Molecular and morphological evidence for recognition of two species within Harpagonella (Amsinckiinae, Boraginaceae)

Abstract Recent taxonomic treatments of the genus Harpagonella have included only one lower taxon, Harpagonella palmeri A. Gray. However, a larger-fruited variety of Harpagonella palmeri from Arizona and Sonora was described by I.M. Johnston in 1924. He continued to recognize this taxon – Harpagonella palmeri var. arizonica – in his treatment of the genus in Kearney and Peebles’s Arizona Flora in 1960. Here, we provide two lines of molecular evidence and quantitative morphological evidence from calyx characters showing that plants of Harpagonella from Arizona, Sonora, and central Baja California, corresponding to Johnston’s var. arizonica, are distinct from Harpagonella palmeri of southern California and Baja California. We make the new combination Harpagonella arizonica (I.M. Johnston) Guilliams & B.G. Baldwin, comb. nov. for the plants from Arizona, Sonora, and central Baja California.


Introduction
Harpagonella A. Gray is a genus of Boraginaceae, subtribe Amsinckiinae (see Chacón et al. 2016 andLuebert et al. 2016) that occurs disjunctly in western North America, with populations in southern California, USA, and adjacent Baja California, México and other populations in southern Arizona, USA, and adjacent northwestern Sonora, México (Figure 1). Th e only species recognized in the genus, H. palmeri A. Gray, was described in 1876 from an 1875 collection by Edward Palmer on Guadalupe Island, Baja California. In 1924, Ivan M. Johnston recognized two varieties in H. palmeri, var. arizonica and var. palmeri. Th e former taxon, then known from Arizona and adjacent Sonora, was said to diff er from var. palmeri, of California and Baja California, in hav-  ing longer "cornute processes on the fruiting calyx" and larger nutlets (Johnston 1924). Furthermore, the plants of California and Baja California are often found on clayey soils, while those of Arizona and Sonora often occur in sandy or gravelly soils. In his treatment of the Boraginaceae for the Arizona Flora (Kearney and Peebles 1960), Johnston retained the taxon as a variety, but most other treatments of the genus recognize H. palmeri without varieties (e.g., Munz 1973, Veno 1979, Kelley and Messick 2014. Harpagonella has been regarded as the most morphologically distinctive member of the Amsinckiinae, largely because of ornamentation of the calyx in fruit that is unique to the genus (Johnston 1924, Veno 1979. Th e genus was placed in its own tribe, Harpagonelleae, for this reason (Gürke 1897). In Harpagonella, the calyx is pentamerous, with the two sepals away from the infl orescence axis connate for >80% of their length and the three other sepals free while in fl ower. Th e two fused sepals are strongly accrescent, becoming conduplicate, indurate, and often more or less enveloping one nutlet or sometimes both nutlets at fruit maturity ( Figure 2). As the fruit matures, fi ve to ten subterete appendages with distal retrorse barbs develop on the pair of fused sepals, giving the fruit the appearance and function of a grappling hook, which is the common name for the genus. Th e pedicel is also accrescent. It recurves or rarely coils as the fruit matures, placing the lobes of the fused sepals against the infl orescence axis. As Gray (1876) noted, these modifi cations eff ectively result in the transfer of dispersal function from the nutlet, as is typical in many Amsinckiinae, to the calyx. Th e gynoecium in Harpagonella is also distinctive. It has been reduced from the typical condition in the Amsinckiinae of four ovules and a fruit of four nutlets to two developing ovules and two nutlets, with the other two ovules early abortive. Unlike the nutlets of many close relatives, e.g. Pectocarya, the two nutlets of Harpagonella are largely without ornamentation, bearing only short hairs. We included Harpagonella in broad phylogenetic and taxonomic studies of some members of the Boraginaceae subtribe Amsinckiinae (Guilliams 2015). During the phylogenetic study, we included several samples of H. palmeri from throughout its range with the goal of evaluating phylogenetic structure of the included samples, with attention to historical taxonomy. We also examined herbarium sheets representing both previously recognized varieties of H. palmeri, taking measurements of the calyx appendages and overall size of the fruit. Although a full phylogenetic study will be published later, we present the results of this study here in reduced form so that the resulting new combination can be available for use in the treatment of Harpagonella for the Flora of North America, North of México.

