Genomic Diversity Guides Conservation Strategies Among Rare Terrestrial Orchid Species When Taxonomy Remains Uncertain

Annals of Botany Page 1 of 11 doi:10.1093/aob/mcx022, available online at https://academic.oup.com/aob

Genomic diversity guides conservation strategies among rare terrestrial orchid species when taxonomy remains uncertain

Collin W. Ahrens1,2,*, Megan A. Supple3, Nicola C. Aitken3, David J. Cantrill1, Justin O. Borevitz3 and Elizabeth A. James1 1Royal Botanic Gardens Victoria, Science Division, Melbourne, Victoria 3004, Australia, 2Hawkesbury Institute for the Environment, Western Sydney University, Locked Bag 1797, Penrith, NSW 2751, Australia and 3Australian National University, Research School of Biology, Centre of Excellence in Plant Energy Biology, Canberra, ACT 0200, Australia *For correspondence. E-mail [email protected]

Received: 21 January 2017 Editorial decision: 30 January 2017 Accepted: 12 February 2017

Background and Aims Species are often used as the unit for conservation, but may not be suitable for species complexes where taxa are difficult to distinguish. Under such circumstances, it may be more appropriate to consider species groups or populations as evolutionarily significant units (ESUs). A population genomic approach was em- ployed to investigate the diversity within and among closely related species to create a more robust, lineage-specific conservation strategy for a nationally endangered terrestrial orchid and its relatives from south-eastern Australia. Methods Four putative species were sampled from a total of 16 populations in the Victorian Volcanic Plain (VVP) bioregion and one population of a sub-alpine outgroup in south-eastern Australia. Morphological measurements were taken in situ along with leaf material for genotyping by sequencing (GBS) and microsatellite analyses. Key Results Species could not be differentiated using morphological measurements. Microsatellite and GBS markers confirmed the outgroup as distinct, but only GBS markers provided resolution of population genetic structure. The nationally endangered Diuris basaltica was indistinguishable from two related species (D. chryseopsis and D. behrii), while the state-protected D. gregaria showed genomic differentiation. Conclusions Genomic diversity identified among the four Diuris species suggests that conservation of this taxonomically complex group will be best served by considering them as one ESU rather than separately aligned with species as currently recognized. This approach will maximize evolutionary potential among all species during increased isolation and environmental change. The methods used here can be applied generally to conserve evolutionary processes for groups where taxonomic uncertainty hinders the use of species as conservation units.

Key words: Diuris, genotyping by sequencing (GBS), microsatellite markers, next-generation sequencing, Orchidaceae, population genomics.

INTRODUCTION Peterson et al., 2012]. Having thousands of markers gives re- searchers and practitioners the power to uncover fine-scale Conservation of species relies on evidence-based, accurate and structure of a conservation unit regardless of current taxonomic meaningful data collected with consistency and scientific rigour classification. (Tear et al.,2005; Frankham et al., 2012). Conservation strate- Defining species boundaries can be a difficult task (Fazekas gies can be improved greatly with knowledge of genetic varia- et al.,2009; Hausdorf, 2011). Taxonomic uncertainty can lead tion identified through population genetic methodologies (e.g. to negative outcomes for conservation strategies (Frankham Xiao et al.,2006; Martins et al., 2015; Zhang et al.,2015; et al., 2012). Indeed, erroneous classifications can lead to po- Ahrens and James, 2016; Turchetto et al.,2016). Indeed, popu- tential financial, legal, biological and conservation repercus- lation genetics has recently been used as the basis for the devel- sions (Hey et al.,2003). Biological reality suggests that a opment of a conservation prioritization framework for rare species can be difficult to confirm when morphology is influ- plants (Ottewell et al.,2016). The utility of understanding the enced by phenotypic plasticity, hybridization or intraspecific genetic variation within a species can be crucial for prioritizing ploidy variability, among other characteristics (Fazekas et al., certain populations over others (Ahrens and James, 2015; 2009; Fitzpatrick et al., 2015). If maintaining evolutionary po- Dillon et al., 2015). Particularly with the advent of newer geno- tential is a part of the final conservation management goal, cate- mic technologies over the past 5 years, we are able to clarify gorization that is not constrained by species boundaries and and resolve patterns of population structure that were previ- includes genetic diversity metrics may lead to improved conser- ously unattainable. Reduced-representation molecular tech- vation outcomes. In such cases, next-generation sequencing niques are particularly useful in that they allow low-cost techniques have the potential to differentiate conservation units genome typing of hundreds or thousands of individuals [e.g. by identifying genetic lineages that are part of dynamic evolu- genotyping by sequencing (GBS), Elshire et al., 2011; double tionary processes (i.e. contemporary gene flow, mating dynam- digest restriction site-associated DNA sequencing (ddRADseq) ics and environmental resilience; Allendorf et al., 2010;

VC The Author 2017. Published by Oxford University Press on behalf of the Annals of Botany Company. All rights reserved. For Permissions, please email: [email protected] Page 2 of 11 Ahrens et al.—Diuris population genomics

