The heterogeneous landscape of western North America has been shaped by dynamic geological and climatic processes, including surface uplift (Foster et al. 2010), glaciation cycles (Thorne 1993), erosion (Cather et al. 2012), aridification (Stebbins 1952;Axelrod 1972), and the formation of biologically restrictive soils (e.g., serpentine, gypsum, limestone, deep sand; Kruckeberg 2004;Moore et al. 2014), all of which have been hypothesized to promote diversification of flowering plants within this region. As the western North American landscape became increasingly heterogeneous with isolated habitats during the Miocene (23.03-5.33 Ma), newly-opened areas of niche space promoted the rapid diversification of endemic genera (Qian et al. 2024), particularly in arid and semi-arid regions (Axelrod 1958). The Great Basin, for example, serves as a center of diversity for Eriogonum Michx. (Polygonaceae; ca. 250 spp.) and Penstemon Schmidel (Plantaginaceae; ca. 270 spp.), the two largest angiosperm genera endemic to North America (Reveal 1978;Wolfe et al. 2006), both of which underwent rapid diversification beginning 6 Ma and 2.5 Ma, respectively (Kostikova et al. 2014;Stone and Wolfe 2021;Wolfe et al. 2021).
Indeed, many of the rapid species radiations in western North America have occurred relatively recently (i.e., within the last $10 Ma), including some of the region's most recognizable and charismatic genera: Arctostaphylos Adans. (Ericaceae; Wahlert et al. 2009), Atriplex L. (Amaranthaceae; Brignone et al. 2019), Castilleja Mutis ex L.f. (Orobanchaceae; Tank and Olmstead 2008), Ceanothus L. (Rhamnaceae; Burge et al. 2011), Eriogonum (Polygonaceae; Kostikova et al. 2014) (Gilbert et al. 2005), Polemonium L. (Worley et al. 2009), and Yucca L. (Asparagaceae; Pellmyr et al. 2007). While these lineages are compelling, their biology introduces major challenges to taxonomy and phylogenetic reconstruction, including cryptic speciation, low morphological variability, limited genetic variation, conflicting phylogenetic signal (cytonuclear discordance, homoplasy, ancient and recent hybridization, introgression), and high rates of incomplete lineage sorting (ILS). Mentzelia section Bartonia (Torrey & A. Gray) Bentham & Hooker f. is no exception to these taxonomic and phylogenetic challenges, however, a recent monograph by Schenk and Hufford (2020) has provided a deeper understanding of species diversity compared with other groups, making it more amenable to phylogenomic studies of taxonomically well-understood species.
Mentzelia section Bartonia is a clade of 51 extant species (Schenk and Hufford 2020) that diversified relatively recently (4.9-9.2 Ma crown age; Schenk et al. 2025) out of the Colorado Plateau and expanded throughout the western United States and northern Mexico (Schenk 2013a). The plants are herbaceous biennials or short-lived perennials with actinomorphic flowers and non-stinging trichomes used in herbivory defense (Eisner et al. 1998;Schenk and Hufford 2020). Much of the clade's species diversification was driven by movement onto nutrient-poor and disturbed soils, such as talus slopes or gypsum and limestone outcrops (Schenk 2013a;Schenk and Hufford 2020). Shifts in aneuploidy also commonly coincided with diversification (Schenk and Hufford 2020). Species are diploid with a base chromosome count of either n 5 10 or n 5 11 (Schenk and Hufford 2020), however, there is contention over whether M. leucophylla Brandegee is a polyploid (n 5 18; Reveal and Styer 1973;Schenk and Hufford 2020). Most extant species are narrowly distributed endemics confined to a single floristic province (Schenk 2013a), with 31 out of 51 species currently listed as either threatened or endangered by state or federal agencies (Schenk and Hufford 2020). Like many other recently and rapidly radiating groups in western North America, Mentzelia section Bartonia is in considerable need of a robust phylogeny to inform future conservation efforts. Much like those other groups, however, taxonomic and phylogenetic work in Mentzelia section Bartonia has been challenging.
The first molecular phylogenetic study to include section Bartonia was Hufford et al. (2003) who sampled 12 out of 51 species as part of a larger study of Loasaceae. Using two chloroplast markers (matK and trnL-trnF), the authors found little resolution among species, with 11 out of 12 species belonging to a single large polytomy, demonstrating that a limited number of chloroplast markers are insufficient for phylogenetic inference of the section. Subsequently, Schenk and Hufford (2011) reconstructed the first complete molecular phylogeny of section Bartonia using two nuclear ribosomal (nrDNA) regions (ITS and ETS), which was later updated to include additional sampling (Schenk and Hufford 2020). While the Schenk and Hufford (2011) phylogeny offered significant insights into species relationships and informed ancestral character estimations of reproductive traits (Schenk 2013b;Botnaru and Schenk 2019) and biogeography (Schenk 2013a), the tree had 45% unresolved relationships with 62% of all relationships being weakly supported (Bayesian posterior probabilities , 0.95).
