Phylogeography and classification of Dusty Miller (Spyridium parvifolium; Rhamnaceae): a morphologically variable shrub from south-east Australia

Spyridium parvifolium is a widespread and morphologically variable shrub from south-eastern Australia. Several varieties have been recognised, and there is disagreement on the accepted taxonomy between Australian states. This study investigated the phylogeography of the species and assessed genetic distinctiveness of its morphological variants. Nuclear ribosomal DNA and complete chloroplast genomes from seventy-two samples of S. parvifolium and seven samples from closely related species were sequenced and analysed using both Bayesian and maximum likelihood phylogenetic methods. The results showed incongruence in the placement of several associated taxa (S. cinereum, S. obcordatum and S. daltonii), plausibly due to long branch attraction, introgression or incomplete lineage sorting. Spyridium parvifolium was resolved as paraphyletic in both phylogenies, with accessions from west of the Murray Darling Depression divergent from those east of the Depression. We found evidence of isolation within S. parvifolium on the inland side of the Great Dividing Range and recent gene flow across Bass Strait. The variants of S. parvifolium were not supported as genetically distinct, and with the prevalence of several variants at single sites and morphological intergrades between variants, we conclude that the taxon is a single, morphologically variable species and that no infraspecific classification is warranted.


Introduction
Rhamnaceae is a medium-sized family of trees, shrubs, and climbers found throughout the world, excluding Antarctica and the Artic. It includes c. 960 species across c. 60 genera and 11 tribes (Hauenschild et al. 2016;Ladiges et al. 2005;Medan and Schirarend 2004). Diagnostic characters for Rhamnaceae are their small flowers with stamens opposite the (often) hooded petals, which in part cover the anthers . Most Australian Rhamnaceae are from a single tribe, Pomaderreae (c. 240 species), the second largest tribe in the family, consisting of species endemic to Australia and New Zealand. The tribe is easily distinguished within Rhamnaceae by the characteristic stellate hairs found on at least some (if not all) of the plant, which can include the floral parts (Kellermann 2020). Molecular studies focused on the family Rhamnaceae have consistently resolved the tribe as monophyletic (Hauenschild et al. 2016;Onstein et al. 2015;Richardson et al. 2000;van Santen and Linder 2020). One genus of the tribe Handling Editor: Christoph Oberprieler.
Pomaderreae, Spyridium, is a group of c. 45 species endemic to semi-arid and temperate regions of southern Australia (Kellermann and Barker 2012). Spyridium species are commonly prostrate-medium shrubs, with tomentose stems and inflorescence of small flowers clustered in cymose heads (Coates 1996). The name Spyridium is derived from the Greek spyridion meaning a small basket, a reference to the inflorescence which is surrounded by leaf-like bracts (Perrin 2018). In many species, these leaf-like bracts (often referred to as floral leaves) are distinct from the stem leaves, covered in velvety hairs and therefore appearing white (Coates 1996). The monophyly of Spyridium is supported in multiple molecular phylogenies (Hauenschild et al. 2016;Kellermann and Udovicic 2007;Kellermann et al. 2005;Richardson et al. 2004).
Spyridium parvifolium is a morphologically variable and widespread shrub from south-eastern Australia ( Fig. 1a; Atlas of Living Australia 2020; Canning and Jessop 1986;Curtis and Morris 1993;PlantNET 2021;VicFlora 2021). Found in dry sclerophyll forests and extending to heathlands (VicFlora 2021), S. parvifolium has a disjunct distribution with two main divides: one across Bass Strait between the states of Victoria and Tasmania and the second across the Murray Darling Depression (MDD) between western Victoria and south-central South Australia (SA; Fig. 1a; Atlas of Living Australia 2020). The Bassian Plain underlying the Bass Strait, the body of water separating mainland Australian from Tasmania, has been repeatedly exposed during glacial periods, including the last glacial maximum c. 16-18 ka (Bowler 1982) with the land bridge exposed until c. 13.5-12 ka (Galloway and Kemp 1981). Many other Australian dry forest plant species have disjunct distributions across Bass Strait (Worth et al. 2017). Conversely, the MDD which is currently above sea level has experienced several marine incursions including during both the Miocene, c. 15 mya, and Pliocene, c. 6 mya (Bowler 1982). The most recent shoreline retreat exposed the full extent of the MDD c. 1 mya and gave rise to geomorphology and edaphic features that still impact species distributions in the MDD (Bowler et al. 2006). Many Australian plants have disjunct distributions across the MDD (French et al. 2016), and several studies have found that many of these species show a major genetic divergence across the MDD. This includes Eucalyptus behriana (Fahey et al. 2021), Hardenbergia violacea (Larcombe et al. 2011), Zieria veronicea (Neal et al. 2019), Eucalyptus globulus subsp. bicostata (Jackson et al.  (d) in south-eastern Australia based on filtered records from Atlas of Living Australia (2020). Australian state abbreviations are as follows: SA South Australia, Vic. Victoria, Tas. Tasmania, NSW New South Wales. Also labelled are the southern section of the MDD, Lowan Sands of the MDD, the NCP, and the southern highlands/alps section of the GDR ▸ 1999) and Themeda triandra (Hayman 1960). Spyridium parvifolium is also discontinuously distributed across Victoria from its western to eastern boundaries and into southeastern New South Wales.