Funk et al., 2012; Hoffmann et al.,2015). The use of evolution- GBS and microsatellite markers to test our hypothesis by ex- arily significant units (ESUs) provides an alternative categori- ploring genetic divergence and interspecific population struc- zation of biological entities (e.g. ecotypes, subspecies or ture at fine scales. The genetic results were compared with populations) and may be suitable as conservation units for man- morphological characteristics to guide the identification of ap- aging closely related species. ESUs represent distinct lineages propriate conservation targets, and we suggest new strategies and take into account genetic interactions among species that for conservation and recovery of this Diuris species complex. influence gene pools but are not constrained by species delimitations (Moritz, 1994). Classifying populations that have sub- stantial reproductive isolation is important for maximizing their MATERIALS AND METHODS potential for adapting to future environmental change (Funk Study species et al., 2012). The use of ESUs instead of species classifications has the power to improve conservation strategies by incorporat- Four co-occurring, closely related terrestrial orchid species on ing genetic diversity in addition to biological and morphologi- the Victorian Volcanic Plain (VVP; Fig. 1)wereusedtoexam- cal information. As currently constituted, existing international ine the relationship between current species delineations, mor- conservation policy tends to overlook genetic diversity as an phological differentiation and genetic divergence. The VVP important component of the species (Laikre, 2010). (Fig. 1) is a basalt-derived bioregion in south-eastern Australia Consider the family Orchidaceae, which exhibits highly di- that contains <1 % of its original vegetation due to fragmenta- verse floral morphology and is under pervasive threat from tion and urban development (Williams et al.,2015). Diuris global change (Swarts and Dixon, 2009). This floral diversity is basaltica is a nationally endangered terrestrial orchid that oc- the result of unprecedented genetic diversity that originated curs in only three known locations on the urban outskirts of the early in orchid evolution (Mondragon-Palomino and Theissen, greater Melbourne district. A fourth historical population is 2009). The breadth of floral variation within genera may create now extinct. At the time of sampling, two locations contained the illusion of greater species diversity than is actually present, only 2–5 individuals. The third site supposedly had hundreds of and has led to a history of taxonomic exaggeration (Pillon and individuals, but none flowered during the 2 years of the study. Chase, 2007). It has been found that taxonomically difficult However, an ex situ population, maintained at the Royal groups tend to have more endangered species within them Botanic Gardens Victoria glasshouses, contained several hun- (Pilgrim et al.,2004). Further, species delimitations in the dred individuals cultivated for conservation from plants sourced Orchidaceae remain tenuous in some groups (Flanagan et al., originally in the late 1990s and early 2000s from both the third 2006; Devey et al., 2007; Pillon and Chase, 2007; Bateman population and the fourth now-extinct population. Of the re- et al., 2010; Hedre´n et al.,2012), suggesting that using an ap- maining species, D. gregaria is considered endangered in the proach other than species boundaries may sometimes be war- state of Victoria and is mostly confined there, with some occur- ranted to guide conservation. Under circumstances where rences in South Australia (Fig. 1). It exhibits a clumping habit orchid species are hard to define, understanding patterns of ge- not found in the other species. Diuris chryseopsis and D. behrii netic diversity has been successful for developing conservation have large distributions that extend beyond Victoria (Fig. 1). strategies (Swarts et al., 2009; Chung et al., 2012; Chen et al., These two species exhibit morphological floral and leaf charac- 2013; Qian et al., 2014; Yang et al.,2014; Pandey et al.,2015; ters that tend to be larger in size than the preceding two species. Pedersen et al., 2016; Ross and Travers, 2016). In many cases, Neither one is considered to be threatened or endangered in the genetic differentiation (FST) within rare orchid species remains state of Victoria or nationally in Australia. This Diuris group very low (0092) compared with more common species (0279), mimics nectar-bearing flowers of other plants in the area and suggesting that there is generally high gene flow and little pop- lures the same generalist pollinators via simple food deception ulation structure in rare orchids (Phillips et al.,2012) but no (Backhouse and Lester, 2010). Species descriptions focus only studies were found that recorded FST among closely related on morphological characters and geographic locations as diag- orchid species. Despite the power of reduced-representation nostic features (Jones, 1998, 2006). Diuris monticola was used next-generation sequencing protocols to resolve fine-scale pop- as an outgroup because it has similar floral morphology to the ulation structure, there have been few, if any, studies that use other four species and occurs outside of the VVP in an Alpine these new techniques to guide the conservation of orchids with region, but within the distribution of D. chryseopsis and D. beh- ‘fuzzy’ taxonomic margins. By using these new techniques, we rii (Fig. 1). can instead focus on protecting evolutionary processes in orchid Sample collection took place in Spring of 2012 and 2013. A populations, which shifts the conservation question from ‘which total of 298 individuals were collected from 17 populations species should be conserved?’ to ‘which populations are evolu- across Victoria (Table 1; Fig. 1). Six other populations, re- tionarily important?’ corded within the past 10 years, were visited but no flowering We used five putative orchid species from south-eastern individuals were present during the two field seasons. At least Australia to test the reliability of using current species delinea- 20 samples were collected from each population, except where tions, made with morphological characteristics, as the units of populations were too small [4–12 individuals (Table 1)]. One conservation. We hypothesized that morphologically diagnosed population (PAR) exhibited high morphological variation so species boundaries in the study group within Diuris sampling was increased to cover the variation observed. At (Orchidaceae) do not reflect the evolutionary lineages identified least one representative voucher specimen (3–4 in most cases) by population genomics. Our goal was to look beyond taxo- was taken from each population for post-hoc identification at nomic classifications and investigate genomic diversity to cre- the Royal Botanic Gardens Victoria by botanical experts and ate a robust conservation strategy. We used morphometrics, lodged at the National Herbarium of Victoria. While we used Ahrens et al. — Diuris population genomics Page 3 of 11

Boundaries and distributions Political boundaries Collected Diuris species N Populations D. basaltica distribution 04590 180 Diuris behrii D. gregaria distribution Diuris basaltica D. chryseopsis distribution km Diuris chryseopsis D. behrii distribution Diuris gregaria Victorian volcanic plain bioregion Diuris monticola

FIG.1.Diuris populations of four species sampled in the Victorian Volcanic Plain (VVP) and an outgroup sampled from the alpine region. Historical distributions for the four VVP species were generated from the Atlas of Living Australia (ala.org.au). post-hoc identifications as putative population classifications, features of this Diuris group were based on floral morphology, the presence of more than one ‘species’ was possible. However, eight of the measurements were taken from the flower. These if two or more species were present within a population, we included the number of flowers per flowering stalk, width (W) would expect morphological and/or genomic data to distinguish and length (L) of the dorsal sepal, W and L of the lateral petal, them. Of the 17 populations sampled, two were identified as D. W and L of the labellum, and flower diameter. The non-floral basaltica,sixasD. gregaria,fiveasD. chryseopsis,threeasD. measurements included the W and L of the leaves and height of behrii and one as D. monticola (Table 1). Only individuals the plant. Length and width ratios (L/W) for all traits where W whose flower morphology could be measured were sampled. and L were measured were used for morphological analyses. Morphology was measured and recorded in situ,withtheex- Using the ratio of some traits allowed us to account for petal ception of the glasshouse- (GH) grown D. basaltica population. shape (e.g. short and wide or long and thin) and variation in Only one D. basaltica population (eight plants) was measured size due to growing environment, as some individuals were in situ. The remaining 24 D. basaltica plants included in the measured while growing in the glasshouse. Size for other traits analysis were measured in the ex situ GH population. Leaf ma- (height and flower width) informed us if size was a determinis- terial was collected and stored for DNA extraction. tic character. A principal co-ordinates analysis (PCoA) was performed on the morphological data set using the Vegan package (Oksanen et al.,2013) and the ‘dist’ function in R v3.1.0 (R Core Team, 2014) to calculate the distance matrix for the Morphology PCoA with the ‘dudi.pco’ function. In addition, we analysed For each individual sampled, we measured 11 vegetative and the 11 measurements as a principal components analysis (PCA) floral characteristics considered diagnostic for the species and using the ‘prcomp’ function in R and a PCoA using the compared them with species descriptions. As diagnostic ‘dudi.pco’ function in R. Page 4 of 11 Ahrens et al.—Diuris population genomics