High throughput sequencing methods have not necessarily been more effective at equivocally resolving relationships. Moore et al. (2023) implemented genome skimming of plastid genomes and the nuclear ribosomal cistron region for 20 out of 51 species and found low phylogenetic signal and high levels of homoplasy. One striking result of the study was that out of $159,000 bp recovered for the chloroplast genome, only 76 sites were phylogenetically informative, illuminating how little genetic variation exists amongst species. To address these limitations, Cohen and Schenk (2022) used ddRADseq to conduct a population-level study of five species endemic to the Mojave Desert. While their approach obtained resolution among tips, their study also revealed evidence of putative hybridization and ILS.
While high-throughput sequencing has enhanced our ability to detect genetic variation, the abundance of genetic data introduces further complications (Moore et al. 2023). Incomplete lineage sorting and paralogy pose substantial challenges for phylogenetic inference of recently and rapidly diversifying lineages. The likelihood of ILS, which is characterized by the persistence of ancestral polymorphisms after speciation events (Maddison 1997), increases with younger lineages, resulting in incongruent gene tree topologies and diminished support values in species trees. Gene duplications in large genomic data sets also obscure homologous relationships that are essential for accurate phylogenetic inference. Paralogs are genes that evolve from duplication events within a lineage and do not coincide with duplication events during speciation. In some situations, a random paralogous gene may be lost, resulting in a single-copy gene that can be mistaken for an ortholog. Leaving these "pseudoorthologs" in a data set can lead to an overestimation of branch lengths and discordant gene tree topologies (Smith and Hahn 2022). In addition, recently and rapidly evolved lineages have not had time for species to accrue synapomorphies before speciation events occurred, hindering our ability to detect phylogenetic signal (Moore et al. 2023). Therefore, we must implement methods that maximize our ability to detect phylogenetic signal while minimizing common issues that confound relationships, such as paralogy and ILS.
Based on the results of Cohen and Schenk (2022), a robust phylogeny of Mentzelia section Bartonia subclades using RADseq is plausible, however, we chose to implement a HybSeq approach for four reasons. First, RADSeq and HybSeq methods can find congruent phylogenies and divergence times (e.g., Zhou and Xiang 2022). Second, HybSeq can outperform RADSeq when DNA quality and concentration are low (Zhou and Xiang 2022), which is often true for genomic extractants of Mentzelia section Bartonia because of secondary metabolite contamination (Carey et al. 2023). Third, RADSeq approaches often fail to recover orthologous SNPs among distantly related species (Leach e and Oaks 2017), and we aim for the compatibility of our data with future studies of older lineages, including Mentzelia (70-35.3 Ma;Schenk and Hufford 2010), Loasaceae (91.60-57.96 Ma;Schenk and Hufford 2010), and tree of life projects (e.g., Baker et al. 2022). Finally, the recent development of the Angiosperms353 probe set (Johnson et al. 2019) provides a relatively easy and costeffective HybSeq method that can be as equally phylogenetically informative as lineage-specific probes (Ufimov et al. 2021). Angiosperms353 probes have been successfully implemented in recent and rapidly radiating lineages outside of western North America, including Burmeistera H. Karst. & Triana (Campanulaceae;Bagley et al. 2020), Cyperus L. (Cyperaceae; Larridon et al. 2020), Tillandsia L. (Bromeliaceae; Yardeni et al. 2022), and Veronica L. (Plantaginaceae; Thomas et al. 2021). Taken together, HybSeq with Angios-perms353 is a promising method for resolving species relationships in Mentzelia section Bartonia.
The goal of the present study is to elucidate species relationships in Mentzelia section Bartonia, a recently and rapidly evolving angiosperm clade from western North America. To achieve our goal, we applied a HybSeq approach with Angiosperms353 using a novel bioinformatics pipeline that aimed to minimize noise in our data while maximizing phylogenetic signal.
To robustly resolve phylogenetic relationships among species of Mentzelia section Bartonia, we employed methods that address the myriad challenges of recent and rapidly radiating lineages, including having a low phylogenetic signal-to-noise ratio, a high chance of ILS, and the potential for hybridization.