Phylogenetic analyses
DNA was extracted from 12 samples of Harpagonella and 2 samples of Pectocarya using a modifi ed CTAB protocol (Doyle and Doyle 1987). Samples included in this analysis are given in Table 1 and were selected on the basis of geographic distribution of the two putative taxa and recency of collection. Six of these samples were from Arizona and were morphologically consistent with H. palmeri var. arizonica sensu Johnston (1924). Th e other six samples were from California and adjacent Baja California and were morphologically consistent with H. palmeri var. palmeri. One sample each of Pectocarya linearis DC. var. ferocula I.M. Johnst. and P. recurvata I.M. Johnst. were included as outgroup taxa.
Polymerase chain reaction (PCR) was used to amplify the internal transcribed spacer (ITS) and the external transcribed spacer (ETS) of nuclear ribosomal DNA, and the rpl16, rps16, trnK-rps16, and trnL-trnF regions of the chloroplast genome. All PCR reactions except for those targeting the ETS region were performed using previously published primers and reaction conditions (see Baldwin et al. 1995, Shaw et al. 2005, Shaw et al. 2007. Th e 5' ETS primer was designed following the protocol of Baldwin and Markos (1998). PCR products were cleaned using USB ExoSAP-IT (Aff ymetrix, Santa Clara, CA, USA) using the standard protocol. Bidirectional sequencing was performed on an Applied Biosystems 3730xl DNA Analyzer at the Barker DNA Sequencing core facility at UC Berkeley. Contigs were assembled and edited in Geneious R6 (Drummond et al. 2013). Sequences were initially aligned under the default parameters using the Geneious alignment tool in Geneious, then further refi ned by hand.
For each DNA region, models of sequence evolution were estimated using jMod-elTest (Posada 2008). Bayesian phylogenetic analyses were performed and summarized using the BEAST suite of programs. Four separate analyses of 10 million generations were performed in BEAST v.1.7.4 , with the fi rst 25% of trees discarded as burn-in. Convergence was assessed using Tracer v.1.7.4 . Post burn-in runs were combined using Log Com- Table 1.

Specimens of
Harpagonella and outgroups used in phylogenetic analysis, including collector and collection numbers, herbarium accession numbers, and GenBank accession numbers by DNA region. Separate maximum likelihood analyses for nrDNA and cpDNA were performed using RAxML v1 plug-in in Geneious v8.1.8 (Drummond AJ et al. 2015). Maximum likelihood bootstrap values resulting from these analyses were added to the MCCT.

Morphological analyses
Morphological data were taken from a total of 32 physical specimens of Harpagonella palmeri var. arizonica and 27 physical specimens of H. palmeri var. palmeri. Physical specimens measured were those available from the ARIZ, JEPS, and UC herbaria with mature fruits. We also measured high quality digital scans of type material of both taxa. For each specimen, we measured and averaged values from up to fi ve fruits for maximum fruit length along an axis oriented from the pedicel base to the most distant point (including subterete appendages; mm), maximum fruit width along an axis perpendicular to maximum fruit length (including subterete appendages; mm), and maximum length of subterete appendages (mm). Measurements of physical specimens were taken with a digital caliper to the nearest hundredth of a millimeter. Measurement of digital specimens were made in ImageJ (Abramoff MD et al. 2004). Nutlet length has been reported as diff erent between the two varieties, but measuring this feature would have required occasional destructive sampling and was therefore avoided.
Morphological data were explored using boxplots and basic descriptive statistics. Student's t-tests were performed to evaluate the statistical signifi cance of the diff erences between the varieties for the features measured. All statistical analyses were performed in R (R Development Core Team 2008).