TABLE 1. Population identity, assigned species category and estimates of population diversity estimates from microsatellite markers are given for all populations and number of samples, reads and reads per sample for the genotyping by sequencing (GBS) analysis

Population Putative species Microsatellite markers GBS

NNe Ho He FIS Nþreps N reads Reads/Ind GH D. basaltica 24 286 045 053 014 23 5 770 048 250 8717 SK D. gregaria 4211 041 047 012 4 1 372 623 343 1558 ML D. gregaria 12 292 044 053 017 12 3 543 504 295 2920 CG D. gregaria 11 237 030 044 031 11 2 442 792 222 0720 Woo D. gregaria 10 232 031 044 029 9 2 898 598 322 0664 Par D. behrii 36 383 037 057 034 44 14 427 416 327 8958 YP D. gregaria 8239 031 051 039 5 1 054 672 210 9344 CW D. chryseopsis 20 253 042 053 022 19 4 530 488 238 4467 IV D. chryseopsis 20 259 034 053 036 25 7 087 302 283 4921 DK D. chryseopsis 20 201 041 038 008 27 8 244 430 305 3493 WG D. basaltica 7185 034 033 003 6 4 303 107 717 1845 RC D. chryseopsis 20 277 044 050 012 27 16 276 802 602 8445 PF D. chryseopsis 20 241 032 049 035 28 14 751 125 526 8259 CR D. behrii 20 246 039 056 030 15 7 661 890 510 7927 CM D. behrii 24 317 032 053 040 24 18 737 356 780 7232 AC D. gregaria 20 264 037 042 013 23 16 184 928 703 6925 ALP D. monticola 25 15 11 246 681 749 7787 Overall 298 228 037 048 024 317 140 533 762 443 3242

N, number of individuals; Ne, number of effective alleles; Ho, observed heterozygosity; He, expected heterozygosity; Fis, inbreeding coefﬁcient, Nþreps, number of individuals þ the number of replicates that made it through the GBS pipeline; N reads, number of reads for all individuals; Reads/Ind, average number of reads per individual.