Sample Preparation and Sequencing-We sampled 50 of 51 species in Mentzelia section Bartonia, including five infraspecific taxa, and four of six species from its sister clade, section Bicuspidaria S.Watson (Hufford et al. 2003;Brokaw et al. 2020). Most ingroup taxa were represented by a single individual; however, eleven taxa were represented by up to three individuals to reflect both phenotypic and geographic variation. We prioritized sampling the same vouchers as Schenk andHufford (2011, 2020) to make direct comparisons of our results to theirs. The complete list of 76 total vouchers is detailed in Appendix 1.
Genomic DNA was obtained from previously published work (Schenk andHufford 2011, 2020;Brokaw et al. 2020;Moore et al. 2023) and by fresh extraction of 5-10 mg of silica-dried leaves or those preserved as herbarium specimens using a modified CTAB protocol (Schenk et al. 2023, protocol v4). Genomic DNA was quantified with gel electrophoresis and a Qubit fluorometer (Thermo Fisher Scientific, Waltham, Massachusetts, USA). Per sample, $500 ng of genomic DNA was used as input for genomic libraries that were constructed with the NEBNext Ultra II FS DNA Prep Kit for SYSTEMATIC BOTANY 68 [Volume 50 Illumina (New England BioLabs, Ipswich, Massachusetts, USA), which also enzymatically fragmented DNA to 200-450 bp fragments, following manufacturer instructions for inputs $ 100 ng. Size selection was performed for a targeted library size of $300 bp. Fragment analyses of the prepared libraries were performed with a TapeStation (Agilent Technologies, Santa Clara, CA, USA). Libraries were indexed with NEBNext Multiplex Oligos for Illumina (New England BioLabs, Ipswich, Massachusetts, USA) and either enriched with Angiosperms353 v5 probe set (myBaits, Daicel Arbor Biosciences, Ann Arbor, MI, USA; Johnson et al. 2019) or left unenriched to increase sequence complexity during the sequencing reaction. Hybrid enrichment was performed following the manufacturer's standard protocol for plants with the following modifications: hybridization temperature was 62 C, library amplification ran for 14 PCR cycles, and wash temperature was 65 C. Enriched and unenriched libraries were pooled to a final concentration of 2.25 nM in 220 ml in accordance with sequencing facility recommendations and sequenced together at a 1:10 ratio of unenriched to enriched libraries.
We performed three separate runs of Illumina sequencing: Illumina NovaSeq 6000 (San Diego, California, USA) on a single lane with a target length of 250-bp paired-end reads (Nationwide Children's Hospital, Columbus, OH, USA), Illumina MiSeq on two lanes with a target length of 300-bp paired-end reads (Psomagen Inc., Rockville, MD, USA), and Illumina NextSeq 2000 on two lanes with a target length of 250-bp pairedend reads (Nationwide Children's Hospital, Columbus, OH, USA). Both enriched and unenriched libraries were simultaneously sequenced. MiSeq and NextSeq runs were initially performed as exploratory methods for a small proportion of samples before we ultimately chose NovaSeq for the remainder of our sequencing. Identical vouchers that were sequenced on more than one platform had their raw reads bioinformatically combined prior to locus assembly.
Phylogenetic Inference-The complete bioinformatics pipeline is illustrated in Fig. 1. Select data and command-line code is available on Dryad (Fabre et al. 2025).
Quality control of raw, paired-end reads was performed with fastp v0.23.4 (Chen et al. 2018) using the following adjustments to default parameters: minimum qualifying phred score per base $ 20, minimum read length 5 50, maximum number of unqualifying bases per read 5 10, and removal of PCR duplicates (fastp code: -q 20 -n 10 -l 50 -D). We used fastp over Trimmomatic (Bolger et al. 2014) to remove homodimers from the ends of Illumina NextSeq and Illumina NovaSeq reads (https://github. com/mossmatters/HybPiper/wiki). The post-fastp reads are hereafter called "clean" reads.
HybPiper v2.1.6 (Johnson et al. 2016) was used with BWA (Li and Durbin 2009) to map clean, paired-end reads to a customized nucleotide target file and assemble supercontigs (i.e., exons with flanking introns). Our target file contained Cornales sequences filtered from Mega353 v1 (McLay et al. 2021) in addition to published exons from Mentzelia decapetala (Pursh) Urb. and Mentzelia involucrata S.Watson (https://treeoflife.kew.org), which increased our recovery rates in preliminary analyses. We checked the target file for low-complexity regions using the "check_targetfile" function in HybPiper with default settings and removed the offending sequences before locus assembly. After locus assembly, we used the "stats" command to summarize gene recovery statistics, and the "paralog_retriever" command to summarize putative paralog recovery and retrieve paralogous sequences.
