsorted (Fig. 7a, b). Stem width shows a ing to cluster 1 of Figs 5–7), P. boyntoni (corre- smaller increase in proportion to the total height sponding to cluster 2 of Figs 5–7) and P. than other characters (Fig. 7b). Division into three aelfwynnia (corresponding to cluster 3 of Figs 5–7). groups is also borne out by simple scatterplots (Fig. 7c, d), where the groups assigned by the principal component analyses (reduced and full character Conclusions matrix) follow separate trends for many pairs of measurements. An underlying ontogenetic cause Statistical techniques such as those described here for these differences can be ruled out by the fact provide a powerful way to analyse large datasets that the large specimens do not cluster together or with missing values and, importantly, are a means plot close to each other (Figs 4 & 7c–f). of analysing both categorical and continuous char- There are therefore three dominant clusters that acters in tandem, while being free from any assump- can be described within Primocandelabrum. The tion as to the taxonomic weight of any of the question is then to what taxonomic level these clus- characters. They also provide a way of identifying ters should be assigned: genus, species, sub-species and quantifying variation within a taxon. There are or variety? The clear discrimination of dumbbells three possibilities for interpreting the variation from Primocandelabrum is not achieved for clus- within the group of individuals referable to Primo- ters within Primocandelabrum. Few clusters are candelabrum. First, that it is a supra-generic group defined by a 100% assignment for individuals that containing upwards of 15 genera, where each genus share the same categorical character state, or with has a distinct branching architecture; second, that it 100% of individuals within a cluster sharing all is a single species within which there is a consider- the same character states. Only two or three cate- able degree of variation in both categorical and con- gorical characters statistically significantly contrib- tinuous characters; and third, that it is a single genus ute to the definition of each cluster, with the other containing three species, within each of which there categorical characters showing no significant corre- is variation in the continuous and categorical char- lation with one cluster over another, even when acters. That three natural clusters can be identified only categorical characters are used. This is in con- within the group leads us to favour the latter inter- trast with the discrimination of dumbbells from pretation. We further interpret the dumbbells as a Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 245

Fig. 8. Primocandelabrum boyntoni, sp. nov. from Charnwood Forest. (a) Plastotype, 6A2l (GSM105969); (b) interpretative drawing of (a); (c) paratype 10A2 (GSM 106046); (d) interpretative drawing of (c); (e) paratype10B8 (GSM 106049); (f) interpretative drawing of (e). Scale bars 2 cm. Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

246 C. G. KENCHINGTON & P. R. WILBY

Fig. 9. Primocandelabrum aethelflaedia, sp. nov. from Charnwood Forest. (a) Plastotype 4A1b (GSM105953); (b) interpretative drawing of (a); (c) paratype 19C4 (GSM106049); (d) interpretative drawing of (c). Scale bars 2 cm. distinct genus from Primocandelabrum based on its the sum of the characters within the whole popula- consistent separation from that group based on all tion. The findings that the combination of continu- continuous and categorical characters. ous and categorical characters is more powerful These techniques may prove especially useful than either set considered in isolation, and that in elucidating the taxonomy of the feather dusters certain characters are more useful taxonomic dis- (Mason & Narbonne 2016). They allow taxa to criminants (Table 5), has important ramifications be discriminated in an unbiased way based on for rangeomorph taxonomy. The application of Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 247