Microsatellite markers amplified and normalized individual libraries before pooling Genomic DNA was isolated for both genetic protocols from and sequencing, which reduced the range in per sample cover- 20 mg of silica-dried leaf tissue from each individual using the age (Nicotra et al.,2016). Two pooled GBS libraries of 154 DNeasy Plant Mini Kit (QIAGEN, Hilden, Germany) by the and 178 multiplexed samples, respectively, were sequenced on Australian Genome Research Facility (AGRF; Adelaide, South an Illumina HiSeq 2500 (Illumina, San Diego, CA, USA) using Australia). The 13 microsatellite markers used in this study 100 bp paired-end reads. Technical replicates (n ¼ 34)ofthe were developed previously for this Diuris group (Ahrens and first lane were added to the second lane. James, 2014). Loci were amplified in three multiplex PCRs us- AXE (v0.2.6, https://github.com/kdmurray91/axe) was used ing a three-primer approach (Blacket et al.,2012). PCRs were to demultiplex the reads. For each demultiplexed sample, a cus- carried out under the following conditions: 5 min denaturation tom script was used to remove adaptor contamination due to at 95 C, followed by 32 cycles of 95 C for 30 s, annealing at fragment sizes <100 bp by aligning each read to its pair and re- 60 C for 1 min 30 s and 72 C for 30s, with a final extension at moving read through if present. Overlapping reads were merged 60 C for 30 min. Fragment analysis was performed via capil- with FLASh v1.2.11 (Magoc and Salzberg, 2011), with maximum overlap 70. Reads were filtered by quality, with 50 trim- lary electrophoresis on an ABI 3130l (Applied Biosystems, ¼ Foster City, CA, USA) by Macrogen (Seoul, South Korea). ming turned off, and trimmed to 64 bp with sickle se (v1.33, Alleles were visualized and scored in Geneious Software ver- https://github.com/najoshi/sickle). For compatibility, pseudo sion 7.0 (Biomatters Ltd, Auckland, New Zealand). barcodes were added to each read and those reads were im- Allele size data were imported into the Polysat (Clark and ported into TASSEL v3.0.172 (Glaubitz et al.,2014). Jasieniuk, 2011) package in R for PCoA on Bruvo’s genetic Candidate single nucleotide polymorphisms (SNPs) were iden- distance. This distance metric was developed to measure ge- tified in TASSEL using the UNEAK pipeline with default set- netic distance similar to dominant data while taking into ac- tings, including a minimum coverage per tag of 5, error count mutational distances between alleles. Observed tolerance of 003, maximum frequency of 05andaminimum heterozygosity (H ), expected heterozygosity (H ) and effective frequency of 005 for minor alleles. The output genotype matrix o e was further filtered in R using custom script (https://github. number of alleles (Ne) were calculated on allele size data in Genodive (Meirmans and Van Tienderen, 2012). Inbreeding com/borevitzlab/GBSFilteR/blob/master/Diuris_hmn_cleaning Fig.R) first by removing failed samples with <4000 SNPs (FIS) was calculated in Genodive using Weir and Cockerham’s (1984) methodology for f. called (273 unique samples remained) and secondly by removing SNP loci with <200 (‘stringent’) or 70 (‘relaxed’) samples called. Potential paralogous loci with >45 samples being called as ‘het’ were also removed. Technical replicates were used to Genotyping by sequencing confirm reproducibility. Only one sample of each replicate was The GBS library was prepared following a method adapted retained for downstream population analyses. The ‘relaxed’ from Elshire et al. (2011) using 300 ng of genomic DNA per in- data set was used whenever a genetic distance matrix was cal- dividual and the HpaII restriction enzyme. We barcoded, culated, as genetic distances can be calculated with good results Ahrens et al. — Diuris population genomics Page 5 of 11 for genomic data with up to 90 % missing data (Fu, 2014). The separation from other species. The x-axis explained 9495 % of ‘stringent’ data set was used for analyses with complex model- the variation, and the y-axis explained 483 % of the variation. ling components (i.e. STRUCTURE and SNAPP) where large The PCA and PCoA with all 11 measurements show the same amounts of missing data can give spurious results or make pattern as the PCoA with ratios (Supplementary Data Figs S2 them difficult to interpret. and S3). The ellipses in the PCA indicated that all species had Both the ‘relaxed’ and ‘stringent’ data sets were used for ge- overlapping morphological characteristics, but D. basaltica, D. netic distance calculations. The ‘dist’ function in R was used to gregaria and D. monticola tended to be slightly smaller. calculate the genetic distance (Euclidean) of both data sets with and without the outgroup, D. monticola. PCoAs were created for the ‘stringent’ and ‘relaxed’ data sets with adegenet v1.4-2 Microsatellite analysis (Jombart, 2008). An unrooted dendrogram was created using all Genotypes based on 13 microsatellite markers were obtained populations. A second unrooted dendrogram was built using for 270 samples. None of the markers amplified in the D. mon- paired F values between populations calculated in adegenet. ST ticola samples. Some individuals had more than two alleles per We used the coalescent model in SNAPP, an add-on to BEAST locus in three of the 13 loci, which could be due to locus dupli- v2.3 (Bryant et al., 2012; Bouckaert et al., 2014), to investigate cation or polyploidy. Putative polyploidy was locus specific genetic divergence. The data set used for SNAPP included six and not population or species specific. Global F was