Divergent lineages in a semi‐arid mallee species, Eucalyptus behriana, correspond to a major geographic break in southeastern Australia

Abstract Aim To infer relationships between populations of the semi‐arid, mallee eucalypt, Eucalyptus behriana, to build hypotheses regarding evolution of major disjunctions in the species' distribution and to expand understanding of the biogeographical history of southeastern Australia. Location Southeastern Australia. Taxon Eucalyptus behriana (Myrtaceae, Angiospermae). Methods We developed a large dataset of anonymous genomic loci for 97 samples from 11 populations of E. behriana using double digest restriction site‐associated DNA sequencing (ddRAD‐seq), to determine genetic relationships between the populations. These relationships, along with species distribution models, were used to construct hypotheses regarding environmental processes that have driven fragmentation of the species’ distribution. Results Greatest genetic divergence was between populations on either side of the Lower Murray Basin. Populations west of the Basin showed greater genetic divergence between one another than the eastern populations. The most genetically distinct population in the east (Long Forest) was separated from others by the Great Dividing Range. A close relationship was found between the outlying northernmost population (near West Wyalong) and those in the Victorian Goldfields despite a large disjunction between them. Conclusions Patterns of genetic variation are consistent with a history of vicariant differentiation of disjunct populations. We infer that an early disjunction to develop in the species distribution was that across the Lower Murray Basin, an important biogeographical barrier separating many dry sclerophyll plant taxa in southeastern Australia. Additionally, our results suggest that the western populations fragmented earlier than the eastern ones. Fragmentation, both west and east of the Murray Basin, is likely tied to climatic changes associated with glacial‐interglacial cycles although it remains possible that major geological events including uplift of the Mount Lofty Ranges and basalt flows in the Newer Volcanics Province also played a role.


| INTRODUC TI ON
Across the Australian continent, significant biogeographical barriers have been identified that have played a role in shaping the diversity of its unique biota (Bryant & Krosch, 2016;Schodde & Mason, 1999).
These barriers range in age and porosity, leading them to influence patterns of biodiversity at many taxonomic levels; however, in this paper we focus on the infra-specific level. In eastern Australia, phylogeographic studies have primarily been conducted on taxa of wet forests and the influence of barriers associated with intervening drier vegetation (see Bryant & Krosch, 2016), and on taxa disjunct between the Australian mainland and Tasmania (e.g., Freeman et al., 2001;Worth et al., 2017). There have been few phylogeographic studies of species from the drier vegetation that covers most of southeastern Australia, and of barriers affecting the distribution of such vegetation (e.g., French et al., 2016;Larcombe et al., 2011).
A prominent feature in the landscape of southeastern Australia is the Murray Basin. The lower portion of the basin encompasses the boundary between the states of South Australia, Victoria, and the southwest corner of New South Wales (Figure 1). It differs both climatically and edaphically from areas to its north-west and southeast (Bowler et al., 2006), and has been identified as a major biogeographic barrier often referred to as the Murravian Barrier (Joseph & Omland, 2009;Schodde & Mason, 1999). The Lower Murray Basin has a complex environmental history since the late Cenozoic, with marine inundation that peaked before 7 Mya, followed by gradual marine regression (McLaren et al., 2011). Tectonic uplift ~2.4 Mya dammed the Murray River, creating a mega lake, Lake Bungunnia, covering an estimated 68,000 km 2 of the Basin (Zhisheng et al., 1986). This starved the coastline of sediments and led to development of the Kanawinka escarpment, now forming a divide between a siliceous formation (Loxton-Parilla Sands Strandlines) on the inland side and the calcareous Gambier Coastal Plain on the seaward side (McLaren et al., 2011). Lake Bungunnia drained within the last 700 kyr giving rise to the current channel of the Murray River and a return to depositional coastal environments (Bowler et al., 2006;McLaren et al., 2011). These processes, plus later alluvial and aeolian sediment movements, produced deep sedimentary soils and dunefields. Ongoing aridification over the last 3 Myr, overlayed with ~100 kyr glacial cycles, also affected both temperature and rainfall in the region (Mills et al., 2013).
Multiple species show a geographic disjunction across the Lower Murray Basin, but few phylogeographic studies have assessed patterns of genetic structuring associated with this disjunction. Examples of plant phylogeographic studies include those of Hardenbergia violacea (Schneev.) Stearn (Larcombe et al., 2011), Eucalyptus globulus , and the genus Correa (French et al., 2016). All of these highlight a deep genetic divergence associated with the basin, which could be associated with history of vicariant separation of populations on either side, although in Correa subsequent reconnection of populations across the Murray Basin was also inferred. Genetic studies of additional taxa with disjunct distributions across the Lower Murray Basin could assess whether common genetic patterns can be identified in the history of the vegetation of this region, associated with either vicariance, where disjunct species predate formation of the barrier, or dispersal subsequent to formation of the barrier.
Eucalyptus behriana F. Muell., a mallee (lignotuberous, multistemmed shrub) in Eucalyptus section Adnataria (Boxes and Ironbarks), is a species that occurs both to the west and in the eastern half of the Murray Basin ( Figure 1) (Brooker et al., 2015). It occurs mainly as a member of tall mallee woodlands in areas of 350-600 mm annual rainfall with shallow soils (Benson, 2008;Myers et al., 1986;Nicolle, 2014). The distribution of the species shows several other large disjunctions, making its biogeographical history potentially informative regarding the broader environmental evolution of southeastern Australia, and a good test case for whether there are recurring patterns of genetic relationships in dry sclerophyll plant taxa across the Murray Basin. Regionalisation for Australia version 7 (IBRA7), Australian Government, 2017). Within these regions, E. behriana mainly occurs in depressions with heavier soils and higher water holding capacity than surrounding sand plains (Nicolle, 2014;White, 2006), and it is rare in the adjoining Lowan Mallee bioregion dominated by younger aeolian dunes (Conn, 1993). Further east, sporadic populations occur on Ordovician and Devonian sandstone outcrops in the Victorian Goldfields (Raymond et al., 2012). Other isolated populations occur at Long Forest, the only area of mallee vegetation south of the Great Dividing Range (GDR), and around the town of West Wyalong it remains possible that major geological events including uplift of the Mount Lofty Ranges and basalt flows in the Newer Volcanics Province also played a role.

