Deciphering the sexual diploid members of the Boechera suffrutescens complex (Brassicaceae, Boechereae)

Abstract Boechera is a model genus that is of particular interest for understanding apomixis due to the presence of numerous apomictic diploid lineages that are tightly correlated with hybridisation events. Boechera includes many narrowly distributed endemics and apomictic hybrid lineages that obscure morphological boundaries amongst taxa. In this study, we focus on the Boechera suffrutescens complex, a phylogenetically well-supported but taxonomically complex north-western United States clade whose diploid species currently include the widespread B. suffrutescens and two narrowly distributed serpentine endemics, B. constancei and B. rollei. Using a 15-locus microsatellite dataset, we infer ploidy and sexual vs. apomictic reproduction for all individuals and then assess species limits for all sexual diploid samples. Our results support the recognition of B. rollei and B. constancei as distinct species and reveal three divergent sexual diploid lineages within B. suffrutescens sensu lato. The latter three lineages exhibit geographic, genetic and morphological coherence and consequently warrant recognition at the species rank. These include Boechera suffrutescens s.s., which is restricted to Idaho and eastern Oregon, Boechera botulifructa, a newly described species distributed along the Cascade Mountain Province from Lassen County, California north to Deschutes County, Oregon and the heretofore dismissed species Boechera duriuscula (basionym ≡ Arabis duriuscula), which occurs along the Sierra Nevada Province from Plumas County southwards to Fresno County, California. Our data also reveal substructure in B. constancei that is likely attributable to the highly fragmented distribution of its serpentine habitat. This refined taxonomic framework for the B. suffrutescens complex enhances Boechera as a model system, adds to our knowledge of speciation in edaphically extreme environments and provides information on ongoing conservation efforts for these taxa.