esti- individuals selected randomly per population from the ‘strin- ST mated at 003 within the VVP group (four species on the VVP), gent’ data set to increase computational speed. For the priors, indicating high gene flow and little genetic isolation. F fell we calculated the mutation rates based on the data set as de- IS between –008 and 040, with an overall F of 024, which in- scribed in the SNAPP manual (Bouckaert and Bryant, 2012). IS dicated that inbreeding was moderate, and two populations had We ran the MCMC (Markov chain Monte Carlo) algorithm for a slightly negative coefficient. The PCoA did not show signifi- 300 000 generations, recording every 1000 tree and log of the cant species or population differentiation (Fig. 2B). Three spe- output for visualization and convergence, respectively. Tree to- cies (D. basaltica, D. chryseopsis and D. behrii)formone pology was visualized in DensiTree. Branch lengths are relative overlapping cluster although the D. gregaria individuals because we only use polymorphic SNPs (Bouckaert and showed off-centre aggregation. The two-axis PCoA accounted Bryant, 2012; Bryant et al., 2012). The ‘stringent’ data set was for 727 % of the variation on axis 1 and 640 % on axis 2. used in STRUCTURE v2.3, a Bayesian clustering MCMC method (Pritchard et al.,2000) to determine the number of genetic clusters (K) in the data. A burn-in length of 100 000 itera- Genotyping by sequencing tions and a run length of 150 000 iterations were used in an admixture model with correlated allele frequencies among pop- In total, 141 million reads were distributed amongst 273 ulations for each K-value between 1 and 10. Each K-value was unique samples, 25 samples removed prior to analysis and 34 run 12 times, and ten runs with the highest lnP(D) scores were replicated samples for 332 total samples (Table 1). We identi- used for later consensus. The optimal K-value was determined fied 118 764 SNPs across the 332 samples which is within using the Evanno DKmethod(Evanno et al., 2005)in range for GBS studies (Poland and Rife, 2012; Grabowski STRUCTURE Harvester (Earl and vonHoldt, 2012). A consen- et al., 2014). Two replicates were removed due to low read sus of the optimum K-value was created in CLUMPP count. In a dendrogram of the genetic distance matrix generated (Jakobsson and Rosenberg, 2007), and the graphical parameters with the ‘relaxed’ data set (Supplementary Data Fig. S4), all 32 were drawn in the program DISTRUCT (Rosenberg, 2004). replicates were closest to their replicate counterpart. We re- To understand the relationship between morphological and moved each replicate counterpart for all of the following analy- genetic differentiation, the natural log of the genetic distance ses. From the 273 unique genotypes, we isolated 9793 SNPs for matrix computed from the ‘relaxed’ data set was plotted in R the ‘relaxed’ data set and 874 SNPs for the ‘stringent’ data set. against the natural log of the morphological pairwise distance The PCoA with the ‘relaxed’ data set showed similar patterns matrix. A generalized linear model was fit to test the relation- to the PCoA calculated from the ‘stringent’ data set (Fig. 2C, ship between morphological distance and genetic distance. D). The two-dimensional PCoA from the ‘stringent’ data set explained 644 % of the variation (Fig. 2C), whereas the PCoA from the ‘relaxed’ data set explained 754 % of the variation RESULTS (Fig. 2D). The PCoA from the ‘relaxed’ data set showed greater separation between groups, especially between the D. gregaria Morphology and the main cluster of D. chryseopsis, D. behrii and D. basal- During field collections, morphological variation was apparent tica. In both GBS PCoAs, one D. behrii population (DK) sepa- within populations (Supplementary Data Fig. S1). Most charac- rated from the main cluster. This was due to low levels of ters measured in situ were in agreement with those given in population structure with a small amount of genetic differentia- published descriptions, except for D. basaltica, which were tion (y-axes explain 246 and 338 %, respectively). consistently different (Fig. S1). We controlled for size differ- The unrooted dendrogram showed the genetic relationship ences between glasshouse-grown and in situ plants as described among individuals (Fig. 3). Individuals of the D. monticola in the Materials and Methods. The PCoA showed that there was population were the most distinct group. However, there were no clear distinction between VVP species, but D. gregaria and less obvious clusters present. A cluster consisting of mostly D. behrii did show a small degree of separation (Fig. 2A). The D. gregaria individuals with one D. behrii individual and one D. monticola samples grouped together but showed no clear D. chryseopsis individual was isolated from the other three Page 6 of 11 Ahrens et al.—Diuris population genomics

AB40

20 1

0 0

–20

Axis 2 (4·83 %) Axis 2 (6·40 %) –1

–40 D. basaltica D. gregaria D. behrii –60 –2 D. chryseopsis D. monticola

–300 –200 –100 0 100 200 –1 01 Axis 1 (94·95 %) Axis 1 (7·27 %) CD

1·0 10

0·5 0 0

–10 –0·5

Axis 2 (2·46 %) –1·0 Axis 2 (3·38 %) –20

–1·5 –30

–2·0 –40 –1 0 12 –10 02010 Axis 1 (3·98 %) Axis 1 (4·16 %)

FIG. 2. Principal co-ordinates analyses for (A) morphology, (B) microsatellites, (C) genotyping by sequencing (GBS) ‘stringent’ data set and (D) GBS ‘relaxed’ data set. Colours represent putative species, and the shapes within colour groups represent different populations. Inset in (D) PCoA of the ‘relaxed’ datasetwithall17 populations (including D. monticola).