Known commonly as Dusty Miller (VicFlora 2021), S. parvifolium is distinguished from other species of Spyridium by grey-green floral leaves and elliptic-obovate (or orbicular) stem leaves with conspicuous secondary venation and an upper surface that usually has a hispid-sub-velutinous indumentum (VicFlora 2021). Four varieties of the species have been recognised based on morphological features (examples provided in Fig. 2): var. molle with grey-green hairy leaves and described from the Flinders and Cape Barren islands, Tasmania; var. grande with large leaves (> 25 mm long) described from the Dandenong Ranges, Victoria (Fig. 2a); var. hirsutissimum with long hispid hairs on the stems and described from the Grampians, Victoria (Fig. 2b); and the typical var. parvifolium which accommodated all other variation including plants with stem leaves that are "soft and often hairy on the upper side" (Bentham 1863). Many intermediates between the varieties have been noted (Vic-Flora 2021), and consequently, herbaria in Australian mainland states no longer recognise these varieties as distinct (CHAH 2020). However, in Tasmania two varieties (var. parvifolium and var. molle) are still recognised as separate taxa and both are listed, and protected, as Rare in Tasmanian legislation (Threatened Species Section, 2016a, b). To complicate matters further, additional variants of the species have been informally described in Victoria (Bull 2014;Vic-Flora 2021). These include a 'Brisbane Ranges form', with small (3-5 mm long), notched leaves (Fig. 2c); a 'Rocky Sites form', with grey-green hairy leaves (similar to var. molle), found in rocky sites in Gippsland (Fig. 2d;VicFlora 2021); and a small, spreading variant from East Gippsland, marketed in nurseries as 'Prostrate form' or 'Nimbus' (Bull 2014).
This array of historical varieties and informal variants (both here forth referred to as 'variants') and conflicting taxonomy in different Australian states has led to suggestions that a more detailed investigation into the species is required (VicFlora 2021). There has been little molecular research into the species to date, with only a few samples included in broader phylogenetic studies focused on the tribe and family. Kellermann et al. (2005) included one sample of S. parvifolium in a study of the tribe based on nuclear ribosomal DNA (nrDNA) sequences (ITS). That sample was found to be divergent from all other south-east Australian Spyridium species included in the study. Kellermann and Udovicic (2007), extending the previous study by using chloroplast DNA (cpDNA) sequences (trnL-trnF), found S. parvifolium formed an unsupported clade with S. thymifolium (from SA). Hauenschild et al. (2016) included three samples of S. parvifolium in a phylogeny of Rhamnaceae with the combined  (ITS) and cpDNA (trnL-trnF) analysis placing all three samples in a clade with S. daltonii and S. thymifolium. More recently, Clowes et al. (2022) included eight samples of S. parvifolium in a molecular (nrDNA and chloroplast genomes) phylogeny of Spyridium, with the species found in an unsupported clade (Bayesian posterior probability, 0.73) with S. daltonii, S. obcordatum and S. cinereum in the nrDNA phylogeny and resolved in a strongly supported clade (polytomy; Bayesian posterior probability, 1.0) with S. daltonii in the cpDNA tree. The three most closely related species in the nrDNA tree are narrow range endemics with overlapping (or nearly so) distributions with S. parvifolium ( Fig. 1b-d). Spyridium daltonii is also known to hybridise with S. parvifolium, resulting in the named hybrid S. ×-ramosissimum (Kellermann 2006).
The current study aims to extend the work of Clowes et al. (2022), with a specific focus on S. parvifolium. Spyridium parvifolium is sampled from across its geographic range, including all named varieties and known morphological variants (VicFlora 2021), together with representatives of other species of Spyridium. Both nrDNA and complete chloroplast genomes (cpDNA) are used to investigate the phylogeographic history of the species and to compare this to other species of dry sclerophyll forests in south-eastern Australia. The data are also used to assess patterns of genetic variation in S. parvifolium and whether they provide grounds for recognition of distinct species or infraspecific taxa.

Taxon selection and sampling
In total, 213 individuals representing 49 species of Spyridium and three outgroup genera of tribe Pomaderreae (Pomaderris, Cryptandra and Trymalium) were sampled, including several phrase name taxa. Of these, 66 samples were newly sequenced accessions of S. parvifolium (64 samples) and S. daltonii (two samples) and 145 were previously included in Clowes et al. (2022).
Most samples were collected during fieldwork. Where plants at a site appeared morphologically homogeneous, one voucher specimen (with duplicates) was collected; however, where a site showed morphological variation, more than one voucher was collected and lodged at The University of Melbourne Herbarium (MELU). Based upon State collection and associated permit requirements, duplicates were sent to the National Herbarium of Victoria (MEL), the Tasmanian Herbarium (HO) or the State Herbarium of South Australia (AD). Leaf material was also collected from several individuals at each site (including the vouchered plant/s) and dried in silica gel, with one to four samples from each site included in this study.
Seventy-two samples of S. parvifolium from 36 locations were included in this study (Table 1; Fig. 3), of which eight were also included in Clowes et al. (2022). Samples were collected from across the geographic range and morphological variants of the species. Reference works used to determine the variety or variant of samples are provided in Table 1. For the Brisbane Ranges form, sites identified as this variant were collected from the Brisbane Ranges area and had small, sometimes notched leaves, as described by VicFlora (2021). However, in some cases the leaves were more than 5 mm long, which is longer than the description suggests. Despite this, we have referred to these samples as the Brisbane Ranges form.
Three samples of S. daltonii were also included, collected from two different locations within the Grampians ranges: Grampians Road and Mt William ( Table 1). One of these accessions was also included in Clowes et al. (2022).