Fig. 10. Primocandelabrum aelfwynnia sp. nov. from Charnwood Forest. (a) Plastotype, 3D8 (GSM105963); (b) interpretative drawing of (a); (c) paratype 19B1 (GSM106040); (d) interpretative drawing of (c). Scale bars 2 cm. these techniques to other taxa would, in an unbiased specimens, with the only variation observed clearly way, confirm or question some recent taxonomic attributable to taphonomic processes (Wilby et al. revisions within rangeomorphs. These include syno- 2015). Charnwood Forest has yielded few other cur- nymizations (Liu et al. 2016) and the creation of rently assignable unifoliate rangeomorphs and so new genera based on single differences in branching their variety in branching pattern is impossible to character states alone (e.g. the differentiation of Vin- constrain. The only other multifoliate taxon known landia and Trepassia from Charnia). from more than a handful of individuals is Bradga- The variety of branching characters in specimens tia, but its branching pattern is not distinguishable whose gross morphology is otherwise essentially across large parts of many of these specimens due indistinguishable and, likewise, the shared branch- to their complex three-dimensional shape and their ing characteristics of individuals with at least a preservation as compound fossils. Analysis of taxa superficially distinguishable morphology is surpris- such as Trepassia, Vinlandia and Beothukis using ing. This is especially so given the fact that few these techniques would reveal the extent of variation other published descriptions acknowledge variabil- in branching characters in unifoliate taxa and would ity in branching pattern within a taxon, unless attrib- test whether intra-specific variability in branching uted to processes such as ontogeny (Brasier et al. pattern is a feature unique to multifoliate taxa. The 2012). Certainly, Charnia masoni seems to have a techniques presented here have the potential to very consistent branching pattern across multiple revise the taxonomy of not just rangeomorphs, but Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

248 C. G. KENCHINGTON & P. R. WILBY also a host of other Ediacaran groups, including This species is characterized by a proportionally small arboreomorphs, dickinsoniomorphs, erniettomorphs disc (c. 25% (10–30%) of the total height), a short stem and even the palaeopascichnids. Problematic Phan- (average 25% (10–30%) of the total height), and a rela- erozoic groups would be similarly amenable to ana- tively wide crown (approximately equal to the total height, lysis using this approach. but 75–130%). The proportions of this taxon are close to the average for the genus Primocandelabrum (as defined The authors thank D. McIlroy for inviting this contribution from Charnwood Forest). Its first order branches have to the volume and for his considerable help in preparing it proximal inflation and are typically unconcealed and its for publication. We also thank A.G. Liu and M. Laflamme for their helpful and insightful comments on this paper, second order branches are typically displayed, concealed and N.J. Butterfield and E.G. Mitchell for their discussions and show proximal inflation, but may be unconcealed on this work. We thank our colleagues Sue Martin, Louise and either subparallel or radiating. Its third order branches Neep, Scott Renshaw and Paul Shepherd for their help in are concealed and are typically displayed and unfurled, but casting and moulding the specimens. Thanks also to may be subparallel or radiating, and with median or S. Harris and to the British Geological Survey for providing distal inflation. photographs of the fossil specimens. CGK and PRW were funded by National Environment Research Council grant P. aethelflaedia sp. nov. NE/1005927/1. CGK also acknowledges a Research Stu- Etymology. Named after Lady Aethelflaed, an ancient dentship funded by the Cambridge Philosophical Society. queen of the historical kingdom of Mercia (within which Access to sites was kindly facilitated by Natural England Charnwood Forest is located). and landowners in Charnwood Forest. Material. This species is described from 12 complete spec- Appendix A: Systematic descriptions imens, all from Bed B in Charnwood Forest (Wilby et al. 2011). Master moulds and casts are housed at