Although the Angiosperms353 probes were designed to capture singlecopy nuclear loci (Johnson et al. 2019), they can still recover paralogs, which HybPiper (Johnson et al. 2016) (the most popular pipeline for assembly of Angiosperms353 loci) can fail to detect (Nauheimer et al. 2021;Zhou et al. 2022). In an exploratory analysis, we reviewed the paralogs recovered by HybPiper, selected a locus with only one specimen flagged as paralogous, and assembled a new target file that contained only the paralogous sequences from that specimen. When we re-assembled the locus based on the paralog-only target file, we discovered additional paralogous copies that were not initially reported. Based on this finding, we decided against removing individual paralogous copies and instead conservatively removed entire loci from downstream analyses if they were flagged with one or more paralogous sequences.
After supercontig assembly with HybPiper, we ran HybPhaser v2.1 (Nauheimer et al. 2021) to evaluate genomic samples for additional hiddenparalogy, hybridization, and contamination. In short, HybPhaser uses the supercontigs generated by HybPiper as a reference to re-map clean reads, discover SNPs, and remove low-quality or outlying individuals and loci.
Raw, pairedend reads from Illumina Remove paralogous and lowquality samples/loci with HybPhaser Quality control of raw reads with fastp Assemble supercontigs with HybPiper Find and assess SNPs with HybPhaser Align orthologous consensus sequences with MAFFT Remove alignment outliers with SpruceUp Infer gene trees with IQ-TREE Obtaining quality sequences Inferring gene trees Concatenate loci with AMAS Partition alignment with AMAS Hypothesis testing of hybrids with HyDe Calculate gene tree and site discordance with IQ-TREE Calculate gene tree statistics with SortaDate Detecting Discordance and Hybridization Collapse branches with < 10% BS support with newick_utils Infer species tree with ASTRAL-III (75 loci) Remove all loci flagged with paralogs from both HybPiper and HybPhaser Inferring Species trees Root gene and species trees with Phyx Re-align loci; continue to IQ-TREE Filter loci by > average bipartition support Infer species tree with ASTRAL-III (238 loci) Collapse branches with < 10% BS support with newick_utils Collapse branches with < 10% BS support with newick_utils Re-infer gene trees; continue to SortaDate HybPhaser consensus sequences Remove sequences with < 25% mean recovered length with filter_by_length.py Infer species tree with ASTRAL-III (108 loci) Remove long branches with TreeShrink Concatenate alignments with AMAS
1. Bioinformatics pipeline that was implemented in our study to minimize noise while maximizing phylogenetic signal. To maximize the low signal-to-noise ratio present in recently and rapidly evolving lineages, we implemented a rigorous bioinformatics pipeline to address paralogy, noise from low-quality sequences, species, and loci, incomplete lineage sorting, and hybridization. Bootstrap support is abbreviated as BS and single nucleotide polymorphisms are abbreviated as SNPs.
HybPhaser flags both samples and loci as statistical outliers if they have a relatively high proportion of SNPs (i.e., more than 1.5 Ã IQR [interquartile range] above the 3rd quartile of mean), indicating putative paralogy. Loci and samples flagged as statistical outliers/paralogs were removed from downstream analyses. Loci that contained # 15% of samples, and samples that were found in # 15% loci, were also removed. When completed, Hyb-Phaser generated consensus sequences with ambiguity codes for heterozygous sites, which reduces conflicting phylogenetic signal from hybrids (Nauheimer et al. 2021).
To reduce the deleterious effects of fragmentary data on downstream analyses of gene and species trees (Sayyari et al. 2017;Mirarab 2019), we ran the Python script filter_by_length.py (https://github.com/ mossmatters/phyloscripts/tree/master/HybPiperUtils) on the HybPhaser "consensus_length" file to flag samples with sequences shorter than 25% of the mean length recovered for a particular locus. To do this, we modified the "consensus_length" file to match the format of the HybPiper "seq_lengths" file and calculated the mean length of recovered sequences (i.e., sequences with recovered length . 0) for each locus. We used the mean recovered length rather than the default mean targeted length to ensure that loci with low recovery across all taxa were still retained. Samples that were flagged as being too short per locus were then removed from the data set.
Putative orthologous consensus sequences were aligned with MAFFT v7.520 (Katoh and Standley 2013) using 1000 maximum iterations. Resulting gene alignments were concatenated with AMAS v1.0 (Borowiec 2016) and analyzed with spruceup v2022.2.4 (Borowiec 2019), utilizing default settings to identify and remove outlier sequences (i.e., individual poorly aligned sequences or sequence fragments that were outside of the 95% confidence interval) using a preliminary ASTRAL (Zhang et al. 2018) tree as a guide. The preliminary tree was comprised of 247 putative orthologous supercontigs from HybPiper and inferred under identical conditions as described below. Following spruceup, the concatenated alignment was partitioned with AMAS (Borowiec 2016) utilizing the "remove-empty" command to remove samples that were characterized by only gaps or ambiguous nucleotides. To reduce regions with excessive gaps, gene alignments were re-aligned with MAFFT under identical conditions outlined above.
