A new species endemic to Hawaiʻi Island, Tetramolopium stemmermanniae, is described and illustrated. Molecular and morphological evidence support Tetramolopium stemmermanniae as being distinct from Tetramolopium arenarium var. arenarium, Tetramolopium consanguineum ssp. leptophyllum, and Tetramolopium humile ssp. humile, which occur at Pōhakuloa Training Area, Hawaiʻi Island. Tetramolopium stemmermanniae shares an upright and multibranched habit as that of Tetramolopium arenarium var. arenarium and Tetramolopium consanguineum ssp. leptophyllum. It differs in ray and disc flower color and number, and in having an open, paniculate inflorescence. We provide a description of the new taxon, include a key to the Tetramolopium species of Hawaiʻi, and a brief description of the habitat where it occurs.

Plant specimens were collected from 10 individuals of the new form of Tetramolopium at Pōhakuloa Training Area, (Fig. 1)  from 2009–2020. Five collections for type specimens were made from one locality in 2020 because of the small number of individuals at the other locations. Collected specimens were compared with voucher specimens of other Tetramolopium taxa held at the Bishop Museum Herbarium (BISH). Voucher specimens of field-collected plants are deposited at Bishop Museum Herbarium, United States National Herbarium at the Smithsonian Institution (US), Colorado State University Center for Environmental Management of Military Lands Herbarium (CMML) at Colorado State University, and Pōhakuloa Training Area Herbarium (PTA). All material was legally collected with appropriate permits.

Plant leaf tissue of 68 individuals from three Tetramolopium species found at PTA was collected into silica gel in August and September 2014 for a molecular population study. The collections include 32 individuals of the new form representing six sites, 9 individuals of T. arenarium representing three sites, and 27 individuals of T. consanguineum representing eight sites.

Seven Tetramolopium species (total of 14 samples from the Hawaiian Islands including four samples used in population study) were obtained for the phylogenetic analysis; included two herbarium samples contributed by Dr. Timothy Lowrey; Keysseria erici (Forbes) Cabrera for use as an outgroup taxon in the phylogenetic work (Appendix 1). DNA was extracted from all samples and purified following previously established laboratory protocols described in Morden et al. (1996). Each sample was assigned an accession number in the Hawaiian Plant DNA Library (HPDL; Morden et al. 1996; Randell and Morden 1999; http://www.botany.hawaii.edu/hawaiian-plant-dna-library; Appendix 1). The concentration and quality of the extracted DNA were assessed using a Nano Drop Spectrophotometer (ND-1000, ver. 3.6.0, Thermo Fisher Scientific, Waltham, Massachusetts). All DNA samples were diluted to 10–15 ng/μl and stored at -20°C until used.

Sequence-related amplified polymorphism (SRAP) markers were utilized to investigate genetic variation within and among populations of T. arenarium, T. consanguineum and the new form of Tetramolopium (Li and Quiros 2001; Robarts and Wolfe 2014; Zagorcheva et al. 2020). DNA from one to a few individuals of each species were surveyed with 100 combinations of 10 forward and 10 reverse SRAP primers (Budak et al. 2004). SRAP analyses were conducted using a 15 μl PCR reaction mixture consisting of: 1xPCR buffer [10 mM Tris-HCl (pH 9.0 at 25°C), 50 mM KCl and 0.1% Triton X-100, Promega, Madison, Wisconsin], 1.5 mM MgCl₂, 0.25 mg BSA, 0.2 mM dNTPs, 0.5 mM of each forward and reverse primers (IDT, Coralville, Iowa), 1 unit of Taq DNA polymerase (Promega), and approximately 10–15 ng of total DNA. All reactions were carried out in an MJ Research (GMI, Inc. Ramsey, Minnesota) or Eppendorf thermal cycler (Eppendorf AG, Hamburg, Germany) with cycling conditions following that established by Li and Quiros (2001). PCR amplified products were mixed with loading dye and separated on a 2% agarose gel, stained with ethidium bromide (EtBr), and visualized with a UV light source. Negative control reactions were run without DNA for all PCR amplifications to ensure reaction components were uncontaminated. Each primer combination was repeated at least once with selected samples to confirm the reproducibility of the genetic markers. Size of amplification products was estimated using either a 100 bp ladder (Promega) or a pBS plasmid (Stratgene, La Jolla, California) digested with restriction enzymes to produce fragments in a size range from 0.448–2.96 Kb. Final gel products were viewed using a Gel Doc XR (BIO-RAD, Hercules, California) and digitally recorded using Quantity One software (BIO-RAD, ver. 4.5.1). SRAP markers were scored either present (1) or absent (0). The scored data were entered into a binary matrix and assessed for polymorphic, unique loci. From this observed and expected heterozygosity were calculated using GenAlEx6.502 (Peakall and Smouse 2006, 2012). Principal coordinates analysis (PCO) using Gower general similarity coefficients (Gower 1971) were calculated using MVSP 3.1 (Kovach 2007). Pairwise similarities were averaged for individuals within and among species. The SRAP binary matrix was also used to resolve an unrooted neighbor joining (NJ) tree from a mean Euclidean distance matrix with 1,000 bootstrap replicates to assess clade support using DARwin v6.0.021 (Perrier et al. 2003). 

