Eriostemon

Orel, Harvey K., McLay, Todd G. B., Neal, Will C., Forster, Paul I. & Bayly, Michael J., 2023, Plastid phylogenomics of the Eriostemon group (Rutaceae; Zanthoxyloideae): support for major clades and investigation of a backbone polytomy, Australian Systematic Botany 36 (5), pp. 355-385 : 371-373

Investigation of the Eriostemon group polytomy

Analysis of tree space

Our approach identified the optimal number of gene tree clusters in our data as three. Cluster 1 contained 72 loci (33 protein-coding, 39 non-coding) with a mean length of 955 bp, and a mean number of parsimony-informative sites of 141. Cluster 2 contained 50 loci (21 protein-coding, 29 non-coding) with a mean length of 276 bp, and a mean number of parsimony-informative sites of 44. Cluster 3 contained 26 loci (17 protein-coding, 9 non-coding) with a mean length of 378 bp, and a mean number of parsimony-informative sites of 66. Re-estimating the phylogeny for each of the three clusters produced trees with no supported incongruence between, for both IQ-TREE (Supplementary Fig. S3) and ASTRAL phylogenies (see files under Data availability). RAxML gene trees from Clusters 1 and 2 were significantly longer than those in Cluster 3 (Dunn’s test; Bonferroni adjusted P = 8.99 × 10 −6 and P = 5.52 × 10 −4 respectively; Supplementary Fig. S4). Bootstrap support was significantly different among all clusters, with support values in Cluster 1 being markedly higher than those in Clusters 2 and 3 (Dunn’s test; Bonferroni adjusted P = 2.97 × 10 −13 for C 1– C 2, P = 2.31 × 10 −21 for C 1– C 3, P = 1.87 × 10 −3 for C 2– C 3; Fig. S4).

Likelihood-mapping analysis

Across all loci, the mean percentage of fully resolved quartets (i.e. quartets falling into Regions 1, 2 and 3 of the likelihood-mapping plot) was 53%. The means for partly resolved quartets (Regions 4, 5 and 6) and unresolved quartets (Region 7) were 4 and 43% respectively. The subset of loci with>50% fully resolved quartets contained 102 loci with a mean length of 851 bp, and a mean number of parsimony-informative sites of 140. The subset of loci with>60% fully resolved quartets contained 63 loci with a mean length of 1056 bp, and a mean number of parsimony-informative sites of 178. The subset of loci with>70% fully resolved quartets contained 22 loci with a mean length of 1490 bp, and a mean number of parsimony-informative sites of 302.

Phylogenetic analysis of each of the subsets resulted in trees with identical topologies that were congruent with the phylogeny produced in the analysis of the unpartitioned full dataset (i.e. Fig. 3). In general, all subset trees had slightly lower branch support than the full-dataset tree. Among subset trees, the>70% tree had the lowest branch support, with exception to the placement of Philotheca angustifolia sister to P. tomentella and P. difformis with low-to-moderate support (UFBoot: 91; SH-aLRT: 82.9). The>50 and>60% trees did not resolve the position of this species. Despite the removal of potentially noisy or discordant loci, none of the subset trees contained a supported resolution for the backbone polytomy; the>60% subset tree displayed the highest support for short branches of the polytomy, but these branches were still unsupported and resulted in star-like resolution of the polytomy in our network analysis ( Fig. 5 b). We found a significant difference in the number of resolved quartets between the three gene tree clusters from the tree space analysis (one-way ANOVA test; F (2,145) = 18.82, P = 5.39 × 10 −8; Kruskal– Wallis rank sum test; H = 40.437, d.f. = 2, P = 1.657 × 10 −9). The mean number of resolved quartets for Clusters 1, 2 and 3 was 3019 (σ = 586), 2434 (σ = 541) and 2213 (σ = 1030) respectively, indicating that Cluster 1 contained the most phylogenetically informative loci on average.

Tree topology tests

Owing to high UFBoot and posterior probability support for four clades forming the backbone polytomy, our tree-topology testing investigated support for the 15 possible relationships among these clades ( Fig. 6 a). Log-likelihood values for each topology ranged from −494 970.83 to −494 975.48. Expectedly, the topology of the most likely tree used as input was found to have the highest log-likelihood (−494 970.83); this topology was identical to the 15th possible resolution of the polytomy ( Fig. 6 a). Five topologies were found to have slightly higher log-likelihoods than the others ( Fig. 6 b: Topologies 15, 5, 10, 13, 14). The three most likely topologies (15, 5, 10) were the only topologies to resolve the Clade 1 sister to the rest of the group. Clades 1 and 2 were grouped sister to Clades 3 and 4 in Topology 13, and in Topology 14 Clade 2 was sister to the rest of the group, with Clade 1 being sister to Clades 3 and 4. The AU test rejected four topologies ( P <0.05): 7, 9, 12 and the hard polytomy. The hard polytomy was found to be one of the least likely topologies, having a low log-likelihood score (−494 975.27) and being statistically supported by

(a) 1.00

0.75

0.50

0.25

0.00 (b)

Chorilaena

Asterolasia

Nematolepis

P. sect. Erionema

Geleznowia Halfordia

Drummondita

P. sect. Philotheca

only the unweighted and weighted SH tests ( P = 0.225 and P = 0.648), which did not reject any topologies ( P -values greater than 0.05 for all topologies). The bootstrap proportion and equal likely weights test did not place the hard polytomy topology in the 95% confidence set of trees. Summary statistics from topology testing are provided as Supplementary Table S4 .