DNA extraction and library preparation
Total genomic DNA was extracted from c. 60 mg of silica dried leaf material following a modified CTAB (cetyl trimethylammonium bromide) DNA extraction protocol (McLay 2017; Shepherd and McLay 2011) based on Doyle and Doyle (1987). Where possible young, dried leaf material from stem tips or dried floral leaves were selected for extraction. DNA quality and quantity were recorded using a Nanodrop 2000 (NanoDrop Products) and Qubit 2.0 fluorometer (Invitrogen). For some samples, in an attempt to improve DNA quality, an additional clean-up was performed using a sodium acetate/ethanol wash following Lamitina Lab Protocols (2007). Genomic DNA was prepared for multiplexed sequencing using the library preparation protocol of Schuster et al. (2018) with modifications per Clowes et al. (2022). Sequencing was performed using an Illumina NextSeq 500 (2 × 150 bp) sequencing platform at the Walter and Eliza Hall Institute of Medical Research Genomics Hub (WEHI).

Sequence assembly and alignment
Quality filtering and base calling were conducted at WEHI with Illumina pipeline software (v.1.7 or later) and pre-processed with custom scripts, as per Schuster et al. (2018). Sequence reads were imported into Geneious 10.0.1 or newer (Kearse et al. 2012) and trimmed, paired and assembled as per Clowes et al. (2022). Sequences were aligned in MAFFT v7.308 (Katoh et al. 2002;Katoh and Standley 2013) using automatic algorithm selection and default settings. Aligned sequences were reviewed in Geneious and manually realigned following the protocol described in Clowes et al. (2022). The final nrDNA alignment was 6,386 bp long and the cpDNA alignment totalled 168,343 bp. The nrDNA sequences were partitioned as follows: partial external
For the nrDNA dataset, the best fit models selected according to the Akaike information criterion (AIC) were: 5'ETS GTR + G, 18S HKY + I + G, ITS1 SYM + G, ITS2 HKY + G, 26S GTR + I + G and 3'ETS + NTS HKY + G. For the cpDNA data, the best fit models according to AIC were: CDS GTR + I + G, tRNA K80, rRNA HKY and spacers GTR + I + G. For the cpDNA data, the best fit models according to AIC were: CDS GTR + I + G, tRNA K80, rRNA HKY and spacers GTR + I + G. Bayesian inference was undertaken in MrBayes XSEDE (Ronquist and Huelsenbeck 2003) using the CIPRES portal (Miller et al. 2010). For the nrDNA alignment, two independent analyses with four chains (Markov chain Monte Carlo) were run for 10 M generations, sampling every 1,000 steps, with a burnin of 25% and Dirichlet distribution unlinked. For the cpDNA alignment, the analysis was run for 2.5 M generations and sampled every 500 steps. Output files were viewed in Tracer (Rambaut et al. 2014) to check for convergence. The average standard deviation of split frequencies was also reviewed and confirmed to be below 0.01 upon completion of analyses. Branches with values of < 0.95 PP were considered unsupported. Maximum likelihood (ML) analysis was completed with IQ-Tree using default settings (Nguyen et al. 2015). Model testing was automated in IQ-Tree, with ModelFinder (Kalyaanamoorthy et al. 2017) used to select models for each partition using Bayesian information criterion (BIC). Best fit models for the nrDNA data were: 5'ETS TPM2u + F + G4, 18S K2P + I + G4, ITS1 TIM2e + G4, ITS2 TNe + G4, 26S TIM3 + F + I and 3'ETS + NTS TN + F + G4. Best fit models for the cpDNA sequences were: CDS TVM + F + I, tRNA K2P, rRNA HKY + F and spacers TVM + F + I + G4. The collected from the same location as other samples, the voucher numbers provided may be another representative collected from that location; the voucher collecting numbers have therefore been provided in brackets next to each voucher number for those species. Herbarium abbreviations follow New York Botanic Gardens (2020). To standardise location details, for samples collected from mainland Australian, the applicable state has been provided and either the reserve name or the nearest town (where the sample was collected outside a reserve system). For samples collected on islands, location details include the island name and associated state. Location abbreviations used in figures and text are underlined, and latitude and longitude coordinates are provided for each voucher. GenBank accession numbers are also provided (nrDNA or cpDNA). References: Clowes et al. (2022) and New York Botanic Gardens (2020) a A population of S. parvifolium was also observed (and collected) from this site, as were samples of S. ×ramosissimum nrDNA analysis contained 274 parsimony-informative characters and the cpDNA analysis included a total of 3,468 parsimony-informative characters. Branches with values of < 95% ultrafast bootstrap support (UFBS) were considered unsupported. Branches that were supported in one analysis (for example BI) but unsupported in the other analysis (for example ML) were considered unsupported overall.