the British Group Rangeomorpha Pflug, 1972 Geological Survey, Keyworth. Genus Primocandelabrum Hofmann, O’Brien & King, 2008 Plastotype is designated as 4A1b (GSM105953; Fig. Type species P. hiemaloranum Hofmann, 9a, b), paratype is designated as 19C4 (GSM106049; Fig. O’Brien & King, 2008 9c, d). This species has a proportionally long stem (average 35% (range 30–40%) of the total height) and a crown that Emended diagnosis for the genus, Primocandelabrum. is both short (average 65% (range 50–75%) of the total Multifoliate rangeomorph consisting of a stem, holdfast height) and narrow (average 80% (range 60–90%) of the and crown. The crown is wider than it is tall, comprising total height). It is characterized by folia and second order around 70% (60–80%) of the total height of the organism branches that are concealed. First order branches show taken from the base of the stem. Its width may be up to 1.5 dominantly proximal (but sometimes median) inflation. times the total height of the organism. The stem comprises Second order branches show dominantly median (but the remaining 30% (20–40%) of the total height and its sometimes proximal) inflation, may be displayed or rotated width is around 10% (7–15%) of the total height. The and subparallel or radiating. Third order branches are con- width of the holdfast is around 30% (20–40%) of the cealed, but may be rotated or displayed, furled or unfurled, height of the organism. The crown consists of three first subparallel or radiating and with no or distal inflation. order branches (folia), which emanate from one point at the distal end of the stem. Folia are rotated, unfurled and P. aelfwynnia sp. nov. typically show proximal inflation. Second order (primary) Etymology. Named after Lady Aelfwynn, the last queen of branches are typically unfurled, displayed and show the historical kingdom of Mercia (within which Charn- median inflation. Both orders may be unconcealed or con- wood Forest is located). cealed at their bases, typically concealed towards the distal tips. Third order (secondary) branches are typically dis- Material. This species is described from 10 complete played, concealed and unfurled, with either median or specimens, all from Bed B in Charnwood Forest (Wilby distal inflation. et al. 2011). Master moulds and casts are housed at the P. boyntoni sp. nov. British Geological Survey, Keyworth. Plastotype is designated as 3D8 (GSM105963; Fig. 10a, b); paratypes are Etymology. Named for Helen Boynton, in recognition of 19B1 (GSM106040; Fig. 10c, d) and BB (GSM105872; her work on the biotas of Charnwood Forest. Fig. 1b). This species is characterized by a crown that is wider Material. This species is described from 18 complete spec- than the organism is tall (average 140% (range 110– imens, all from Bed B in Charnwood Forest (Wilby et al. 175%) of the total height) and a proportionally large disc 2011). Master moulds and casts are housed at the British (average 50% (range 30–90%) of the total height). Its Geological Survey, Keyworth. first order branches are unconcealed and inflate proximally Plastotype is designated as 6A2l (GSM105969; Fig. and its second order branches are displayed, concealed or 8a, b), paratypes are 10A2 (GSM106046; Fig. 8c, d) and unconcealed and may be furled or unfurled and may 10B8 (GSM106049; Fig. 8e, f). show distal inflation. Its third order branches are Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

RANGEOMORPH CLASSIFICATION SCHEMES 249 concealed, and may be displayed or rotated, furled or Erwin, D.H., Laflamme,M.,Tweedt, S.M., Sperling, unfurled, subparallel or radiating and with no, median or E.A., Pisani,D.&Peterson, K.J. 2011. The Cambrian distal inflation. conundrum: early divergence and later ecological success in the early history of animals. Science, 334, 1091–1097, https://doi.org/10.1126/science.1206375 References Hofmann, H.J., O’Brien, S.J. & King, A.F. 2008. Edia- caran biota on Bonavista Peninsula, Newfoundland, Antcliffe, J.B. & Brasier, M.D. 2007. Charnia and sea Canada. Journal of Paleontology, 82, 1–36, https:// pens are poles apart. Journal of the Geological Society, doi.org/10.1666/06-087.1 London, 164, 49–51, https://doi.org/10.1144/0016- Hoyal Cuthill,J.F.