K E Y W O R D S
climatic cycles, ddRAD-seq, Eucalyptus, phylogeography, population fragmentation, species distribution model, vicariance several hundred kilometers north in the South-West Slopes region of New South Wales (Benson, 2008). At Long Forest, the species occurs on solodic soils developed on an outcrop of Ordovician sediments, largely surrounded by richer soils on significantly younger basaltic volcanic flows (Myers et al., 1986).
Our aim in the current study was to assess patterns of genetic variation in E. behriana to build hypotheses regarding evolution of major disjunctions in the species' distribution and to expand understanding of the biogeographic history of southeastern Australia.
Specifically, we sought to assess the significance of the Lower Murray Basin as a barrier to gene flow, relative to the other geographic disjunctions in E. behriana, to contribute to understanding of the importance of this feature in the history of the flora. We also sought to assess whether genetic patterns showed any support for recent dispersal, over a null hypothesis of vicariance, in the history of the species. To achieve these aims, we analyzed anonymous nuclear genome loci generated through double digest restriction site-associated DNA sequencing (ddRAD-seq) and built species distribution models (SDMs).

| Sampling
Populations were sampled across the range of E. behriana ( Figure 2). We sampled 1-11 individuals from each sampling location, depending on population size (Table 1). Samples were collected ~100 m apart to minimize the chances of sampling siblings, as most eucalypt seeds fall within 20 m of parent trees (Booth, 2017). Leaves showing minimal signs of disease or damage were collected into coffee filters and silica desiccating beads, and at least one voucher herbarium specimen was taken per sampling location (Table 1).