Long branch attraction tests
Preliminary nrDNA phylogenies placed samples of two species (S. cinereum and S. obcordatum) together on comparatively long branches and in positions incongruent with those on cpDNA trees. To assess whether long branch attraction (LBA) could be influencing the positions of these species in the nrDNA trees, BI and ML analyses were re-run (as described above), with each species separately removed from the dataset, following an approach suggested by Bergsten (2005). The resulting consensus trees, as viewed in FigTree (Rambaut and Drummond 2012), are provided in the Electronic Supplementary Material (Online Resources 1-4).

Nuclear rDNA phylogeny
Overall, the topology in the BI (Fig. 4a) and ML (Fig. 4b) nrDNA trees was similar, though there were several unsupported (< 0.95 PP, < 95% UFBS) incongruencies between them. Clades relevant to S. parvifolium are shown in detail in Fig. 4, with all other clades collapsed. The BI tree (Fig. 4a) resolved two sister clades (B and C) that included all S. parvifolium samples, along with samples of S. daltonii (nested in clade B) plus a clade (D) of S. obcordatum and S. cinereum nested within clade C. Those same clades were also present on the ML tree, but differed in their relationships. In the ML tree (Fig. 4b), clade C was not sister to clade B, but rather was sister, with 57% UFBS, to a clade (E) of other Spyridium species mostly endemic to SA. The relationship of S. daltonii samples to other members of clade B, and that of clade D to other members of clade C also differed in the ML tree, but none of the relevant nodes had strong support. Other differences between the BI and ML trees were also unsupported and made no notable difference to the results. Spyridium parvifolium was not resolved as monophyletic in either the BI or ML trees (Fig. 4a, b). Four samples from west of the MDD in SA, Kaiserstuhl and Mt Remarkable, formed a clade (B) with the three S. daltonii samples; this clade (B) was sister to the main S. parvifolium clade (C) in the BI tree. The previously identified morphological variants of S. parvifolium do not form clades in the tree (Fig. 4a). Spyridium obcordatum and S. cinereum were both resolved on long branches within clade D (Fig. 4a, b). When samples of S. cinereum were removed from the alignment and the analyses re-run, the placement of S. obcordatum changed; in both the BI and ML re-analyses, S. obcordatum was placed as sister (0.93 PP; 95% UFBS) to a clade (0.87 PP; 77% UFBS) containing all other samples of clade A (Online Resources 1, 2), congruent with its placement in the cpDNA phylogeny (Fig. 5). When S. obcordatum was removed from the alignment and the analyses re-run, S. cinereum was found in a basal polytomy in clade C in both BI and ML trees (Online Resources 3, 4), congruent with the BI tree Fig. 4a.

Chloroplast genome phylogeny
Overall, the relationships of S. parvifolium samples in the BI and ML analyses of cpDNA genomes were congruent, with differences between the trees unsupported (Fig. 5). Clades relevant to S. parvifolium are shown in detail on the BI tree (Fig. 5), with all other clades collapsed. Despite all S. parvifolium samples being found in a single clade (clade G1), the species was rendered paraphyletic as Spyridium daltonii was nested within clade G2. The three accessions of S. daltonii do not form a monophyletic grouping, all falling near the basal polytomy of this clade (Fig. 5). Two samples of S. daltonii (one from Mt William and one from Grampians Road) were placed together with strong support, while the third sample (also from Mt William) was placed separately on a comparatively long terminal branch. Spyridium obcordatum was resolved as sister to clade G1, within clade G, whereas S. cinereum was resolved in a distant clade (F).
Within clade G1 there was some geographic clustering of S. parvifolium samples (Fig. 5, 6). Where more than one sample of S. parvifolium was included for a site, they mostly grouped together in the tree; exceptions were samples from Jancourt, Wilsons Promontory, Distillery Creek, Croajingolong and Scott Creek sites (Fig. 5). Samples of S. parvifolium collected from west of the MDD (Mt Remarkable, Kaiserstuhl and Scott Creek) formed a grade at the base of clade G1 (Figs. 5, 6a), subtending a clade (G2) including all accessions from east of the MDD (Fig. 5). Within G2, an inland GDR (Great Dividing Range) clade G8 was resolved, as was a south-eastern Victorian (coastal NSW) only clade (G6). Tasmanian and Flinders Island accessions were found in four clades (Figs. 5,6), and south-eastern Victorian samples were found throughout clade G2, including in three of the four Tasmanian and Flinders Island clades (Fig. 6a-c).
Of the morphological variants of S. parvifolium, only one, var. hirsutissimum was limited to a single clade (G11, Fig. 5). All samples of the Brisbane Ranges form were found in clade F3, although this clade also contained some samples of var. parvifolium and var. molle from Tasmania. All other variants were found scattered throughout the tree within clade G2.