Phylogenetic patterns in Harpagonella
Th e nuclear dataset comprising ITS and ETS was 1,082 total bases in length. For these loci, jModelTest determined a best-fi t model of sequence evolution of GTR+I. In the matrix, 79 positions were variable and phylogenetically informative, 29 were variable and not phylogenetically informative, and 974 were invariant.
Th e MCCT resulting from the analysis of the concatenated nuclear DNA matrix is given in Figure 3A Th e chloroplast dataset comprising rpl16, rps16, trnK-rps16, and trnL-trnF was 3,442 total bases in length. For these loci, jModelTest determined a best-fi t model of sequence evolution of GTR+I. Of these, 51 positions were variable and phylogenetically informative, 30 were variable and not phylogenetically informative, and 3,361 were invariant.
Th e MCCT resulting from the analysis of the concatenated chloroplast DNA matrix is given in Figure 3B. Samples of each variety of Harpagonella are reciprocally monophyletic and clades by taxon are strongly supported. Th e clade of samples of var. arizonica was supported with a posterior probability of 0.96, and a maximum likelihood bootstrap value of 100. Th e clade of samples of var. palmeri was supported with a posterior probability of 1 and a maximum likelihood bootstrap value of 100. Support for phylogenetic relationships within each clade was poor.
Th e split between Harpagonella and outgroup sequences as well as the branches subtending varieties of Harpagonella palmeri were all supported by a number of shared nucleotide substitutions as well as insertion/deletions (indels). Th e Harpagonella-outgroup split was supported by 68 substitutions in the nuclear dataset, and 46 substitutions and 31 indels in the chloroplast dataset. Th e branch subtending the clade of var. arizonica samples was supported by 4 nucleotide substitutions in the nuclear dataset, and 1 substitution and 5 separate indels in the chloroplast dataset. Th e branch subtending the clade of var. palmeri samples was supported by 3 nucleotide substitutions in the nuclear dataset and 3 substitutions in the chloroplast dataset.

Discussion
Th e separate phylogenetic analyses of nrDNA and cpDNA presented here each recover two clades within H. palmeri corresponding to the two named varieties. Statistical support for these groupings was very high, with posterior probabilities above 0.96 and maximum likelihood bootstrap values of 100 in all cases. Th e Harpagonella-outgroup split as well as clades of samples by variety were each supported by numerous nucleotide substitutions and indels. We take this as strong evidence for two evolutionary lineages in the genus. Morphologically, these two lineages diff er in all measured aspects of fruit size. Plants primarily from Arizona and Sonora are signifi cantly larger in maximum fruit length, maximum fruit width, and appendage length. Box and whisker plots for these features show that the ranges of measurements of these characters between the two lineages are mostly non-overlapping. Although unmeasured here, nutlet size in Harpagonella was suggested by Johnston (1924) to be larger in plants from Arizona and Sonora than in plants from California and Baja California. Th ese diff erences are quantitative, not qualitative, and absent a formal statistical analysis of morphology, Veno (1979) advocated for recognizing no infraspecifi c taxa in H. palmeri, stating that "this feature is variable and somewhat clinal, and does not provide a signifi cant or reliable basis for taxonomic delimitation." Th e data presented here suggest instead that these quantitative characters appear to be suffi cient for reliable delimitation of two taxa corresponding to the evolutionary lineages recovered in the phylogenetic analysis.
Herbarium study of 366 specimens representing 291 gatherings of Harpagonella has permitted the evaluation of the geographic range of these morphologically distinct evolutionary lineages, which is especially critical for specimens collected on the Baja California Peninsula, where both named varieties have been reported. Specimens of plants with larger fruits corresponding to Johnston's H. palmeri var. arizonica are almost entirely from Arizona and Sonora, with two collections attributable to this taxon made from desert regions of Baja California at mid-peninsula (Moran 12682, 28.29007, -113.12146;Moran 12845, 28.28333, -113.65). We have observed and confi rmed the taxonomic identity of a specimen of the former (DS598325) but not the latter. Specimens of plants with smaller fruits corresponding to Johnston's concept for H. palmeri var. palmeri are known primarily from southwestern California and the adjacent western coastal areas of the Baja California Peninsula, with collections ranging as far to the south as the Vizcaino Peninsula on the Pacifi c Coast in Baja California Sur.
Th e biogeographic pattern displayed by Harpagonella -a disjunction between the California Floristic Province sensu Howell (1957) and central, southern Arizona and adjacent Sonora -is somewhat common yet underexplored. Raven and Axelrod (1978) describe this pattern briefl y in their important paper on the origin of the California fl ora, and provide a table of 35 genera, species, or species pairs that have this pattern. To their list of taxa, we add Harpagonella based on evidence presented here.

Taxonomic treatment
Based on complete and well-supported reciprocal monophyly in two unlinked genomic partitions, statistically signifi cant morphological diff erences, and essentially nonoverlapping geographic ranges, the two lineages of Harpagonella resolved here merit recognition at the species level under the criteria of phylogenetic species concepts (see Mishler and Th eriot 2000) as well as longstanding taxonomic practice. To recognize a taxon at species rank for the large-fruited plants found primarily in the deserts of Arizona and Sonora, the following new combination is needed.