| DNA extraction and library preparation
The modified CTAB protocol of Schuster et al. (2018)

| Read trimming, quality control, and data assembly
Reads were paired in Geneious v11.1.5 (https://www.genei ous. com, Kearse et al., 2012) and trimmed to remove restriction site residues and adaptor diversity bases using the Cutadapt 2.8 python script (Martin, 2011) with a maximum error rate of 0.5. Paired and trimmed reads from the 97 samples successfully sequenced were put through the ipyrad pipeline (v.0.9.33) (Eaton & Overcast, 2020) to assemble loci (see Table S1.1 available on Dryad at https:// doi.org/10.5061/dryad.v9s4m w6sm for parameters used). Data were treated as paired-end ddRAD data and assembled using the reference option, mapping to the eleven chromosomes of the Eucalyptus grandis genome (Myburg et al., 2014). We targeted only loci from the nuclear genome to avoid any potential confounding of results due to the well-established cytonuclear discrepancy in the eucalypts (Alwadani et al., 2019;Nevill, Després, et al., 2014).
Multiple values for several ipyrad parameters were tested to find the most informative dataset possible. The analysis using a 0.85 clustering threshold, 6 read minimal coverage, and a minimum of 75% of samples represented at each locus was identified as the optimized dataset, as it contained as much data as possible while still allowing for confidence in the base calls and analyses not being computationally limited.

| Species distribution modeling
Maxent (Phillips et al., 2006)  Grid of Australia (Viscarra Rossel et al., 2015), measured at a depth of 30-60 cm where applicable: clay percent content, sand percent content, soil depth, soil water holding capacity and soil pH. As historic soil data are not available, we built two models, one limited to climatic variables which was projected on glacial maximum (~20 kya) climate variables from WorldClim 1.4 (Hijmans et al., 2005), and one for current conditions only that included climatic and soil variables.
The MaxentVariableSelection package for R (Jueterbock et al., 2016) was used to incrementally reduce the number of variables while testing regularization factors between 0.5 and 15 in 0.5 increments, and the Akaike information criterion was used to choose optimal models. For the climate only model, the optimized model included four predictors (annual mean temperature, mean temperature of wettest quarter, annual precipitation, precipitation of warmest quarter) and regularization parameter of 1.5; the soil and climate model included five predictors (annual mean temperature, mean temperature of wettest quarter, annual precipitation, soil clay content, soil pH) and a regularization parameter of 1.

| Species distribution models
The species distribution models (Figures 4 and 5) show a good fit to the current distribution of the species, although the soil and climate model ( Figure 5) was a tighter fit to the species distribution in the Murray Basin and inland slopes of the GDR, suggesting that soil factors indeed contribute to these disjunctions. When the climate only model was projected onto modeled last glacial maximum climatic conditions (Figure 4b), the areas of potentially suitable environment moved far to the north, with low suitability across the areas the species currently occupies. Overall, the highest population level divergences were observed between populations separated by the Lower Murray Basin (Table 2); divergence between the two largest western populations, separated by the Spencer Gulf, was also high.

| Eucalyptus behriana population variation
The observed support for isolation by distance in E. behriana

| Mechanisms and timing of population disjunctions
Mallee eucalypts, such as E. behriana, typically have long life spans, associated with a large lignotuber rich in dormant buds and energy reserves that provide substantial capacity to recover from fires and droughts (Noble, 2001). They also have low recruitment rates (Wellington & Noble, 1985) and limited capacity for seed dispersal (Booth, 2017

| East-west divergence across the Murray Basin
We found a substantial genetic divide between the populations to the east and west of the Murray Basin ( Figure 3); however, without an outgroup, we are unable to determine with certainty whether this reflects monophyletic groups, or one clade nested within the other.
If the latter is true, based upon the level of genetic diversity, we suggest there is a stronger case for the eastern clade being nested within the west than the reverse. The large differentiation between the populations in the west suggests that fragmentation occurred in these western populations earlier and more completely than in the currently larger eastern populations. These results are largely congruent with patterns in Hardenbergia violacea (Larcombe et al., 2011) and Correa spp. (French et al., 2016), adding weight to the hypothesis that this deep east-west split in the Murray Basin may be a recurring F I G U R E 5 Optimized Maxent model predictions of environmental suitability under current climatic and edaphic conditions. This optimized model used a regularization factor of 1 and five predictor variables: mean annual temperature, mean temperature of wettest quarter, mean annual precipitation, soil clay content (30-60 cm below soil surface), and soil pH (30-60 cm below soil surface). Point records include both those downloaded from the Atlas of Living Australia and collections used in this study as per Figure 1 pattern among dry sclerophyll plant taxa, reflecting the historical environmental change in the region.