After the above conditions were met, individual-locus alignments were used to infer maximum-likelihood gene trees with 1000 ultrafast bootstrap approximations in IQ-TREE v2.2.2.3 (Nguyen et al. 2015;Minh et al. 2022) using the ModelFinder command (Kalyaanamoorthy et al. 2017) to determine the best-fit substitution-model for each locus. Unusually long branches were detected with TreeShrink (Mai and Mirarab 2018) and offending tips were removed from gene trees and corresponding alignments for the offending locus. To re-estimate bootstrap support after TreeShrink, gene trees were then re-estimated with IQ-TREE under identical conditions as above. Gene tree nodes with , 10% bootstrap support were collapsed with newick_utils v1.6 (Junier and Zdobnov 2010;Mirarab 2019), and resulting gene trees were used to infer a multispecies coalescent tree with ASTRAL-III v5.7.8 (Minh et al. 2022) utilizing default parameters. We did not implement a concatenation-based approach because it underperforms in the face of ILS and is therefore an inferior method for groups that have undergone recent and rapid speciation (Mirarab 2019). The final ASTRAL tree was visualized in FigTree v1.4.4 (http://tree.bio. ed.ac.uk/).
Discordance and Hybridization-To evaluate the consistency among gene trees, we rooted the gene trees using pxrr in Phyx (Brown et al. 2017) and used SortaDate (Smith et al. 2018) to calculate tip-to-root variation, tree length, and bipartition support. Tip-to-root variation indicates clocklikeness, tree length is the total sum of branch lengths, and bipartition support measures the similarity between the gene trees and species trees (Smith et al. 2018). Bipartition support has also been used as a metric to remove paralogs that are otherwise only detectable by eye (Frost et al. 2024). Based on this, we subset our data to include only loci with higher than average bootstrap support (Frost et al. 2024; Fig. 1) to remove additional putative paralogs. Simultaneously, we also generated a data subset to include only loci that were not flagged as paralogous by either HybPiper or HybPhaser (Fig. 1). Alignment statistics, including alignment length, percent missing data, proportion of invariable sites, and proportion of parsimony informative sites, were calculated with AMAS (Borowiec 2016) for all three data sets.
To better understand gene tree discordance in our three data sets, gene concordance factors (gCF) and site concordance factors (sCF) were calculated with IQ-TREE (Minh et al. 2022), with the latter utilizing the Model-Finder function (Kalyaanamoorthy et al. 2017) and maximum likelihood (Mo et al. 2023) with 100 quartets. Gene concordance factors are the proportion of gene trees that are concordant with the species tree at a given node and signify the proportion of the genome that shares this evolutionary history (Baum 2007). Site concordance factors represent the proportion of alignment sites that are decisive for a particular branch (Minh et al. 2020).
Putative hybrids were detected by HybPhaser if they had a high proportion of loci containing SNPs (indicative of high locus heterozygosity [LH]) in conjunction with a high proportion of SNPs distributed across all loci (indicative of high allele divergence [AD]). We designated taxa as putative hybrids worthy of further exploration if they (i) had a high LH/AD ratio based on HybPhaser analyses, (ii) were recovered as nonmonophyletic in one or more of our ASTRAL trees, or (iii) had discordant relationships between our ASTRAL trees. Using only the data subset containing loci with above-average bipartition support, we then tested explicit hypotheses of hybridization with HyDe v0.4.3 (Blischak et al. 2018;
We produced genomic sequence data for 77 individuals, comprising 96% of taxa in Mentzelia section Bartonia and 67% of taxa in section Bicuspidaria. In total, we recovered 498 billion clean, paired-end reads, ranging from 23 thousand to 26 million reads per sample (median 5 5.4 million; Table S1).
HybPiper recovered sequence data for 351 total loci with an average of 336 loci per specimen (Table S1). HybPiper flagged 104 loci as containing at least one paralog (Table S1). Based on both HybPiper and HybPhaser analyses, no individuals recovered entirely paralogous loci, suggesting there are no allopolyploids in our data set (including M. leucophylla, the only hypothesized polyploid). HybPhaser failed to detect SNPs for 20 individuals; however, of the 57 individuals with SNP data, 236 loci were flagged as containing at least one individual with a paralog (Table S1), and 16 loci were flagged as paralogous across all individuals. Based on quality filtering by HybPhaser, two loci were removed from the data set for containing fewer than 15% of samples, and one specimen (M. canyonensis) was removed from the data set completely for being recovered in fewer than 15% of loci (Table S1).