Discussion
In this study we have presented a molecular analysis of S. parvifolium, across the species range and incorporating all morphological variants previously discussed in the literature. These results provide information about the phylogeographic history of the species and enable an interpretation of the classification of the variants. However, several differences between the nrDNA and cpDNA phylogenies related to the placement of associated taxa (S. cinereum, S. obcordatum and S. daltonii) warrant discussion first.

Incongruent placement of Spyridium cinereum and S. obcordatum in nrDNA and cpDNA phylogenies
The placement of S. cinereum and S. obcordatum differed markedly between the nrDNA and cpDNA phylogenies (Figs. 4,5), as previously found in the analyses of Clowes et al. (2022). The causes of this incongruence could be biological (e.g. resulting from incomplete lineage sorting or introgression between taxa), as commonly seen in many plant groups (Barrett et al. 2018;Rieseberg and Soltis 1991;Wiley and Lieberman 2011), or could be an artefact of analysis resulting from LBA (Bergsten 2005) or, in the case of nrDNA sequences, from combining different paralogues in the analysis (Bailey et al. 2003;Bayly and Ladiges 2007). The positions of these species in the cpDNA tree (Fig. 5) are more in line with current species-level taxonomy, and morphological features with S. obcordatum resembling, but are distinguished from the Prostrate form of S. parvifolium, and with S. cinereum being considerably more distinctive, in line with placement in clade F with a number of other species (several with similar notched leaves like S. cinereum, for example).
It is the position of S. cinereum and S. obcordatum in the nrDNA tree (Fig. 4a) that seems most anomalous, especially given the very long branches on which they sit relative to surrounding ones. The fact that placement of S. obcordatum moves, and matches that seen in the cpDNA, when S. cinereum is excluded from the nrDNA dataset (Online Resources 1, 2) could be considered evidence that LBA is at play (Bergsten 2005). Other explanations are less plausible, but cannot be ruled out. Of those explanations, incomplete lineage sorting, especially in nrDNA arrays which have small effective population sizes (Buckler and Holtsford 1996) and, therefore, short coalescence times as a result of concerted evolution (Arnheim 1983), seems unlikely and would not account for the unusual branch lengths of S. cinereum and S. obcordatum. Confusion of nrDNA paralogues also seems unlikely, as we scrutinised sequence reads and assemblies closely and saw no evidence of distinctive rDNA paralogues within samples, and especially in S. cinereum and S. obcordatum. Introgression, on the other hand, which can have variable outcomes for sequences of nrDNA arrays (Álvarez and Wendel 2003) could conceivably result in distinctive sequences that combine features of different parental lineages. Although putative hybrids between S. parvifolium and S. cinereum or S. obcordatum have not been reported, S. parvifolium does grow in close proximity to the other two (Atlas of Living Australia 2020), making introgression plausible. The example of S. ×ramosissimum shows that S. parvifolium can hybridise with relatives when they co-occur; however, in the case of S. cinereum, it would have to have been historical rather than recent hybridisation to account for the monophyly of the two highly disjunct samples of S. cinereum (Fig. 1c) in the nrDNA tree.
Broader population sampling and additional nuclear DNA data, preferably from a broad range of loci, are needed to more fully understand the relationships of S. cinereum and S. obcordatum to S. parvifolium and other species in the genus. They could confirm or refute the nrDNA relationships resolved here and could be used to test for evidence of introgression. Target sequence capture methods (e.g. Hyb-Seq; Dodsworth et al. 2019) could be ideal for this purpose. Spyridium daltonii, endemic to the Grampians National Park in western Victoria, was nested within S. parvifolium in both the cpDNA and nrDNA trees (Figs. 4a, 5). Although there is uncertainty about its exact placement because nodes at the base of clades B and C in the nrDNA tree are very weakly supported (allowing the possibility that S. daltonii could be sister to S. parvifolium), and because it is placed in a large polytomy near the base of clade G2 in the cpDNA tree, its nested position in the cpDNA tree is strongly supported.

Placement of Spyridium daltonii among samples of S. parvifolium
Spyridium daltonii co-occurs with S. parvifolium at several locations in the Grampians (Atlas of Living Australia 2020) and the two species are also known to hybridise, producing the hybrid taxon S. ×ramosissimum (Kellermann 2006). Given this, the nesting of S. daltonii in the S. parvifolium clade, especially in the cpDNA phylogeny, could be attributed to introgression resulting in chloroplast capture. Although the molecular data presented here might suggest that S. daltonii could be synonymised with S. parvifolium, the two are readily distinguished on the basis of morphology to the extent that treating them as conspecific would be difficult to justify. As such, further sampling of S. daltonii individuals and of additional nuclear DNA markers, is needed to assess whether there is any incongruence between nuclear and cpDNA relationships and the extent to which there is evidence of nuclear differentiation or introgression between the two species.