&Conway Morris, S. 2014. Fractal 76492006-080 branching organizations of Ediacaran rangeomorph Bonder, M.J., Abeln, S., Zaura,E.&Brandt, B.W. fronds reveal a lost Proterozoic body plan. Proceedings 2012. Comparing clustering and pre-processing in of the National Academy of Sciences, 111, 13122– taxonomy analysis. Bioinformatics, 28, 2891–2897, 13126, https://doi.org/10.1073/pnas.1408542111 https://doi.org/10.1093/bioinformatics/bts552 Husson, F., Leˆ,S.&Page`s, J. 2010. Exploratory Multi- Brasier, M.D. & Antcliffe, J.B. 2009. Evolutionary variate Analysis by Example Using R. CRC Press/ relationships within the Avalonian Ediacara biota: Chapman & Hall/CRC Computer Science & Data new insights from laser analysis. Journal of the Geo- Analysis. logical Society, London, 166, 363–384, https://doi. Jardine,N.&Sibson, R. 1971. Mathematical Taxonomy. org/10.1144/0016-76492008-011 Wiley, San Francisco, CA. Brasier, M.D., Antcliffe, J.B. & Liu, A.G. 2012. Josse,J.&Husson, F. 2012. Selecting the number of com- The architecture of Ediacaran fronds. Palaeontology, ponents in principal component analysis using cross- 55, 1105–1124, https://doi.org/10.1111/j.1475-4983. validation approximations. Computational Statistics 2012.01164.x & Data Analysis, 56, 1869–1879, https://doi.org/10. Brasier, M.D., Liu, A.G., Menon, L., Matthews, J.J., 1016/j.csda.2011.11.012 McIlroy,D.&Wacey, D. 2013. Explaining the Josse, J., Chavent, M., Liquet,B.&Husson, F. 2012. exceptional preservation of Ediacaran rangeomorphs Handling missing values with regularized itera- from Spaniard’s Bay, Newfoundland: a hydraulic tive multiple correspondence analysis. Journal of model. Precambrian Research, 231,122–135,https:// Classification, 29, 91–116, https://doi.org/10.1007/ doi.org/10.1016/j.precamres.2013.03.013 s00357-012-9097-0 Clapham, M.E. & Narbonne, G.M. 2002. Ediacaran epi- Laflamme,M.&Narbonne, G.M. 2008. Ediacaran faunal tiering. Geology, 30, 627–630, https://doi.org/ fronds. Palaeogeography, Palaeoclimatology, Palaeo- 10.1130/0091-7613(2002)030,0627:EET.2.0.CO;2 ecology, 258, 162–179, https://doi.org/10.1016/j. Coghlan, A. 2014. A Little Book of R for Time Series. palaeo.2007.05.020 Wellcome Trust Sanger Institute, Cambridge. Laflamme, M., Narbonne, G.M. & Anderson, M.M. Darroch, S.A.F., Laflamme,M.&Clapham, M.E. 2004. Morphometric analysis of the Ediacaran frond 2013. Population structure of the oldest known macro- Charniodiscus from the Mistaken Point Formation, scopic communities from Mistaken Point, Newfound- Newfoundland. Journal of Paleontology, 78, 827–837. land. Paleobiology, 39, 591–608, https://doi.org/10. Laflamme, M., Narbonne, G.M., Greentree,C.& 1666/12051 Anderson, M.M. 2007. Morphology and taphonomy Dean,B.&Bhushan, B. 2010. Shark-skin surfaces for of an Ediacaran frond: Charnia from the Avalon Pen- fluid-drag reduction in turbulent flow: a review. Philo- insula of Newfoundland. In: Vickers-Rich,P.& sophical Transactions of the Royal Society of London Komarower, P. (eds) The Rise and Fall of the Edia- A: Mathematical, Physical and Engineering Sciences, caran Biota. Geological Society, London, Special Pub- 368, 4775–4806, https://doi.org/10.1098/rsta.2010. lications, 286, 237–257, https://doi.org/10.1144/ 0201 SP286.17 Dillon, W.R. & Goldstein, M. 1984. Multivariate Ana- Laflamme, M., Xiao,S.&Kowalewski, M. 2009. lysis: Methods and Applications. Wiley, New York. Osmotrophy in modular Ediacara organisms. Proceed- Dufour, S.C. & McIlroy, D. 2016. Ediacaran pre- ings of the National Academy of Sciences of the United placozoan diploblasts in the Avalonian biota: the role States of America, 106, 14 438–14 443, https://doi. of chemosynthesis in the evolution of early animal org/10.1073/pnas.0904836106 life. In: Brasier, A.T., McIlroy,D.&McLoughlin, Laflamme, M., Flude, L.I. & Narbonne, G.M. 2012. N. (eds) Earth System Evolution and Early Life: A Cel- Ecological tiering and the evolution of a stem: the old- ebration of the Work of Martin Brasier. Geological est stemmed frond from the Ediacaran of Newfound- Society, London, Special Publications, 448. First pub- land, Canada. Journal of Paleontology, 86, 193–200, lished online September 15, 2016, https://doi.org/10. https://doi.org/10.1666/11-044.1 1144/SP448.5 Lajus, D., Sukhikh,N.