Introduction
The genus Boechera Á.Löve & D.Löve was first recognised in 1976, but it was not widely accepted as distinct from Arabis L. prior to 2003(Al-Shehbaz 2003. This largely North American lineage represents a complex assemblage of ±83 sexual diploid (S2X) taxa that have given rise to hundreds of apomictic hybrids, a situation that has confounded morphological classification since the first species were described in the 1820s (Li et al. 2017). The causes of this complexity include recent divergence and ongoing hybridisation, limited morphological disparity, edaphic shifts and the presence of both apomictic diploid (A2X) and triploid (A3X) hybrid lineages that are common and persistent across the distribution (Beck et al. 2011;Windham and Al-Shehbaz 2006;Windham and Al-Shehbaz 2007a;2007b). Boechera is nearly unique amongst flowering plants in possessing numerous diploid apomictic lineages (Bicknell and Koltunow 2004;Koltunow and Grossniklaus 2003). Understandably, the confluence of these characteristics has attracted considerable attention and Boechera has become a focal point for studies of biogeography, speciation, adaptation, apomixis and ecological genomics.
Despite widespread interest in Boechera as a model system (e.g. Rushworth et al. 2011) ongoing research has been hindered by limited understanding of species-level diversity, biogeography and phylogeny. Only recently, through a combination of molecular phylogenetic and population genetic studies, has genuine progress been made towards a coherent Boechera classification. This has involved a modified "diploids first" approach (Brown et al. 2002), acknowledging that it is nearly impossible to identify and study apomictic hybrids without an in-depth understanding of the sexual diploid species that gave rise to them. This approach has proven highly effective in Boechera, documenting cryptic biodiversity and bringing new clarity to both the B. fendleri and B. lignifera species complexes . Here, we apply this method to another poorly known group, the B. suffrutescens complex.
The ability to distinguish amongst different ploidy levels and reproductive modes in Boechera rests on several well-documented correlations derived from chromosomal, microsatellite heterozygosity and pollen data Beck et al. 2011;Li et al. 2017). Initially, pollen morphology was used as the primary indicator (Al-Shehbaz and Windham 2010; Windham and Al-Shehbaz 2006;Windham and Al-Shehbaz 2007a;2007b). S2X lineages produce pollen in tetrads through normal meiosis; the individual grains are mostly uniform, narrowly ellipsoid, 13-16 μm wide, with three symmetrical colpi (Suppl. material 1: fig. 1A). A3X lineages produce any functional pollen in dyads by means of apomeiosis; these grains are more irregular, ovoid-spheroid, 22-30 μm wide, with more than three asymmetric colpi (Suppl. material 1: fig. 1B) (Windham and Al-Shehbaz 2006) . A2X lineages usually produce predominantly malformed pollen (resulting from irregular meiotic events) mixed with functional meiotic and/or apomeiotic pollen (Suppl. material 1: fig. 1C) (Beck et al. 2011). More recently, an extensive 15-locus microsatellite database, encompassing nearly all known sexual diploid taxa and over 4400 accessions (Li et al. 2017), has made it possible to determine both ploidy level and reproductive mode through microsatellite analysis on a simple DNA sample. This dataset has confirmed previous reports of a bimodal distribution of heterozygosity across the genus Beck et al. 2011). Comparative meiotic studies of over 134 individuals representing 84 lineages of Boechera reveal that the left peak of this bimodal distribution (heterozygosity <0.5) consists almost entirely of S2X individuals while the right peak includes mostly apomicts. Amongst the apomicts, A3X lineages can then be distinguished from A2X lineages by the presence of three alleles at one or more of the 15 microsatellite loci Beck et al. 2011).
Our improved ability to sort Boechera specimens into natural groups, combined with cluster analysis of microsatellite data and phylogenetic analysis of DNA sequence data, have greatly improved our understanding of several S2X species complexes Windham et al. 2015). Nevertheless, there are many groups that require additional study to characterise extant sexual diploid diversity. One such group is the B. suffrutescens complex. This complex currently includes three S2X species (B. constancei (Rollins) Al-Shehbaz, B. rollei (Rollins) Al-Shehbaz and B. suffrutescens (S. Wats.) Dorn) that formed a maximally supported clade in genus-wide molecular phylogenetic analyses (Alexander et al. 2013). Two A3X species (B. horizontalis (Greene) Windham & Al-Shehbaz and B. rigidissima (Rollins) Al-Shehbaz) are believed to be hybrids between members of the B. suffrutescens complex and more distantly related species of Boechera (Al-Shehbaz and Windham 2010).
The group takes its name from Arabis suffrutescens S. Wats., which has been broadly defined to include populations from the Sierra Nevada, Trinity Alps, Cascades and isolated mountain peaks across the northern Great Basin, southern Columbia Plateau and Rocky Mountains of central Idaho. This highly variable taxon includes both S2X and A3X populations (Al-Shehbaz and Windham 2010), which occur in close proximity near the type locality along the Snake River Gorge in eastern Oregon. Eighteen years after Watson named A. suffrutescens, Greene described a segregate species, Arabis duriuscula Greene. The taxon was typified based on collections from Donner Lake, California, which Rollins (1941) subsequently treated as a taller and less suffrutescent phenotype of A. suffrutescens. Arabis dianthifolia Greene, described from the vicinity of Crater Lake (Greene 1910), has also been viewed as synonymous with A. suffrutescens (Al-Shehbaz and Windham 2010).
Two other taxa were segregated from Arabis suffrutescens by Rollins (1993b) and subsequently transferred to Boechera by Windham and Al-Shehbaz (2006). Boechera rollei is the most narrowly distributed taxon in the group, known only from the Trinity Mountains in Siskiyou County California (Fig. 1) and an isolated population along upper Beaver Creek, Jackson County, Oregon. The relative showiness of its flowers indicates that it is likely an outcrossing S2X lineage (Schmidt and Bancroft 2011). The other commonly accepted segregate is B. constancei, a narrow endemic apparently confined to Plumas and Sierra Counties, California. This taxon is known to be diploid based on a published chromosome count from the type locality (Rollins and Rüdenberg 1971) and it exhibits protogyny with distinctly elongated styles, which is suggestive of outcrossing (Schmidt and Bancroft 2011). In addition to being of conservation concern, both B. constancei and B. rollei appear to be restricted to serpentine soils, a model substrate for studying the links between edaphically extreme environments and divergent plant speciation (Kruckeberg 1951;1984;2002).
Although B. constancei and B. rollei are generally separable from the wide-ranging B. suffrutescens, there are some collections that appear to be morphologically intermediate (Rolle, pers. comm.). There also are unresolved questions regarding the placement of Arabis duriuscula and A. dianthifolia in synonymy under Boechera suffrutescens, as well as the relationship between S2X and A3X populations of the latter (Al-Shehbaz and Windham 2010). The purpose of this study is to identify and characterise the S2X lineages (taxa) within the B. suffrutescens complex to provide a framework for future investigations into the origins of related A2X and A3X lineages. Along with traditional macro-morphological and pollen analyses, we apply the set of 15 microsatellite loci, previously employed by Beck et al. (2011) andAlexander et al. (2015), to both herbarium specimens and extensive recent field collections. Pollen and microsatellite data are used to infer the ploidy and reproductive mode of each accession. The S2X individuals singled out by this process are used in a series of population genetic analyses to identify genetically coherent lineages worthy of species-level recognition.