species. The remaining four D. gregaria individuals were inter- analyses. The D. monticola population, supported by 100 % of mingled with the other three species. The remaining clusters all the trees, was separated from the main group with the greatest showed a mixed representation of the three described species number of mutations (tree height) and was in agreement with (D. basaltica, D. chryseopsis and D. behrii). The unrooted den- the previously built Diuris phylogeny (Indsto et al.,2009). The drogram calculated from population pairwise FST values only other structure detected by the coalescent model was the showed a similar pattern. For some populations (ALP, WG, bifurcation between the D. gregaria populations and the other SK, CG and YP), FST values could not be calculated due to three species from the VVP, which was supported with 100 % missing data and/or small population size, but the pattern ob- of the trees. We considered this to be evidence of a separate ge- served in the unrooted dendrogram and the PCoA showed dif- netic lineage. There was no structure between the remaining ten ferentiation between D. gregaria and the other populations. populations of D. basaltica, D. chryseopsis and D. behrii.The The SNAPP analysis (Fig. 4; coalescent model) showed a tree topology showed that the VVP populations were all con- similar pattern to those found with the genetic distance nected within the 001 grid, meaning that there were few Ahrens et al. — Diuris population genomics Page 7 of 11 mutational differences between populations or the described VVP genetic clusters, which is in agreement with the separation species. Bayesian clustering analysis of the ‘stringent’ data set of this population in the PCoAs. indicated that an optimal K-value of 2 was sufficient to explain The log of genetic distance was plotted against the log of the results via a strict Evanno interpretation (Fig. 5; morphological distance with (Fig. 6A) and without (Fig. 6B) Supplementary Data Fig. S5). We also showed a K-value of 3 D. monticola. Two distinct clusters were present when D. mon- because there was a secondary peak at three in the DK graph. ticola was included (Fig. 6A). The cluster to the right com- Biologically, the K ¼ 3 analysis showed finer structure within prised D. monticola individuals whereas the cluster to the left the VVP populations by separating D. gregaria populations comprised the four VVP species. The generalized linear model from the other three species and showing admixture between D. accounted for a small percentage of the variation in the data gregaria and three of the D. chryseopsis populations. The DK (r2 ¼ 0023) but was significant (P ¼ 21 10–16). Here, ge- population showed a nearly perfect admixture between the two netic distance did have a weak relationship with morphological distance. When D. monticola was removed (Fig. 6B), even this weak relationship was lost, indicating that morphology was varying within a larger well-connected population.

D. basaltica D. gregaria DISCUSSION D. behrii The five species we investigated accentuate the potential con- D. chryseopsis servation risk and management shortcomings associated with D. monticola reliance on species delineations based solely on non-exclusive morphological characters. Morphological variation within the group had a large amplitude and the characters used for species identification overlapped in the descriptions of currently recog- IIVV nized taxa (Fig. 2A; y-axis). On the basis of our results, mor- GGHH WooWoo CMCM phological features alone were poor indicators of species ParPar ACAC PPFF divergence and not suitable to define conservation units. This is RMRM MLML not surprising as Diuris is described as having few reliable mor- CDCD CWCW 0.001 phological characters for classification (Indsto et al.,2009). Our genetic data have provided evidence of some isolation FIG. 3. Phylogenetic representation of the ‘relaxed’ data set using genetic dis- between D. gregaria and the other three Diuris species found tance. Unrooted dendrogram calculated using PCoA on the ‘stringent’ data set. on the VVP. However, our findings did not confirm previously The inset shows the FST dendrogram from the ‘stringent’ data set. Some populations were removed from the FST dendrogram because the populations were ei- described groupings at the species level, nor did the findings ther too small or had too much missing data. correspond to population structure. There are important differences when comparing the FST between the microsatellite data

0·150 0·125 0·100 0·075 0·050 0·025 0 ALP AC ML 100 CG 100 YP Woo SK 100 IV CW CM DK Par PF CR WG RC GH

FIG. 4. A species complex tree obtained from a coalescent MCMC analysis of 97 individuals using SNAPP and BEAST on the ‘stringent’ data set. Support values are given for clades present in > 80 % of the iterations. Branch lengths are interpreted as relative time. Branches are identiﬁed by population abbreviations given in Table 1. Species are represented by colours. Page 8 of 11 Ahrens et al.—Diuris population genomics

IV GH SG ML CG YP AC DK PF CD CM WG Woo CW RW Par ALP

N Australia state boundaries

BIOREGION 0 50 100 200 km Victorian volcanic plain

FIG. 5. Bayesian clustering analysis of the ‘stringent’ data set using (A) two genetic clusters (k ¼ 2) and three genetic clusters (k ¼ 3) in STRUCTURE. Each column (individual) is assigned a specific suite of colours corresponding to the membership coefficient. Populations are labelled according to Table 1. The line above STRUCTURE figures represents the five species (from left to right D. basaltica, D. gregaria, D. behrii, D. chryseopsis and D. monticola). (B) The posterior probabil- ity of the STRUCTURE analysis in geographical space. Genetic clusters are averaged across populations, and genetic cluster are analogous to Fig. 4A.

sets provided here and in other studies. The FST among all four species delineations. Instead, our genetic results suggest that VVP species (003) is much lower than the average within- current species delimitations are not based on consistently dif- species FST (0092) for 22 studies of rare orchid species ferent morphological characters, and are unsupported by ge- (Phillips et al., 2012). This is indicative of a well-connected netic structure. Therefore, current species delimitations are not meta-population that should be conserved as such. Indeed, the suitable as conservation units. genetic data indicated a high level of gene flow and a low level The measured divergence of D. gregaria implied some level of inbreeding in the Diuris complex, including the nationally of reproductive isolation despite sympatry with other members endangered D. basaltica populations. Current genetic structure of the VVP group. However, the degree of differentiation be- suggests that gene flow has not been constrained significantly tween D. gregaria and the other VVP species is minimal because by reproductive, geographic or environmental isolation and is it represents only a tiny fraction of the total variation within the consistent with recent habitat fragmentation. Using the current GBS data sets (416 %). This leads us to an important decision species as our pre-defined populations, we would expect to find by which we must determine what demarcates an ESU and evidence of correlation between genetic differentiation and where the ESU threshold lies. In the case of this Diuris group, Ahrens et al. — Diuris population genomics Page 9 of 11

A B 6

2 log(morphological distance) Model: 1 log(m.dist) = –0·80 + 1·14(log(g.dist)) P = 2·1×10–6 R2 = 0·023 0 3·5 4·0 4·5 3·8 4·0 4·2 4·4 4·6 log(genetic distance) log(genetic distance)