Our most inclusive data set contained 238 loci, the data set based on above-average bipartition support contained 108 loci, and our strict ortholog data set contained 75 loci (Table S2). On average, the three datasets exhibited similar proportions of missing characters and variable sites; however, the 75-locus dataset contained 70.8% fewer parsimony informative sites than the 238-locus dataset and 38.5% fewer than the 108-locus dataset (Table S2). In general, support values decreased with the reduction of loci, but the major relationships were consistent across all three phylogenies (Fig. 2; Figs. S1, S2). The greatest relationship differences we observed between the phylogenies included the placement of M. decapetala, M. pumila Nutt., M. densa Greene, M. springeri (Standl.) Tidestr., and M. longiloba var. chihuahuaensis J.J.Schenk & L.Hufford. We will base the remaining Results and Discussion on the above-average bipartition data set of 108 loci unless noted otherwise. We focus on the 108-locus data set to maintain a balance between retaining phylogenetic signal and removing noise associated with paralogy.
Each locus contained an average of 69 taxa and had an average alignment length of 6397 bp (Table S2). A concatenated alignment was 690,863 bp long with 60% undetermined FABRE ET AL.: PHYLOGENOMICS OF MENTZELIA SECTION BARTONIA 71 2025]
with varying levels of support (Fig. 3). The largest clade was supported by a local posterior probability (LPP) of 0.81 with nodes along the backbone averaging an LPP of 0.56, a gCF of 0.51, and an sCF of 33.53 (Fig. 3). In contrast, the second-largest lineage was supported by an LPP of 1 with nodes along the backbone averaging an LPP of 0.76, a gCF of 4.93, and an sCF of 41.75 (Fig. 3). The monotypic lineage comprised of M. decapetala was supported by an LPP of 0.51, a gCF of 4.4, and an sCF of 43.93 (Fig. 3). In general, support values tended to increase towards the tips (Fig. 3). Only 7% of total relationships across
1/99.1/76.19 0.36/0/36.59 1/59.78/85.48 0.72/0/30.46 0.66/28.28/32.65 0.59/2.68/29.07 0.72/7.21/27.23 1/99.1/76.19 0.62/6/31.31 1/56.52/52.61 0.62/5.1/26.09 1/39.6/23.52 1/56.76/48.1 1/73.81/81.42 0.98/14.29/46.39 0.98/7.21/40.5 0.8/4.55/43.95 0.65/0/42.23 0.59/0.91/44.08 0.63/3.64/33.95 0.58/0/28.3 0.94/6.86/42.02 0.81/2.68/29.56 0.47/0/44.06 1/17.59/45.12 1/45.92/68.48 0.7/0/44.33 0.31/1.04/33.31 0.96/11.34/46.84 0.75/16.67/51.09 0.88/3.08/35.35 1/69.09/32.43 1/12.04/41.85 0.46/21.35/24.52 0.87/18.37/76.85 0.69/26.74/20.64 0.94/15.38/25.18 0.54/3.64/48.29 0.91/48/27.72 1/42.11/68.74 0.42/2.63/29.75 0.51/4.4/43.93 0.52/9.52/43.52 0.42/0/17.51 1/47.42/59.2 0.55/25/20.76 0.55/0/38.67 0.54/0/27.94 1/87.16/92.6 1/13.33/40.4 0.44/7.69/30.81 1/39.45/40.74 1/14.29/33.88 0.4/0/33.96 0.55/11.88/31.01 1/78.3/26.94 0.48/21.57/46.16 71/3.39/33.63 0.9/24.75/31.29 0.88/39/49.27 0.58/11.58/36.07 0.97/45.45/46.35 0.63/49.48/43.94 0.85/8.47/28.41 0.97/4.5/39.06 1/10.99/55.45 0.44/0/32.19 1/91.89/59.93 0.78/17.65/49.07 0.65/6.48/34.91 0.79/0/41.87 0.62/0/23.66 0.91/1.64/34.67 0.6/8.49/23.29 subsection Multiflora subsection Multicaulis "Mojave clade" FIG. 3. Cladogram of Mentzelia section Bartonia based on an ASTRAL analysis of 108 nuclear loci. Colored boxes designate newly named subsections. Taxa are designated with a voucher number and place of origin in USA (county, state) or Mexico (state, country). Values at nodes indicate local posterior probability (LPP), gene concordance factors (gCF), and site concordance factors (sCF) and are in the order of LPP/gCF/sCF. SYSTEMATIC BOTANY 72 [Volume 50