| Diversity of western populations
Our findings suggest that the western populations may not have been the small and geographically restricted populations they are today for an extended period, as genetic diversity in these populations would have been reduced by genetic drift and bottlenecking if this were the case (Amos & Harwood, 1998). Indeed, we see the opposite to this with the highest genetic diversity being observed in these currently small, restricted western populations. While the species was likely not significantly more widespread prior to European landscape (Neagle, 2008). The Eyre Peninsula population has also been affected by land clearing but covers a narrower geographic area and E. behriana is not the dominant species where it occurs, rather occurring only along drainage lines, aside from a poorly drained depression west of Cummins where it is co-dominant with E. odorata (Smith, 1963). There is no information on the history of the Flinders While the term refugia is most commonly used to refer to areas of constant occupancy by a taxon through glacial periods, it stands to reason that those species that replace the retreating taxa where glaciation does not occur may retreat to refugial areas under interglacial conditions (Bennett & Provan, 2008). This is particularly likely in the case of Australia as there was not large-scale formation of icesheets across the landscape, rather widespread cooling and drying of the climate (Barrows et al., 2001). Here we put forward the hypothesis that the western populations may have been initially fragmented by a climatic shift associated with glaciation cycles with the higher rainfall of interglacial conditions allowing taller eucalypt species to replace E. behriana in areas between the current populations. The associated sea-level rise also likely played a role, with the Spencer Gulf that sits between the Eyre Peninsula, and Barossa and Light River valleys populations being above sea-level during glacial climates, thus allowing population connectivity across the region, a hypothesis supported in our SDM which show high environmental suitability under glacial climates in the Spencer Gulf. A second hypothesis regarding this fragmentation is that it has been a gradual process associated with the long-term climatic trends that operate on time spans longer than recent glacial cycles. The Australian climate has been gradually drying over the last ~10 MY, although a wetter peak is known from ~5-3 Mya  continues to the present (Tokarev, 2005), and therefore changes in topography may be one of the factors that contributed to establishing the disjunctions in the taxon's western populations.

| History of the West Wyalong population
The low level of genetic distance between the Goldfields and West Wyalong populations suggests that these populations diverged comparatively recently relative to other eastern populations (Figure 3

| History of Long Forest population
We found only weak evidence for the relationship between the Long Forest population and other eastern populations, although the NeighbourNet network ( Figure 3)  open, grassy ecosystems (Barlow & Ross, 2001) and the basalt flows partially surround Long Forest aside from the north where the GDR represents exposed parts of the Castlemaine Group. It is possible that it was these volcanic events that isolated the Long Forest population, however, the most recent flows (<0.1 Mya) are to the south of the Long Forest area, with many of the flows that surround Long Forest being closer to 2 Myr old (Heath et al., 2020).
This date would seem too old to be the initial cause of the isolation of the Long Forest population given the lack of morphological and genetic differentiation of this population, and the fact that such an old isolation would also infer that the deeper genetic disjunction across the Lower Murray Basin was on the order of several million years. The only possible route of population connection that does not cross basalt flows is across the GDR to the north of Long Forest, which is potentially congruent with a climate driven vicariance event as evidence suggests a less wooded environment on the ranges during the drier, colder glacial climates (Hope, 1994). If E. behriana is primarily responding to rainfall and not temperature, it is possible that the species occurred on the GDR during glacial climates and has been outcompeted by taller communities moving upslope in response to increasing rainfall and temperature during an interglacial period. When considering the balance of the evidence from the genetic analyses, SDMs and the known historical environmental change in the region, we favor a hypothesis of climate driven isolation, likely via either the Kilmore Gap or directly across the GDR.

ACK N OWLED G M ENTS
We thank Pauline Ladiges for comments on the manuscript and

CO N FLI C T O F I NTE R E S T
The authors have no conflicts of interest to declare.