Genetic divergence in Spyridium parvifolium across the Murray Darling Depression
This study has revealed genetic divergence associated with the disjunct distribution of S. parvifolium across the MDD. The SA populations (Mt Remarkable, Kaiserstuhl and Scott Creek; Fig. 3) show genetic differences from more easterly populations based on both cpDNA and nrDNA sequences, although their relationships are resolved differently in the two datasets (Figs. 4a, b, 5). Such divergence is not unexpected as other plant species are disjunct and genetically divergent across the MDD, for example: Eucalyptus behriana (Fahey et al. 2021), Hardenbergia violacea (Larcombe et al. 2011), Eucalyptus globulus subsp. bicostata (Jackson et al. 1999) and Themeda triandra (Hayman 1960). French et al. (2016) highlighted three factors that have created barriers to dispersal across the MDD, potentially leading to geographic disjunctions in this area: past marine incursions, subsequent edaphic factors and hostile climates Fig. 4 a Bayesian inference (BI) 50% majority rule consensus tree based on analysis of nuclear ribosomal DNA (nrDNA). Only clades relevant to Spyridium parvifolium are shown (A-D) with all other unrelated clades collapsed. b A reduced selection of the ML consensus tree shown only the SA accessions of S. parvifolium and those of S. daltonii, S. cinereum and S. obcordatum. The S. parvifolium clade contains the same samples of the species as shown in Fig. 4a. In both Fig. 4a and 4b only values for unsupported nodes are showing (i.e. < 0.95 PP on the BI tree and < 95% UFBS on the ML tree) with the values of supported nodes not shown (i.e. ≥ 0.95 PP, ≥ 95% UFBS). Coloured shapes next to accession labels indicate variants and geographic regions for S. parvifolium samples, as described in the caption for Fig. 3. The contents of collapsed clades, together with all support values, can be viewed by accessing this phylogeny on Dryad (https:// doi. org/ 10. 5061/ dryad. 573n5 tb9t). Clade E contains 70 accessions of Spyridium predominantly endemic to SA and south-eastern Australia more broadly. The NSW/QLD endemics clade contains six samples of Spyridium, the Tasmanian endemics clade contains 19 Spyridium samples, and the predominantly WA endemics clade contains 33 Spyridium samples ◂ associated with Quaternary glacial periods. The MDD and the Naracoorte Coastal Plain (NCP) have experienced multiple marine incursions including during the Miocene (c. 15 mya) and Pliocene (c. 6 mya; Fig. 3; Bowler 1982). Subsequent to these incursions, geomorphology and edaphic factors that remained were also likely a barrier to recolonisation by some plants. During marine incursions in SA, elevated areas including the Lofty Ranges (e.g. Scott Creek and Kaiserstuhl) may have provided refugia for flora species (Byrne 2008), such as S. parvifolium, which is found scattered throughout the southern Lofty Ranges today (Atlas of Living Australia 2020). By c. 1 mya, the shoreline had retreated to the inland margin of the NCP (Fig. 3; Bowler et al. 2006) with the final retreat resulting in additional edaphic barriers (Specht 1972) including calcareous sands and limestone of the NCP. Compounding these incursions and retreat 'barriers', were glacial periods of the Quaternary, which resulted in cooler, arid climates in many places including the MDD that inhibited the establishment of plants in areas such as mobile dunes. Given the history of potential barriers to plant dispersal across the MDD, the geographic disjunctions in species' ranges across this area could be explained by either vicariance (e.g. species predates barrier) or dispersal (i.e. species younger than barrier and disperses across it).
For S. parvifolium, the genetic pattern in the cpDNA phylogeny (Fig. 5) is consistent with either vicariance (with early differentiation in the west and a single monophyletic group in the east) or with dispersal from west to east. The Spyridium phylogeny (Clowes et al. 2022) implied an early divergence of WA taxa during the Miocene; therefore, the age of the clade including S. parvifolium is such that the Pliocene marine incursion, edaphic barriers or aridification are plausible explanations for the observed SA divergence. Spyridium parvifolium is absent from the Lowan Sands and adjoining regions in SA (Fig. 1a), which could either be associated with current edaphic conditions, or due to previous climatic conditions (e.g. with dune mobilisation in glacial times; Bowler et al. 2006;Conn 1993).
Although S. parvifolium is effectively absent from the MDD, it is found in southern parts of the NCP (Fig. 1a), which suggests that limestone soils are not necessarily a barrier for the species. According to the Australian State of Environment Report, less than 40% of native vegetation remains in the NCP region (Metcalfe and Bui 2017), so it is possible that additional populations of S. parvifolium were present in this area and have been cleared since European settlement. Presence of S. parvifolium on the NCP indicates that some recolonisation of this species has occurred since the sea levels dropped around 1 mya, which could have resulted in the reconnection of previously isolated and divergent eastern and western lineages. Such a process was inferred in the genus Correa (Rutaceae) by French et al. (2016), who identified divergent eastern and western cpDNA clades that overlapped, potentially through reconnection of previously separated lineages, on the Fleurieu Peninsula in SA (i.e. west of the MDD, near the western edge of the NCP).
If S. parvifolium has dispersed across the NCP, e.g. since those plains were exposed above sea level in the last million years, reconnection and intermixing of lineages previously separated across the MDD could potentially explain the contrasting placement of Scott Creek samples in the cpDNA and nrDNA trees. Whereas the cpDNA tree places those samples close to others from west of the MDD, as part of a western grade at the base of Clade F1 (Fig. 5), the nrDNA trees place them separate from the other western samples in a clade (C) of samples from east of the MDD (Fig. 4a). Given the location of Scott Creek near the western edge of the NCP (Fig. 3), it is tempting to speculate that these plants might combine "western" chloroplasts and "eastern" nrDNA because of intermixing of lineages facilitated by dispersal east to west across the NCP. Such a hypothesis could be tested by more intensive sampling of populations from the SA and southwest Victoria, and by using variable genomewide nuclear markers (e.g. RADseq or similar; Davey and Blaxter 2010), together with cpDNA sequencing, to investigate genetic patterns across this region.
West of the MDD, a north-south pattern of divergence of S. parvifolium was identified, with the most northern population (e.g. Mt Remarkable; Fig. 3) divergent in both phylogenies (Figs. 4a,5), and especially in the cpDNA tree. Throughout the Lofty Ranges, S. parvifolium has a somewhat continuous distribution with populations separated by relatively short distances (Figs. 1a). North of the Lofty Ranges, S. parvifolium has a discontinuous distribution with only three main localities; Mount Remarkable, Telowie Gorge Conservation Park (both part of the Southern Flinders Ranges) and Tothill Ranges (Atlas of Living Australia 2020). This northern divergence (Mt Remarkable) and discontinued distribution is interesting; however, no other comparative studies with north-south SA samples included Bayesian inference (BI) 50% majority rule consensus tree based on analysis of chloroplast genomes (cpDNA). Posterior probabilities (PP) < 0.95 are shown, followed by ultrafast bootstrap (UFBS) values < 95%; support values are not shown for nodes with ≥ 0.95 PP or ≥ 95% UFBS. Coloured shapes next to accession labels indicate variants and regions of S. parvifolium, described in Fig. 3 caption. Clades relevant to S. parvifolium are labelled F-G and G1-G11. Clades not focused on S. parvifolium that have been collapsed, can be viewed accessing this phylogeny on Dryad (10. 5061/ dryad. 573n5 tb9t). Collapsed clades within clade F contain 37 accessions of Spyridium (predominantly endemic to SA and south-eastern Australia more broadly). The predominantly SA endemics clade contains 37 Spyridium samples, the predominantly WA endemics clade contains 28 Spyridium samples, the NSW/QLD endemics clade contains six Spyridium samples, and the Tasmanian endemics clade contains 33 Spyridium samples ◂ have been identified to determine whether this is an isolated finding or pattern also seen in other members of the flora across these areas. The conservation of S. parvifolium across its range in SA is important, because each population (particularly those isolated north of the Lofty Ranges) may be genetically distinct. Further research into SA populations of other taxa found across south-east Australia is warranted, to investigate whether this finding is repeated in other dry sclerophyll forest species.