Sampling
Samples for the project were obtained from individuals representing the morphology and known geographic range of the complex (Fig. 1), including 150 newly added collections and 348 previously collected herbarium samples. Holotype specimens of Arabis suffrutescens, A. suffrutescens var. perstylosa, A. rollei, A. constancei and an isotype of A. duriuscula were included. The holotype of A. dianthifolia was observed online through the Smithsonian plant database (https://collections.nmnh.si.edu).

Correlation between pollen morphology and reproductive mode
In concert with other data, pollen morphology was used to assign or confirm the assignment of individual plants to S2X, A2X or A3X categories. Pollen samples of adequate quality were obtained from 45 individuals and were analysed and categorised following Beck

DNA extraction
Genomic DNA was extracted using a modified version of the protocol outlined in Alexander et al. (2015). The deviation included: dried leaf samples being homogenised without buffer, eluting the pulverised material in grinding buffer plus 12 μl (>600 mAU/ml) of Proteinase K and an incubation with agitation at 65 °C for 12-15 hours prior to moving on to the next step.

Identification of S2X individuals and populations
All samples with data for at least 8 of the 15 microsatellite loci were retained for analysis. The ploidy level of each sample was then estimated using the criteria outlined by Beck et al. (2011). In short, if an individual exhibited no more than two alleles per locus, it was inferred to be diploid; if three alleles were present at one or more loci, it was inferred to be triploid. Following Alexander et al. (2015), the mode of reproduction amongst diploids (e.g. S2X vs. A2X) was then inferred via an average number of alleles per non-null locus (ANA/NNL) approach. The S2X category was initially set to a mean ANA/NNL ≤1.5  and later reduced to ≤1.35 or less following downstream population genetic analyses that identified hybrid "B. constancei" with values above 1.35 (discussed below).

Analyses of population structure within and amongst S2X taxa
Following Alexander et al. (2015), we employed a hierarchical approach to investigate patterns of microsatellite variation and population differentiation amongst putative species-level lineages. STRUCTURE employs a parametric Bayesian approach to investigate the most likely number of differentiated (K) population systems (Falush et al. 2003;Pritchard et al. 2000). Exploratory STRUCTURE analyses were performed with the admixture model using default settings with 50,000 burn-in and 500,000 postburn in generations with five iterations at each value of K from 1-11. Final STRUC-TURE analyses included 100,000 burn-in and 1,000,000 post-burn in generations with 10 iterations for each value of K from 1-11. The most likely value of K for each analysis was identified using the ∆K method of Evanno et al. (2005) as implemented in STRUCTURE Harvester (Earl and vonHoldt 2012). Null alleles at seven loci (B11, C8, BF15, Brdu266, e9, BF3 and BF19), clearly corresponding to either prior taxonomic assignment and/or geographic structure, were coded in STRUCTURE using RECESSIVE ALLELES = 1 (Falush et al. 2007).
AWclust utilises a nonparametric approach to infer population structure based on allele sharing distance (Gao and Starmer 2008). Critically, this approach does not incorporate a model of within-group Hardy-Weinberg equilibrium, an assumption that is likely unrealistic considering the interspecific, biogeographic and temporal (inclusion of historical specimens) scope of our sample set. Multidimensional scaling plots (MDS) were generated in AWclust to visualise relative coherence and distinctness of clusters based on allele sharing distance. AWclust estimates the optimal number of clusters (K) via the gap statistic (Tibshirani et al. 2001), whereby individuals are assigned to clusters at the optimal K through the implementation of Ward's minimum variance hierarchical clustering. Gap statistics were calculated for a given data set with 100 null simulations for K values 1-11. The aforementioned null alleles were also treated as characters in AWclust.