&Alekseev, V. 2015. Cryptic or Dunn,G.&Everitt, B.S. 2004. An Introduction to Math- pseudocryptic: can morphological methods inform ematical Taxonomy. Courier, Massachusetts. copepod taxonomy? An analysis of publications and Enrı´quez,S.&Pantoja-Reyes, N.I. 2005. Form- a case study of the Eurytemora affinis species complex. function analysis of the effect of canopy morphology Ecology and Evolution, 5, 2374–2385, https://doi. on leaf self-shading in the seagrass Thalassia testudi- org/10.1002/ece3.1521 num. Oecologia, 145, 234–242, https://doi.org/10. Liu, A.G., Kenchington, C.G. & Mitchell, E.G. 2015. 1007/s00442-005-0111-7 Remarkable insights into the paleoecology of the Downloaded from http://sp.lyellcollection.org/ by guest on October 1, 2021

250 C. G. KENCHINGTON & P. R. WILBY

Avalonian Ediacaran macrobiota. Gondwana Research, rangeomorphs from Spaniard’s Bay, Newfoundland. 27, 1355–1380, https://doi.org/10.1016/j.gr.2014.11. Journal of Paleontology, 83, 503–523, http://www. 002 jstor.org/stable/29739123 Liu, A.G., Matthews, J.J. & McIlroy, D. 2016. The Narbonne, G.M., Laflamme, M., Trusler, P.W., Dal- Beothukis/Culmofrons problem and its bearing on rymple, R.W. & Greentree, C. 2014. Deep-water Ediacaran macrofossil taxonomy: evidence from an Ediacaran fossils from Northwestern Canada: taphon- exceptional new fossil locality. Palaeontology, 59, omy, ecology, and evolution. Journal of Paleontology, 45–58, https://doi.org/10.1111/pala.12206 88, 207–223, https://doi.org/10.1666/13-053 Mason, S.J. & Narbonne, G.M. 2016. Two new Edia- Pagni, M., Niculita-Hirzel,H.et al. 2013. Density- caran small fronds from Mistaken Point, Newfound- based hierarchical clustering of pyro-sequences on land. Journal of Paleontology, 90, 183–194, https:// a large scale – the case of fungal ITS1. Bioinformat- doi.org/10.1017/jpa.2016.14 ics, 29, 1268–1274, https://doi.org/10.1093/bioinfor Matthews, J.J., Liu, A.G. & McIlroy, D. 2017. Post- matics/btt149 fossilization processes and their implications for Pflug, H.D. 1972. Systematik der jung-pra¨kambrischen understanding Ediacaran macrofossil assemblages. Petalonamae. Palaöntologische Zeitschrift, 46, 56–67. In: Brasier, A.T., McIlroy,D.&McLoughlin,N. RCORE TEAM. 2014. R: A Language and Environment for (eds) Earth System Evolution and Early Life: A Cele- Statistical Computing. R Foundation for Statistical bration of the Work of Martin Brasier. Geological Computing, Vienna. Society, London, Special Publications, 448. First pub- Wilby, P.R., Carney, J.N. & Howe, M.P.A. 2011. A rich lished online March 17, 2017, https://doi.org/10. Ediacaran assemblage from eastern Avalonia: evi- 1144/SP448.20 dence of early widespread diversity in the deep ocean. Mitchell, E.G., Kenchington, C.G., Liu, A.G., Geology, 39, 655–658, https://doi.org/10.1130/ Matthews, J.J. & Butterfield, N.J. 2015. Recon- G31890.1 structing the reproductive mode of an Ediacaran Wilby, P.R., Kenchington, C.G. & Wilby, R.L. 2015. macro-organism. Nature, 524, 343–346, https://doi. Role of low intensity environmental disturbance in org/10.1038/nature14646 structuring the earliest (Ediacaran) macrobenthic Narbonne, G.M. 2004. Modular construction of early Edi- tiered communities. Palaeogeography, Palaeoclima- acaran complex life forms. Science, 305, 1141–1144. tology, Palaeoecology, 434, 14–27, https://doi.org/ Narbonne, G.M. 2005. The Ediacara biota: Neoprotero- 10.1016/j.palaeo.2015.03.033 zoic origin of animals and their ecosystems. Annual Wood, D.A., Dalrymple, R.W., Narbonne, G.M., Geh- Review of Earth and Planetary Sciences, 33, ling, J.G. & Clapham, M.E. 2003. Paleoenviron- 421–442, https://doi.org/10.1146/annurev.earth.33. mental analysis of the late Neoproterozoic Mistaken 092203.122519 Point and Trepassey formations, southeastern New- Narbonne, G.M., Laflamme,M.,Greentree,C.&Trus- foundland. Canadian Journal of Earth Sciences, 40, ler, P. 2009. Reconstructing a lost world: Ediacaran 1375–1391, https://doi.org/10.1139/e03-048