the tree remain unresolved (Fig. 2). Of the 14 taxa that were sampled with multiple individuals, our analyses revealed six species that are not monophyletic as currently circumscribed: M. longiloba J.Darl, M. lagarosa (K.H.Thorne) J.J.Schenk & L.Hufford, M. perennis Wooton, M. oreophila J.Darl, M. speciosa Osterh., and M. saxicola H.J.Thomps. & Zavort (Figs. 2, 3). Allele divergence varied between 0.04% and 2.36% (avg. 0.58%) and LH varied between 8.17% and 99.7% (avg. 67%; Table S1). Three ingroup individuals had a high LH/AD ratio compared to other samples, indicating possible hybridization: M. longiloba var. longiloba (Schenk 1225), M. humilis var. humilis (Urb. & Gilg) J.Darl (Schenk 1264), and M. saxicola (Schenk 1256; Table S1). Individual hypothesis testing with HyDe inferred significant hybridization in M. longiloba var. longiloba 1225 with both M. procera (Wooton & Standl.) J.J.Schenk & L.Hufford (p , 0.001, g 5 0.50) and M. longiloba var. yavapaiensis J.J.Schenk & L.Hufford (p , 0.001, g 5 0.74), and in M. saxicola with both M. mexicana H.J.Thomps. & Zavort. (p , 0.001, g 5 0.71) and M. longiloba chihuahuaensis (p , 0.001, g 5 0.47; Table S3). No significant hybridization was detected in M. humilis var. humilis 1264 among any of the hypotheses we tested (Table S3).
Below, we propose to formally name subsections of Mentzelia section Bartonia. We will use these newly designated subsection names throughout the Discussion. Please see Schenk and Hufford (2020) for further details of the characters and distribution maps of the species.
Bartonia subsection Decapetala P.Fabre, J.Brokaw, L.Hufford, M.Johnson, & J.Schenk subsection nova-Mentzelia decapetala (Pursh) Urb., Ber. Deutsch. Bot. Ges. 10: 263. 1892. TYPE: UNITED STATES. On the banks of the Missouri, from the river Platt to the Andes, on arid volcanic soil, 1811, T. Nuttall s.n. (lectotype: Bot. Mag. 36: pl. 1487, 1812; by Reveal, Moulton & Schuyler, Proc. Acad. Nat. Sci. Philadelphia 149: 32. 1999). Biennial to short-lived perennials (as rosettes only). Subterranean caudex unbranched. Ovary five-seven carpellate. Stigma lobes short. Seed coat anticlinal cell walls straight. Mentzelia section Bartonia subsection Multicaulis P.Fabre, J.Brokaw, L.Hufford, M.Johnson, & J.Schenk subsection nova-Mentzelia multicaulis (Osterh.) J.Darl., Ann. Missouri Bot. Gard. 21: 156. 1934. TYPE: UNITED STATES. Colorado. Eagle Co.: Wolcott, 11 Jul 1902, G. E. Osterhout 2663 (lectotype: RM!; isolectotypes: CAS!, COLO! [2], NY!, POM! [2], RM!, RSA! [2]). Overwintering annual/biennial (in M. pterosperma only) or short-lived perennials. Subterranean caudex branched if perennial. Ovaries threecarpellate. Stigma lobes short. Seed coat anticlinal cell walls straight. Mentzelia section Bartonia subsection Multiflora P.Fabre, J.Brokaw, L.Hufford, M.Johnson, & J.Schenk subsection nova-Mentzelia multiflora (Nutt.) A.Gray, Mem. Amer. Acad. Arts ser. 2 [Pl. Fendl.] 4:48. 1849. TYPE: UNITED STATES. New Mexico. Santa Fe, sandy hills along the borders of the Rio del Norte, Aug 1841, W. Gambel s.n. (lectotype: K! [K000810481]). Biennials or short-lived perennials. Subterranean caudex unbranched. Ovaries three-carpellate. Stigma lobes long, slender, and tapering. Seed coat anticlinal cell walls wavy, sinuate, or straight.
Mentzelia section Bartonia is one of many lineages that have recently and rapidly diversified in western North America. To date, phylogenetic relationships of recently and rapidly radiating groups from western North America have mostly been investigated with Sanger sequencing (e.g., Gilbert et al. 2005;Drummond 2008;Brignone et al. 2019) or amplified fragment-length polymorphism (AFLP; e.g., Worley et al. 2009), but few have been investigated with high-throughput sequencing methods (e.g., Ottenlips et al. 2021;Wenzell et al. 2021;Overson et al. 2023). Although larger data sets might promise resolution among species trees over traditional sequencing, they do not necessarily resolve all relationships. Phylogenetic studies of Penstemon, for example, highlight the challenges of resolving evolutionary relationships in recently and rapidly radiating groups of western North America despite extensive data sets. While both RADseq (Wessinger et al. 2016(Wessinger et al. , 2019) ) and HybSeq with lineage-specific probes (Wolfe et al. 2021) recovered greater phylogenetic resolution over Sanger sequencing-based approaches in Penstemon, more recently diverged relationships remained unsupported due to substantial discordance caused by ILS, hybridization, and/or introgression (Wolfe et al. 2006).