Morphological assessment a posteriori
Individuals inferred as representing S2X species-level lineages through the aforementioned analyses were subsequently studied in detail to identify diagnostic morphological characteristics for the taxonomic treatment.

Correlation between pollen morphology and reproductive mode
Forty-five accessions harboured pollen of sufficient quality and quantity for morphotyping. Of these, 26 individuals exhibited ovoid-spheroid, multicolpate pollen consistent with apomictic reproduction, 14 exhibited narrowly elliptic, tricolpate pollen consistent with sexual reproduction and five exhibited presumably non-viable pollen with a spheroid, ecolpate morphology. There was 92% agreement between mode of reproduction inferred via pollen morphology and that inferred by the maximum number of alleles per locus (see below). This high correlation is consistent with prior studies by Beck et al. (2011), who reported a 96% correlation in a sample of 330 specimens.

Identification of S2X individuals and populations
The maximum number of alleles per locus criterion Beck et al. 2011) identified 191 triploid (which were excluded from further analysis (see Appendix 1)) and 307 diploid (S2X and A2X) individuals (see Table 1, "Additional specimens examined" and Appendix 1). For ease of data presentation in Table 1 and this text, populations are represented by abbreviations that include a locality prefix (CD = Canyon Dam, CP = Cascade Province, GB = Great Basin, OVR = Onion Valley Reservoir, Table 1. Summary of sexual diploid (S2X), apomictic diploid (A2X), and apomictic polyploid (A3X and A4X) assignments and clusters inferred from preliminary analyses. Polyploids, A2X clusters, and singletons were excluded from the final S2X analyses (see text). PLSI = Plumas and Sierra Counties, SNP = Sierra Nevada Province, TL = type locality region,) and a species suffix (co = constancei, ro = rollei and su = suffrutescens s.l.). If the specific identifier is preceded by an 'x' (e.g. OVR-xco), the group is a putative hybrid lineage assigned to the A2X category. Ploidy assignment for a small number of individuals was inconsistent with prior inferences for their taxon. In particular, 9 of 96 B. constancei individuals were inferred to be triploids despite prior diploid inference from a smaller sample of individuals (Rollins and Rüdenberg 1971).

Subsequent analyses focused on differentiating A2X and S2X individuals
The 1.5 ANA/NNL criterion Beck et al. 2011) identified 238 putative S2X individuals. Preliminary STRUCTURE runs were then employed to fine-tune the ANA/NNL cutoff. These preliminary studies identified nine putative population systems (excluding "singletons") with 16-59 individuals per group (S2X and A2X in Table 1). A small subset of three individuals from Falcon Valley, Washington did not cluster with other population systems. "Falcon Valley" is an anomalous place name used by W.N. Suksdorf and we are unable to determine from where these specimens were collected. Given the geographic uncertainty and poor sampling of this lineage, plants from "Falcon Valley" were excluded from further analysis. A group of individuals (OVR-xco) from Onion Valley Reservoir that is morphologically assignable to B. constancei showed genetic admixture. These individuals, inferred to represent a previously undetected A2X hybrid lineage, exhibited a mean ANA/NNL of 1.36. In light of this, the S2X mean cutoff was reset to <1.35 to provide a more conservative circumscription of the S2X category. After applying these filters, we were left with 235 inferred S2X individuals to be included in the final analyses.

Figure 2.
Results from Analysis 2 (which excluded TL-co). A STRUCTURE bar plot with K = 6 B MDS microsatellite plot of first three axes at K = 6 as found with the gap statistic C MDS plot with TL-su and TL-ro, the most divergent clusters, excluded to demonstrate coherence of the remaining four clusters (gap statistic K = 4).