The Great Dividing Range is a biogeographic barrier in Spyridium parvifolium
The GDR is identified as a barrier to seed-mediated gene flow within S. parvifolium, with samples collected from the inland side of the range forming a clade in the cpDNA tree, distinct from samples collected on the coastal side of the GDR and other parts of Victoria (Fig. 5). This result is interesting because there are relatively few comparable studies of south-east Australian dry sclerophyll forest species with GDR disjunctions, and of those available, the findings were different to those of S. parvifolium. For example, no  (Broadhurst et al. 2018), both widespread in dry sclerophyll forest of south-eastern Australia. Given the limited number of studies in this area, investigation of more species disjunct and separated by the high parts of the GDR is warranted to see whether there are common patterns. For S. parvifolium in the southern highlands and alps, glacial cycles may have exaggerated the physical barrier, with a snowline that dropped 1,000 m below the range peaks (Frakes et al. 1987). The extension of these alpine and subalpine conditions downward is likely to have significantly fragmented forest vegetation (Byrne 2008), creating even greater potential for isolation of sites on either side of the range. Today, variation in climate and elevation over the GDR may act as a continued physical barrier for species (Milner et al. 2012), such as S. parvifolium.

Recent gene flow in Spyridium parvifolium across Bass Strait
In contrast to the barrier associated with the GDR, there is evidence that seed-mediated gene flow in S. parvifolium has continued until relatively recently across Bass Strait, with multiple close cpDNA relationships identified, predominantly between south-western Victoria, Wilsons Promontory, Tasmania and Flinders Island (Figs. 5, 6). Most clades with Bass Strait connections (G3, G4, G5) also show a degree of sequence divergence between mainland and Tasmanian samples within the clade, except for clade G9 in which there is little differentiation. This sequence divergence suggests the most recent connection across Bass Strait occurred between Wilsons Promontory-Cape Liptrap in south Gippsland (Victoria) and Sisters Beach in north-west Tasmania (Clade G9; Figs. 5, 6d). Other studies have found a similar pattern of recent gene flow between Victoria and Tasmania, for example in Correa (French et al. 2016), Zieria veronicea (Neal et al. 2019), Astroloma humifusum, Epacris impressa, Bursaria spinosa and several other dry forest species (Worth et al. 2017). During glacial maxima, sea levels dropped below those of today exposing the Bassian Plain (Larcombe et al. 2011), with glacial/interglacial oscillations occurring every 100 ka over the last 2.5 M years (Frakes et al. 1987). The last glacial maximum occurred 16-18 ka (Bowler 1982) with the land bridge across Bass Strait exposed until c. 13.5-12 ka (Galloway and Kemp 1981). Evidence suggests that at least some areas of the Bassian Plain were covered in eucalypt woodland (Hope 1978(Hope , 1994Kirkpatrick and Fowler 1998), which may have been suitable habitat for S. parvifolium. If so, this could explain the main points of connection between Victoria and Tasmania identified here and for other dry sclerophyll forest species (Worth et al. (2017): the Otways (e.g. Jancourt, Port Campbell and Distillery Creek), south Gippsland (e.g. Cape Liptrap and Wilson Promontory), north-east Tasmania (e.g. Bay of Fires) and to a lesser extent north-west Tasmania (e.g. Sisters Beach). Over-land dispersal at times of low sea level seems more plausible than multiple chance dispersals over water, as Spyridium species lack obvious features to enable seed dispersal (Coates 1996;Coates et al. 1999).