DNA Sequence variation was analyzed for four nuclear rDNA regions and five chloroplast DNA regions. The non-coding nuclear regions included rDNA ITS (Baldwin 1992), ETS (Baldwin and Markos 1998), 5S-NTS (Cox et al. 1992; Sastri et al. 1992), and nuclear NIA (Levin et al. 2009). Chloroplast loci included the rpl16 intron (Shaw et al. 2005), the psbA–trnH spacer (Shaw et al. 2005), trnT-L spacer, trnL intron, and the trnL-F spacer (Taberlet et al. 1991). Samples were PCR amplified in 25 µl volumes with the following components: 25 ng of DNA, ca. 0.2 mM each of dATP, dCTP, dGTP, dTTP, 1X Taq Polymerase buffer (10 mM Tris-HCl [pH 9.0 at 25°C], 50 mM KCl, and 0.1% Triton X-100 [Promega]), 1.5 mM MgCl₂, 0.50 mg BSA, 0.2 mM forward and reverse primers, and ca. 1 unit of Taq DNA Polymerase (Promega). PCR reactions were performed using thermocycling conditions outlined in the original primer references. Positive PCR amplification was confirmed and digitally recorded as described above.

The PCR products were purified using an Exo-Sap-It kit (Affymetrix, Santa Clara, California) according to the manufacturer’s instructions. PCR products were bidirectionally Sanger sequenced using amplification primers on an Applied Biosystems 3730XL DNA Analyzer (Applied Biosystems, Waltham, Massachusetts, USA) at the University of Hawaii’s facility for Advanced Studies in Genomics, Proteomics, and Bioinformatics.

Sanger sequence results were edited and contiged using Sequencher® ver. 5.0 sequence analysis software (Gene Codes Corporation, Ann Arbor, Michigan) and aligned using MEGA6.06-mac (Tamura et al. 2013) using ClustalW with default parameters. Aligned sequences were converted by Mesquite ver. 3.03 (Maddison and Maddison 2015) to Nexus format for use in phylogenetic analyses. Phylogenetic Bayesian analysis was conducted using MrBayes on XSEDE (3.2.6) (Ronquist et al. 2012) from the CIPRES Portal (Miller et al. 2010). The GRT+I+G model was chosen for partitioned and combined data sets as proposed by Abadi et al. (2019) using four heated four chains run for 10 million generations with sampling every 1000 generations and the first 25% of trees discarded as burn-in. Branch support was evaluated by posterior probability. PAUP* v. 4.0b 10 (Altivec; Swofford 2002) was used for parsimony analysis using a branch-and-bound search option and bootstrap resampling (1000 pseudoreplicates) to calculate branch support (Felsenstein 1985). An additional set of phylogenetic analyses were conducted using a concatenated alignment containing only perfectly homologous and parsimony informative loci to ensure lineage-specific loci were not biasing the results. The trimmed matrix was analyzed using the BI procedures described above as well as an NJ method using the Tamura-Nei genetic distance model (Tamura and Nei 1993) and 1,000 jack knife replicates to assess clade support.

Model Finder (Kalyaanamoorthy et al. 2017) and IQ-Tree (Version 2.2.0) (Nguyen et al. 2015) (www.igtree.ort);(http://iqtree.cibiv.univie.ac.at) were used to determine the best-fit model and to construct a maximum-likelihood tree for each locus respectively. Ultrafast bootstrap approximation (UFBoot) (Hoang et al. 2018) was used to obtain branch support values. Trees were visualized using FigTree (v. 1.4.4) (http://Tree.bio.ed.ac.ul/software/figtree/).

Programs that can access .xls and .tif files.