FIG. 6. Results from correlation analyses between the ‘relaxed’ data set [genotyping by sequencing (GBS)] and morphology for the Diuris species. Genetic distance vs. morphological distance along with a best-fit line from a GLM model, with D. monticola (A) where the grouping on the right contains the individuals from the outgroup and (B) without. we are taking into account genetic diversity and current taxon- effects of rapid environmental change (Hoffman et al.,2015; omy. Therefore, for conservation purposes, we propose one ESU Bragg et al.,2015). More generally, a meta-analysis of frag- that includes D. gregaria and the other VVP species. We priori- mented, inbred populations shows that outcrossing of inbred tize one ESU because we consider that the risk of negative con- populations had benefits for 923 % of the cases included sequences that can occur through isolation (unknown) is likely to (Frankham, 2015). Frankham (2015) found that increased gene be higher than through genetic connectedness (already estab- flow reduced inbreeding while increasing fitness effects and lished). In contrast to species growing on the VVP, D. monticola evolutionary potential. While not perfectly analogous to this sit- comprises a distinct genetic lineage despite its morphological uation due to low inbreeding of the Diuris group, the principles similarity. This divergence may be a result of reproductive and of genetic rescue remain germane because ongoing gene flow environmental isolation between D. monticola and Diuris popu- among populations within the group is critical for the mainte- lations on the VVP. In its alpine environment, D. monticola nance of genetic diversity and evolutionary processes. In es- flowers later than the VVP populations (November–January vs. sence, any reduction in genetic diversity could have major September–October). As D. monticola occurs within the geo- ramifications on survival and fitness within the group. If the out- graphic range of D. behrii and D. chryseopsis in eastern come of conservation strategies artificially restricts genetic con- Victoria, phenological data are required to determine the possi- nectivity by protecting few populations or those that contain bility of gene flow between the three species. only a sub-set of a larger well-connected metapopulation, overall evolutionary potential and long-term viability could be se- verely limited for the Diuris species studied here. Evolutionary implications When conservation focuses on one entity of a species Conservation and management complex, such as the nationally endangered D. basaltica,the downstream effects can include long-term negative impacts on When taxonomic uncertainty exists, extinction risks could be evolution within the group (Ellstrand, 2014). By using taxonom- mitigated by managing gene pools more broadly to retain evolu- ically uncertain species as the conservation unit, land managers tionary processes, regardless of species boundaries. This inclu- and conservation practitioners risk inadvertently prioritizing sive strategy has the potential to simulate historical patterns of only a portion of the broader interactive gene pool. The Diuris gene flow as well as minimizing risks associated with inbreed- populations to be conserved tend to be small, and thus are prone ing, low genetic diversity and genetic drift. The benefit of main- to negative impacts from stochastic processes such as genetic taining allelic diversity for the group as a whole facilitates drift and non-random mating (Whiteley et al., 2015). As such, evolutionary processes (Sgro et al., 2010) that could buffer the future speciation, species persistence and the rate of genetic Diuris group from negative impacts from environmental change. change could be compromised by artificial genetic selection High allelic diversity, as found in this study, facilitates genetic (Dynesius and Jansson, 2013). The outcome would be a skewed responses to an uncertain future and can aid conservationists in reduction in genetic diversity, particularly affecting small popu- predicting evolutionary potential (Vilas et al., 2015). Given the lations. A reduced gene pool is likely to be less resilient to the rapid demographic decline but lack of distinct genetic lineages Page 10 of 11 Ahrens et al.—Diuris population genomics in the Diuris populations across the VVP, we propose that treat- ACKNOWLEDGMENTS ing the group as one ESU is a less risky strategy than using taxo- We thank N. Walsh, J. Jeanes and K. Brown for their valuable nomically questionable species as conservation units. input. This work was supported by by Hansons Construction With increased focus on active conservation and restoration Materials Pty Ltd. (Barnosky et al.,2011), it is critical to incorporate genetic diversity and evolutionary potential into management practices. At the same time, funds for on-ground conservation work in the LITERATURE CITED state of Victoria and within Australia more broadly are often Ahrens CW, James EA. 2014. Characterization of 13 microsatellite markers for predicated on the rarity of a species so conservation strategies Diuris basaltica (Orchidaceae) and related species. Applications in Plant must be created within that framework. We are unable to revise Science 2: 1300069. the taxonomic treatment of this Diuris group based on the data Ahrens CW, James EA. 2015. Range-wide genetic analysis reveals limited we have presented, but the challenging taxonomy of this group structure and suggests asexual patterns in the rare forb Senecio macrocar- pus. Biological Journal of the Linnean Society 115:256–69. should not dissuade funding for the protection of the few re- Ahrens CW, James EA. 2016. Regional genetic structure and environmental maining populations. A comprehensive revision of Diuris that variables influence our conservation approach for feather heads (Ptilotus includes the study taxa and considers the results presented here macrocephalus). Journal of Heredity 107: 238–247. is needed. With this, we encourage that decisions made for one Allendorf FW, Hohenlohe PA, Luikart G. 2010. Genomics and the future of species should be made in the context of the others. Future rein- conservation genetics. Nature Reviews Genetics 11: 697–709. Backhouse G, Lester K. 2010. National recovery plan for the Small Golden troductions are planned for D. basaltica, and we believe that it Moths Orchid Diuris basaltica. Melbourne Victoria: Department of would be in the best interest for the species to not be isolated Sustainability and Environment. from other species. Actions from this type of forward thinking Barnosky AD, Matzke N, Tomiya S, et al. 2011. Has the Earth’s sixth mass ex- could mitigate against the detrimental effects of genetic drift tinction already arrived? Nature 471: 51–57. Bateman RM, Devey DS, Malmgren S, Bradshaw E, Rudall PJ. 2010. and other stochastic influences from environmental change. Conflicting species concepts underlie perennial taxonomic controversies in Our data suggest that anthropogenically assisted migration Ophrys. Cahors Societe Francaise d’Orchidophilie 7: 87–101. from any species on the VVP to the WG population is likely to Blacket MJ, Robin C, Good RT, Lee SF, Miller AD. 2012. Universal primers be beneficial for longevity and genetic diversity. for fluorescent labelling of PCR fragments – an efficient and cost-effective approach to genotyping by fluorescence. Molecular Ecology Resources 12: 456–463. Bouckaert R, Bryant D. 2012. A rough guide to SNAPP.https://snap-mcmc.goo CONCLUSIONS glecode.com/files/SNAPPv1.1.pdf. Bouckaert R, Heled J, Ku¨hnert D, et al.2014.BEAST 2: a software platform When categorical species assignment cannot fully explain the for Bayesian evolutionary analysis. PLoS Computational Biology 10: variability of an organism, rigid adherence has the potential to e1003537. result in weakened management outcomes. All organisms are ex- Bragg JG, Supple MA, Andrew RL, Borevitz JO. 2015. Genomic variation posed to changing environmental conditions and must migrate or across landscapes: insights and applications. New Phytologist 207: 953–967. Bryant D, Bouckaert R, Felsenstein J, Rosenberg NA, RoyChoudhury A. adapt if they are to persist. High levels of genetic diversity that 2012. Inferring species trees directly from biallelic genetic markers: bypass- are subject to selection, gene flow and evolutionary processes ing gene trees in a full coalescent analysis. Molecular Biology and provide the best strategy to optimize environmental resilience. Evolution 29: 1917–1932. This work illustrates a robust method that can be used for taxo- Chen X, Guan J, Ding R, et al.2013.Conservation genetics of the endangered terrestrial orchid Liparis japonica in Northeast China based on AFLP nomically difficult organisms to prevent uncertainty from imped- markers. Plant Systematics and Evolution 299: 691–698. ing conservation. We envisage that this work will help to bridge Chung MY, Lopez-Pujol J, Maki M, et al.2012.Genetic diversity in the com- the gap between theory and practice by identifying unique attrib- mon terrestrial orchid Oreorchis patens and its rare congener Oreorchis cor- utes of organisms or populations to maintain biological diversity eana: inference of species evolutionary history and implications for and evolutionary trajectories. Ultimately, managing those attrib- conservation. Journal of Heredity 103: 692–702. Clark LV, Jasieniuk M. 2011. POLYSAT: an R package for polyploid micro- utes in the context of broader circumstances (e.g. climate change, satellite analysis. Molecular Ecology Resources 11: 562–566. fragmentation or higher levels of taxonomy) will lead to fruitful Devey DS, Bateman RM, Fay MF, Hawkins JA. 2007. Friends or relatives? strategies that benefit biological diversity. Phylogenetics and species delimitation in the controversial European orchid genus Ophrys. Annals of Botany 101: 385–402. Dillon S, McEvoy R, Baldwin DS, et al.2015.Genetic diversity of Eucalyptus camaldulensis Dehnh. following population decline in response to drought SUPPLEMENTARY DATA and altered hydrological regime. Austral Ecology 40: 558–72. Dynesius M, Jansson, R. 2013. Persistence of within-species lineages: a ne- Supplementary data are available online at https://academic.oup. glected control of speciation rates. Evolution 68: 923–934. com/aob and consist of the following. Figure S1: (A–F) Earl DA, vonHoldt BM. 2012. STRUCTURE HARVESTER: a website and VariationinfloralmorphologyofsixdifferentDiuris individ- program for visualizing STRUCTURE output and implementing the uals from the Parwan population. (G) Morphological variation Evanno method. Conservation Genetics Resources 4: 359–361. measured during the study vs. that provided in descriptions. Ellstrand NC. 2014. Is gene flow the most important evolutionary force in plants? American Journal of Botany 101: 737–753. Figure S2: principal co-ordinate analysis of the 13 morphologi- Elshire RJ, Glaubitz JC, Sun Q, et al. 2011. A robust, simple Genotyping-by- cal traits. Figure S3: principal component analysis of 13 mor- Sequencing (GBS) approach for high diversity species. PLoS One 6: e19379. phological measurements for all individuals in the study for five Evanno G, Regnaut S, Goudet J. 2005. Detecting the number of clusters of in- species. Figure S4: dendrogram from the ‘relaxed’ data set of dividuals using the software STRUCTURE: a simulation study. Molecular Ecology 14: 2611–2620. the Diuris species complex on the Victorian Volcanic Plain in Fazekas AJ, Kesanakurti PR, Burgess KS, et al.2009.Are plant species inher- south-eastern Victoria with replicates annotated with arrows. ently harder to discriminate than animal species using DNA barcoding Figure S5: line figure of Delta K from STRUCTURE Harvester. markers? Molecular Ecology Resources 9: 130–139. Ahrens et al. — Diuris population genomics Page 11 of 11