Analyses of population structure within and amongst S2X taxa
Our diploid only (see above) and preliminary S2X only analyses revealed conflict and instability in the optimal K inferred by STRUCTURE as well as between AWclust and STRUCTURE. Analysis 1, including all 235 S2X individuals, yielded two equally optimal K (3 and 8) in STRUCTURE and two equally optimal K values (4 and 8) in AWclust. The instability observed within and between these analyses was the result of conflicting assignments for individuals of B. constancei from the type locality (TL-co). TL-co individuals either formed a unique cluster (Suppl. material 2: fig. 2A, K = 6 and 8) or occasionally grouped with CP-su individuals (Suppl. material 2: fig. 2A, K = 5). These findings were consistent with potential introgression involving TL-co and CPsu. Given the instability associated with TL-co, we performed a second round of analyses (Analysis 2) without TL-co, which yielded an unambiguous K = 6 from both STRUCTURE and AWclust. This array specifies B. rollei as a single cluster, but supports two distinct clusters (CD-co and PLSI-co) within B. constancei and three distinct clusters (CP-su, SNP-su and TL-su) within B. suffrutescens s.l. (Fig. 2). Each of these clusters also occupies a discrete geographic range (Figs 3, 4) with possible introgressant populations (TL-co) located in close proximity to the most similar putative parent (B. constancei) but nearly 100 km south of the documented range of the other (CP-su).

Morphological Assessment a Posteriori
Morphological comparisons of the clusters identified by STRUCTURE and AWclust revealed a variety of features useful for distinguishing these putative taxa. Character state differences in petal length, style length and the presence or absence of auricles on the cauline leaves have been used to separate B. rollei and B. constancei from B. suffrutescens s.l. Each of these features is consistent with differentiation amongst these lineages.
Additionally, we have identified a series of morphological features that support the recognition of the three clusters belonging to B. suffrutescens s.l. (CP-su, SNP-su, and TL-su). These include basal leaf pubescence, shape of the fruit apex and the length-towidth ratio of mature basal leaves (see "Taxonomic Account" section below).

Assignment of ploidy and characterisation of S2X
The stepwise process employed to first parse diploids from polyploids and then S2X from A2X, identified 238 S2X individuals out of a total sample of 498 B. suffrutescens complex samples. The fact that more than half of the individuals were A3X or A2X clearly illustrates that the B. suffrutescens complex harbours the hybridisation, apomixis and polyploidy trifecta that has severely hindered species-level taxonomy in the genus as a whole Windham and Al-Shehbaz 2007a;2007b). The analytical approach taken here supports the recognition of five sexual diploid taxa within the B. suffrutescens complex. These lineages include the current circumscriptions of B. constancei and B. rollei and a recircumscription of B. suffrutescens that recognises three distinct taxa. Each of these five taxa is discussed below.

Boechera rollei
All 36 individuals of Boechera rollei were clearly defined as S2X by allele numbers (Table 1) and as a distinct group by cluster analyses (Fig. 2). The taxon is very rare, with just three known populations restricted to serpentine soils in Siskiyou County, California and Jackson County, Oregon. Its distribution overlaps with the broadly distributed B. suffrutescens s.l. (Fig. 1), but the two have not been observed growing together. Boechera rollei is separable morphologically from all other S2X members of the complex by its unusually large (8-11 mm long) cream-coloured (vs. lavender) petals and non-geniculate fruiting pedicels. It is further separable from B. constancei by its auriculate cauline leaves and shorter styles (≤1.5 mm). Genetics, ecology, geography and morphology all support recognition at the species level, a conclusion that is consistent with prior taxonomic treatments (Al-Shehbaz and Windham 2010; Rollins 1993a) and meets the criteria proposed by the genetic species concept (Baker and Bradley 2006;Bateson 1909), the phylogenetic species concept (Nixon and Wheeler 1990) and the genotypic cluster concept (Mallet 1995) for species level recognition.

Boechera suffrutescens sensu lato
Boechera suffrutescens s.l. is by far the most widespread and morphologically heterogeneous taxon in the complex. With regard to the S2X lineages, STRUCTURE and Table 2. Summary S2X only analyses one and two. 'X' indicates inclusion in the analysis using both AWClust and STRUCTURE.