Spyridium parvifolium is a single, morphologically variable taxon
This study did not resolve the morphological variants of S. parvifolium, including the two currently recognised varieties (var. parvifolium and var. molle) from Tasmania and Flinders Island, as monophyletic or as more genetically differentiated than other populations of the species (Figs. 4a,  5). The cpDNA phylogeny did resolve var. hirsutissimum as monophyletic (Fig. 5); however, both samples were collected from a single site and the cpDNA phylogeny showed strong geographic structuring, such that var. hirsutissimum samples were no more distinct than those from a range of other collecting sites. As discussed above, recent gene flow or close cpDNA relationships are evident between many populations from Tasmania, Flinders Island and Victoria, including between var. parvifolium, var. molle and the Brisbane Ranges form (Figs. 5, 6a, b).
Observations made in the field and when reviewing herbarium specimens also suggest that most of the variants of the species are not well supported on morphological or ecological grounds, with instances of several variants recorded at a single site and some morphological intergrades between variants. For example, plants with the morphology of var. parvifolium and var. molle on Flinders Island co-exist at the Hines Road site (and group together based on cpDNA, with no discrimination between them; clade G4, Fig. 5) and plants morphologically matching the Rocky Site form were found side-by-side with var. grande in the Dandenong Ranges (Fig. 2a, d Table 1; and group together in clade G6 based on cpDNA). Several examples of intergrades were observed including examples of var. parvifolium from Flinders Island with a sparse indumentum on the upper surface of the stem leaves (and therefore grading with var. molle), and many examples of the Brisbane Ranges form have small, notched leaves, 1-2 mm longer than the described range for the variant (grading into typical S. parvifolium). Herbarium specimens reveal dozens of collections from throughout the range of S. parvifolium with an indumentum on the upper surface, to varying degrees of density. Hairs on the adaxial surface of the stem leaves appear to be a labile character that randomly occurs in S. parvifolium but might be more common on Flinders Island (hence var. molle is recognised there at the moment).
Given our molecular results, and with consideration to observations made in the field and from herbarium specimens, there is little available evidence to support taxonomic recognition of any of the variants of the species. We therefore recommend S. parvifolium be recognised as a single, morphologically variable and widespread species with no recognition of infraspecific taxa. Our recommendation may impact the Tasmanian legislation (Threatened Species Protection Act 1995), as the varieties currently recognised in Tasmania are both listed as Rare (Threatened Species Section, 2016a, b). A conservation assessment of S. parvifolium as a single taxon may result in the species exceeding the criteria for Rare in Tasmania, and it may therefore require consideration at a different conservation listing level or delisting.
Considering the incongruence of cpDNA and nrDNA phylogenies (Figs. 4a, 5) and the slightly different topologies of the ML and BI trees based on nrDNA (Fig. 4a, b), we do not recommend the merger of the associated species S. cinereum, S. daltonii or S. obcordatum with S. parvifolium. The relationships of all these taxa require further investigation, as our study focused mainly on S. parvifolium. In addition, though the divergence of the Kaiserstuhl and Mt Remarkable was not supported in the nrDNA phylogeny, the morphology of these variants of S. parvifolium should be evaluated. This could be extended to SA variants of the species more generally, given the Scott Creek samples were also unsupported but somewhat divergent in the nrDNA trees and supported and divergent in the cpDNA phylogeny.

Information on Electronic Supplementary Material
Online Resource 1. Bayesian inference (BI) 50% majority rule consensus tree based on analysis of nuclear ribosomal DNA (nrDNA) with Spyridium cinereum samples removed. Clades not discussed have been collapsed. Online Resource 2. Maximum likelihood (ML) consensus tree based on analysis of nuclear ribosomal DNA (nrDNA) with Spyridium cinereum samples removed. Clades not discussed have been collapsed. Online Resource 3. Bayesian inference (BI) 50% majority rule consensus tree based on analysis of nuclear ribosomal DNA (nrDNA) with Spyridium obcordatum samples removed. Clades not discussed have been collapsed. Online Resource 4. Maximum likelihood (ML) consensus tree based on analysis of nuclear ribosomal DNA (nrDNA) with Spyridium obcordatum samples removed. Clades not discussed have been collapsed.