Fitzpatrick BM, Ryan ME, Johnson JR, Corush J. 2015. Hybridization and conservation prioritization and decision-making. Diversity and the species problem in conservation. Current Zoology 61: 204–214. Distributions 22: 174–188. Flanagan NS, Peakall R, Clements MA, Otero JT. 2006. Conservation of taxo- Pandey M, Richards M, Sharma J. 2015. Microsatellite-based genetic diversity nomically difficult species: the case of the Australian orchid, Microtis angu- patterns in disjunct populations of a rare orchid. Genetica 143: 693–704. sii. Conservation Genetics 7: 847–859. Pedersen H, Srimuang KO, Watthana S. 2016. Strengthening the taxonomic Frankham R. 2015. Genetic rescue of small inbred populations: meta-analysis backbone of Thai orchid conservation: genetic fingerprinting and morphom- reveals large and consistent benefits of gene flow. Molecular Ecology 24: etry applied to a species complex in Geodorum. Annals of Botany 118: 2610–2618. 125–133. Frankham R, Ballou JD, Dudash MR. 2012. Implications of different species Peterson BK, Weber JN, Kay EH, Fisher HS, Hoekstra HE. 2012. Double di- concepts for conserving biodiversity. Biological Conservation 153:25–31. gest RADseq: an inexpensive method for de novo SNP discovery and geno- Fu YB. 2014. Genetic diversity analysis of highly incomplete SNP genotype typing in model and non-model species. PLoS One 7:e37135. data with imputations: an empirical assessment. G3: Genesj Genomesj Phillips RD, Dixon KW, Peakall R. 2012. Low population genetic differentia- Genetics 4: 891–900. tion in the Orchidaceae: implications for the diversification of the family. Funk WC, McKay JK, Hohenlohe PA, Allendorf FW. 2012. Harnessing geno- Molecular Ecology 21: 5208–5220. mics for delineating conservation units. Trends in Ecology and Evolution Pilgrim ES, Crawley MJ, Dolphin K. 2004. Patterns of rarity in the native 27: 489–496. British flora. Biological Conservation 120: 161–170. Glaubitz JC, Casstevens TM, Lu F, et al.2014.TASSEL-GBS: a high capacity Pillon Y, Chase MW. 2007. Taxonomic exaggeration and its effects on orchid genotyping by sequencing analysis pipeline. PLoS One 9:e90346. conservation. Conservation Biology 21: 263–265. Grabowski PP, Morris GP, Casler MD, Borevitz JO. 2014. Population geno- Poland JA, Rife TW. 2012. Genotyping-by-sequencing for plant breeding and mic variation reveals roles of history, adaptation, and ploidy in switchgrass. genetics. Plant Genome 5: 92–102. Molecular Ecology 23: 4059–4073. Pritchard JK, Stephens M, Donnelly P. 2000. Inference of population structure Hausdorf B. 2011. Progress toward a general species concept. Evolution 65: using multilocus genotype data. Genetics 155: 945–959. 923–931. Qian X, Li QJ, Liu F, Gong MJ, Wang CX, Tian M. 2014. Conservation ge- Hedreń M, Nordstro¨m S, Sta˚hlberg D. 2012. Geographical variation and sys- netics of an endangered Lady’s Slipper Orchid: Cypripedium japonicum in tematics of the tetraploid marsh orchid Dactylorhiza majalis subsp. sphagni- China. International Journal of Molecular Sciences 15: 11578–96. cola (Orchidaceae) and closely related taxa. Botanical Journal of the R Core Team. 2014. R: a language and environment for statistical computing. Linnean Society 168: 174–193. Vienna, Austria: R Foundation for Statistical Computing. http://www.R-proj Hey J, Waples RS, Arnold ML, Butlin RK, Harrison RG. 2003. ect.org/. Understanding and confronting species uncertainty in biology and conserva- Rosenberg NA. 2004. distruct: a program for the graphical display of population tion. Trends in Ecology and Evolution 18: 597–603. structure. Molecular Ecology Notes 4: 137–138. HoffmannA,GriffinP,DillonS,et al.2015.A framework for incorporating Ross AA, Travers SE. 2016. The genetic consequences of rarity in the western evolutionary genomics into biodiversity conservation and management. prairie fringed orchid (Platanthera praeclara). Conservation Genetics 17: Climate Change Responses 2: 1–23. 69–76. Indsto JO, Weston PH, Clements MA. 2009. A molecular phylogenetic analy- Sgro CM, Lowe AJ, Hoffmann AA. 2010. Building evolutionary resilience for sis of Diuris (Orchidaceae) based on AFLP and ITS reveals three major conserving biodiversity under climate change. Evolutionary Applications 4: clades and a basal species. Australian Systematic Botany 22: 1–15. 326–337. Jakobsson M, Rosenberg NA. 2007. CLUMPP: a cluster matching and permu- Swarts ND, Dixon KW. 2009. Terrestrial orchid conservation in the age of ex- tation program for dealing with label switching and multimodality in analy- tinction. Annals of Botany 104: 543–556. sis of population structure. Bioinformatics 23: 1801–1806. Swarts ND, Sinclair EA, Krauss SL, Dixon KW. 2009. Genetic diversity in Jombart T. 2008. adegenet: a R package for the multivariate analysis of genetic fragmented populations of the critically endangered spider orchid markers. Bioinformatics 24: 1403–1405. Caladenia huegelii: implications for conservation. Conservation Genetics Jones DL. 1998. A taxonomic review of Diuris in Tasmania. Australian Orchid 10: 1199–208. Research 3: 72–85. Tear TH, Kareiva PM, Angermeier PL, et al. 2005. How much is enough? Jones DL. 2006. Miscellaneous new species of Australian Orchidaceae. The recurrent problem of setting measurable objectives in conservation. Australian Orchid Research 5: 45–111. Bioscience 55: 835–849. Laikre L. 2010. Genetic diversity is overlooked in international conservation Turchetto C, Segatto AL, Mader€ G, Rodrigues DM, Bonatto SL, Freitas LB. policy implementation. Conservation Genetics 11: 349–354. 2016. High levels of genetic diversity and population structure in an en- Magoc T, Salzberg SL. 2011. FLASH: fast length adjustment of short reads to demic and rare species: implications for conservation. AoB PLANTS 8:pii: improve genome assemblies. Bioinformatics 27: 2957–2963. plw002. doi:10.1093/aobpla/plw002. Martins EM, Lamont RW, Martinelli G, Lira-Medeiros CF, Quinet A, Vilas A, Pe´rez-Figueroa A, Quesada H, Caballero A. 2015. Allelic diversity Shapcott A. 2015. Genetic diversity and population genetic structure in for neutral markers retains a higher adaptive potential for quantitative traits three threatened Ocotea species (Lauraceae) from Brazil’s Atlantic than expected heterozygosity. Molecular Ecology 24: 4419–4432. Rainforest and implications for their conservation. Conservation Genetics Weir BS, Cockerham, CC. 1984. Estimating F-statistics for the analysis of pop- 16:1–14. ulation structure. Evolution 38: 1358–1370. Meirmans PG, Van Tienderen PH. 2012. The effects of inheritance in tetraploids Whiteley AR, Fitzpatrick SW, Funk WC, Tallmon DA. 2015. Genetic rescue on genetic diversity and population divergence. Heredity 110: 131–137. to the rescue. Trends in Ecology and Evolution 30: 42–49. Mondragon-Palomino M, Theissen G. 2009. Why are orchid flowers so di- Williams N, Marshall A, Morgan J. 2015. Land of sweeping plains. Canberra, verse? Reduction of evolutionary constraints by paralogues of class B floral Australia: CSIRO Publishing. homeotic genes. Annals of Botany 104: 583–594. Xiao LQ, Gong XU. 2006. Genetic differentiation and relationships of popula- Moritz C. 1994. Defining ‘evolutionarily significant units’ for conservation. tions in the Cycas balansae complex (Cycadaceae) and its conservation im- Trends in Ecology and Evolution 9: 373–375. plications. Annals of Botany 97:807–12. Nicotra A, Chong C, Bragg J, et al. 2016. Population and phylogenomic de- Yang Q, Fu Y, Wang YQ, et al 2014. Genetic diversity and differentiation in composition via Genotyping-By-Sequencing in Australian Pelargonium. the critically endangered orchid (Amitostigma hemipilioides): implications Molecular Ecology 25: 2000–2014. for conservation. Plant Systematics and Evolution 300: 871–879. Oksanen J, Blanchet FG, Kindt R, et al. 2013. vegan: community ecology pack- Zhang J, Li Z, Fritsch PW, Tian H, Yang A, Yao X. 2015. Phylogeography age. R package version 2.0-10. http://CRAN.R-project.org/package¼vegan. and genetic structure of a Tertiary relict tree species, Tapiscia sinensis Ottewell KM, Bickerton DC, Byrne M, Lowe AJ. 2016. Bridging the gap: a (Tapisciaceae): implications for conservation. Annals of Botany 116: genetic assessment framework for population-level threatened plant 727–737. 本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP 图书馆。图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具