Formal
Analysis CD-co CP-su TL-co PLSI-co SNP-su TL-su TL-ro AWclust analyses subdivided individuals identified as S2X B. suffrutescens s.l. into three geographically distinct clusters (CP-su, SNP-su and TL-su) with little or no evidence of admixture ( Fig. 2 and Table 2). The population system (TL-su) that includes the type locality for B. suffrutescens is S2X based on ANA/NNL results (mean 1.142) and pollen morphology (Table 1). It is separable form other S2X lineages based on several lines of evidence. It is genetically distinctive, forming a cohesive cluster in both STRU-CURE and AWclust analyses (Fig. 2) and it exhibits unique fixed alleles at both the e9 and BF9 microsatellite loci. Its geographic range, extending ca. 100 km north-south along the Idaho-Oregon border with outlying populations in Fremont County, Idaho and Grant Counties, Oregon (Fig. 3), is separate from those of the other S2X taxa. This lineage is morphologically distinctive as well, characterised by having narrower basal leaves (length/width ratio usually ≥8:1) and sparser pubescence relative to the CP-su and SNP-su lineages (see "Taxonomic Treatment"). The combination of these features support recognition of this cluster as B. suffrutescens s.s. The two remaining S2X lineages currently included within B. suffrutescens s.l. are distributed from the southern Sierra Nevada north into the central Cascades (Fig. 3). Both exhibit mean ANA/NNL values and pollen morphologies consistent with S2X assignment (Table 1). Like the TL-su cluster, members of the SNP-su group formed a cohesive cluster in both STRUCTURE and AWclust analyses (Fig. 2). Our sampling, comprising 59 individuals representing 23 populations of SNP-su, is distributed along the Sierra Nevada from Fresno to Plumas Counties, California (Fig. 3). At the northern end of its range, SNP-su overlaps with the distribution of B. constancei and hybridisation between these taxa may have given rise to the presumed A2X lineage OVR-xco (see discussion below). The SNP-su cluster is separable from sympatric populations of B. constancei by having shorter styles (≤1.5 mm) and auriculate cauline leaves and is distinguished morphologically from the other suffrutescens s.l. S2X taxa by fruit and pubescence characters. This cluster included an isotype of Arabis duriuscula Greene, a taxon that has been treated as a synonym of B. suffrutescens (e.g. Al-Shehbaz 2003; Al-Shehbaz and Windham 2010; Rollins 1993a; Windham and Al-Shehbaz). The findings presented here support the recognition of a distinct taxon requiring the recognition of Arabis duriuscula at the species level in Boechera (see "Taxonomic Treatment").
The second group previously assigned to B. suffrutescens s.l., the CP-su cluster, is represented by 38 individuals. Like the other two suffrutescens s.l. clusters, CP-su is genetically distinct in both STRUCTURE and AWclust analyses (Fig. 2). Current sampling suggests that it occupies a discrete geographic range in the Cascade Moun-tain Province extending from Lassen and Siskiyou Counties, California to Deschutes County, Oregon (Fig. 3). In addition to being genetically and geographically distinct, members of the CP-su cluster are separable from the other S2x taxa previously assigned to B. suffrutescens s.l. based on the distinctive, ovoid-shaped fruit apices. None of the species-level names, previously formalised in Boechera or Arabis, appear to be applicable to this taxon and we therefore propose a new name: B. botulifructa (see "Taxonomic Treatment").

Boechera constancei
Previous studies of the obligate serpentine endemic B. constancei considered it S2X based on chromosome counts (Rollins and Rüdenberg 1971) and pollen morphology (Al-Shehbaz and Windham 2010). Amongst our sampling of 87 individuals representing 28 unique geographic sites, 71 samples from 26 localities were indeed assigned to the S2X category based on ANA/NNL ratios (Table 1). The 16 samples from Onion Valley Reservoir (OVR-xco) were inferred to be A2X. Even with OVR-xco removed, population genetic analyses including all S2X B. constancei revealed considerable instability. STRUCTURE and AWclust analyses recovered as many as three clusters, CD-co, PLSI-co and TL-co (Fig. 4). The apparent instability was associated with an affinity between TL-co and CP-su, potentially indicative of hybridisation or introgression between these S2X groups. The species is well separated morphologically from the other S2X members of the B. suffrutescens complex by its unusually long styles (≥1.5 mm) and consistently non-auriculate cauline leaves.
The isolated island-like biogeography (Ellstrand and Elam 1993;Young et al. 1996) of B. constancei on serpentine soils could explain the patterns of geographically defined genetic sub-structure observed in this taxon (Suppl. material 2: fig. 2B). Future work is especially needed to investigate the complex substructure observed in B. constancei and to determine whether segregate taxa worthy of recognition are contained within the species.

Evaluation of A2X lineages in the Boechera suffrutescens complex
Two major A2X clusters were evident in the complex. The aforementioned OVR-xco cluster consisted of 16 individuals collected from serpentine in the vicinity of Pilot Peak and Onion Valley Reservoir in Plumas County, California. The STRUCTURE allele assignment profiles for these individuals exhibited apparent admixture between SNP-su and B. constancei and they resolved in intermediate positions between these putative parents in AWclust plots (data not shown). No pollen data were available from which to infer mode of reproduction. Further research will be required to confirm this assertion and investigate the origin of this cluster.
The second major A2X cluster (GB-xsu) is broadly distributed across the mountain ranges of the Great Basin in California, Oregon and Idaho. It is morphologically assignable to B. suffrutescens s.l. Both the mean ANA/NNL of GB-xsu and pollen morphology from two individuals indicate that this is an A2X lineage (Table 1). Some additional preliminary analyses suggest it may have arisen through hybridisation between SNP-su and CP-su (Windham, unpubl. data).
The focus of this study was on circumscribing the sexual diploid taxa of the B. suffrutescens complex and our combined data support the recognition of at least five S2X species. Although not discussed in detail here, at least two A2X lineages and an even greater diversity of A3X hybrid lineages were also evident. More than half of the individuals included in this study ultimately were assigned to asexual groups and preliminary analyses imply that some of these lineages incorporate one or more genomes from outside of the B. suffrutescens complex (Morin, unpubl. data). This is consistent with prior observations that hybridisation and a transition to apomixis may be linked (e.g. Beck et al. 2011) and that recognising the impact of these phenomena is a critical part of deciphering Boechera diversity and evolution.

Taxonomic treatment
Members of the B. suffrutescens complex are distinguished from congeneric taxa by having relatively wide (2.5-6 mm) pendent fruits containing a single row of broadly winged (0.3-1.5 mm) seeds per locule. Previous molecular analyses (Alexander et al. 2013) have established that sexual diploid accessions of B. constancei, B. rollei and B. suffrutescens s.l. (represented by the SNP-su lineage) form a wellsupported clade. In addition to the five S2X taxa characterised below, we encountered many A2X and A3X individuals. Some of these exhibit morphological characters that clearly set them apart from S2X taxa, suggesting that they are products of hybridisation with other species groups. More problematic were the apomictic hybrids that have arisen within the B. suffrutescens group, blurring the already subtle distinctions amongst the S2X members of the complex. It should be noted that the taxonomic treatment provided below applies only to S2X individuals and that pollen morphology and/or allelic diversity are the only reliable means for distinguishing closely related sexual and apomictic lineages in Boechera as a whole (Beck et al. 2011;Windham and Al-Shehbaz 2006).

Key to sexual diploid taxa of the B. suffrutescens complex
Given the frequency of hybridisation in Boechera, pollen morphology should be characterised prior to proceeding with this key. An inference of sexual diploidy can be made for individuals that produce mostly well-formed, narrowly ellipsoid symmetrically tricolpate pollen (Suppl. material 1: fig. 1). In terms of the macromorphological characters used in the key, there is considerable variation within species and some inevitable morphological overlap between species. Multiple plants should be examined if possible. Caute procedere.

1
Petals mostly more than 6 mm long, cream white, but occasionally with rosecoloured apices (B. rollei); mature fruiting pedicels curved-descending or reflexed but never distinctly geniculate proximally; plants only known from serpentine ( 6 mm). Within the complex, the species is one of just five that produces narrowly ellipsoid symmetrically tricolpate pollen (Suppl. material 1: fig. 1A) indicative of dip-loid sexual reproduction. Boechera botulifructa is distinguishable from four other sexual diploid B. suffrutescens complex species by the presence of small petals (4-6 mm long), abruptly tapered silique distal apices and a geographic distribution along the Cascade Province in California and Oregon.
Distribution, habitat and phenology. As currently known, the species occupies three distinct regions in the Cascade Province: western Deschutes County, Oregon, near Medicine Lake, Siskiyou County, California and the area west of Eagle Lake in Lassen County, California. It favours rocky slopes and gravelly felsic soils in association with Artemisia tridentata, Purshia tridentata, Pinus jeffreyi, Pinus contorta and Juniperus at elevations of 1300-2100 m; flowering from May to July.
Comments. Morphologically, B. botulifructa is most similar to B. duriuscula and these two taxa are parapatric along the southern distribution of B. botulifructa. The species is distinguished from close relatives primarily by the abrupt tapering of the apex of the fruit, resulting in a sausage-like profile to which the specific epithet refers. Like most other members of the B. suffrutescens complex, it has a suffrutescent habit and wide (>3 mm), reflexed, often secund, fruits. Molecular data suggest that the specimens from the southernmost population in Lassen County, California, have diverged from the northern populations and may have a history of gene flow with B. constancei from the vicinity of its type locality. The latter B. botulifructa individuals, from Lassen County, also have reduced cauline leaf auricles relative to other non-serpentinicolous members of the complex.
Though the species spans a large geographic range, we have only identified nine populations systems thus far, suggesting a need for future investigation of conservation status. Within the populations we visited, individuals were sparsely dispersed across broad areas. The species occurs on public lands with noted impacts from grazing activity and potential impacts from logging of local native forests.

Distribution, habitat and phenology.
Boechera constancei is only known from the western slope of the Sierra Nevada in the vicinity of Lake Delahunty in Sierra County and adjacent southern Plumas County, California. It appears to be confined to a variety of serpentinic substrates in association with Pinus jeffreyi and other "serpentine barren" vegetation types at elevations from 1200-1900 m; flowering from Apr-Jun.
Comments. Boechera constancei was originally treated as a variety of Arabis suffrutescens s.l., but it is distinguished from members of that group by its non-auriculate cauline leaves, longer (6.0-8.0 vs. 4.5-6.0 mm) petals that are creamy white and longer (1.5-5.0 vs. 0.4-1.2 mm) style. Although restricted to serpentine substrates, it generally shows greater local abundance and higher population densities within its narrow geographic range than B. duriuscula, B. botulifructa and B. suffrutescens s.s.  (1882). The repository of the Sonne syntype(s) is unknown at this time; the Michener syntype deposited at NDG has been included in our morphological and molecular analyses and is here designated as lectotype.
Distribution, habit and phenology. Boechera suffrutescens is distributed north and east of the Great Basin; concentrated in the vicinity of the Snake River Gorge (Hells Canyon), but extending from Grant County, Oregon to central Idaho on steep, rocky, basaltic substrates in alpine and subalpine ecozones at elevations from 1800-2500 m; flowering from May-July.
Comments. Although geographically isolated from the other S2X species of the complex, B.suffrutescens s.s. is the least distinct morphologically. The most useful character for distinguishing this species is that individuals usually have basal leaves that are glabrate, with a few 1(-2) rayed trichomes scantily dispersed along the margins and apices. However, plants are occasionally encountered that are pubescent basally with 1-3(-4) rayed trichomes. These individuals often appear stressed or lack flowering stems, suggesting that pubescence may be more prevalent amongst plants growing in unfavourable environments. On the holotype specimen, one of each morphotype is present and the plant lacking a flowering stem is pubescent. All other taxa in the complex are consistently pubescent basally. On robust individuals of B. suffrutescens s.s., the basal leaves are generally narrower and the fruits are generally wider than those of the other S2X taxa.