Of the 14 aforementioned lineages that have recently and rapidly evolved in western North America, we are aware of only two that have been investigated with the Angiosperms353 probe set: Castilleja (Wenzell et al. 2021) and Lomatium (Ottenlips et al. 2021), in which both studies examined shallow-level relationships within species complexes. Wenzell et al. (2021) recovered too low genetic variation to implement tree-based methods, but Ottenlips et al. (2021) recovered enough variation to infer well-supported phylogenies that improved resolution over Sanger sequencing-based trees, although they still found high gene-tree discordance likely caused by ILS or hidden paralogy. Both Wenzell et al. (2021) and Ottenlips et al. (2021) did not rigorously address paralogy beyond removing those flagged by HybPiper. As we discovered in our exploratory HybPiper analyses and subsequent HybPhaser analyses (Table S1), HybPiper does not detect all paralogs, and data sets that remove loci flagged only by HybPiper likely maintain hidden paralogs that can obscure species relationships by increasing discordance and lowering support values (Smith and Hahn 2022).
To elucidate species relationships of Mentzelia section Bartonia, we implemented a hybrid-enrichment approach with the Angiosperms353 bait-set and inferred a phylogenetic tree with only 7% unresolved relationships. In contrast, Schenk and Hufford (2011), who utilized Sanger sequencing approaches, inferred a phylogeny with 45% unresolved relationships. We believe that our ability to resolve most phylogenetic relationships was due to our rigorous bioinformatics pipeline, especially regarding paralogs. In lineages with low phylogenetic signal, relationships are substantially impacted by paralogy, hybridization, and ILS. We detected paralogs because we looked for them in multiple steps of our bioinformatics pipeline (Fig. 1). We also further reduced conflicting FABRE ET AL.: PHYLOGENOMICS OF MENTZELIA SECTION BARTONIA 73 2025] phylogenetic signal by utilizing the HybPhaser consensus sequences to account for hybridization. Furthermore, we accounted for ILS by implementing a multispecies coalescent approach with ASTRAL-III rather than implementing a concatenated supermatrix approach that is insufficient for recently and rapidly radiating groups (Mirarab 2019). One consequence of rigorous paralog detection, however, is that fewer loci decrease support values and alter species tree topology among poorly supported taxa. A custom bait approach could have alleviated this problem by enabling the retention of more loci for phylogenetic analyses following our rigorous quality control steps.
By successfully resolving most relationships in Mentzelia section Bartonia, our study demonstrated that Angiosperms353 holds considerable promise for elucidating relationships in recently and rapidly radiating lineages of western North America, provided a rigorous bioinformatics pipeline is also implemented to account for paralogy, ILS, and other noise. The robust phylogenetic hypothesis of Mentzelia section Bartonia we reconstructed now allows us to explore patterns among species relationships at a deeper level than previous analyses allowed.
Subsectional Taxonomy-Across our data sets, we consistently recovered two well-supported major clades (Fig. 3) that were concordant with what Schenk andHufford (2011, 2020) called clade-1 (now Mentzelia subsection Multiflora) and clade-2 (now Mentzelia subsections Multicaulis and Decapetala). Both clades represent major groupings of species that are consistent with geographic distribution and morphological variation. Subsection Multiflora (clade-1) is especially diverse in the Chihuahuan, Great Plains, Rocky Mountains, and Sonoran floristic provinces (Schenk 2013a), while M. subsection Multicaulis is diverse in the Colorado Plateau, Great Basin, Sonoran Desert, and Mojave Desert. Mentzelia subsection Multicaulis also contains many species with highly limited ranges that are confined to biologically restricted soils, putting them at great risk of extinction (Schenk 2013a;Schenk and Hufford 2020). Most species in M. subsection Multicaulis have adaptations for steep slope habitats with loose substrates, such as having "pseudo-rhizomes" (i.e., aerial shoots that become buried) and a sub-shrubby habit (Schenk andHufford 2009, 2020), with a notable exception in the overwintering annual or biennial M. pterosperma Eastw. Mentzelia pterosperma is the only species in M. subsection Multicaulis that has converged on the M. subsection Multiflora form, but multiple studies (Schenk andHufford 2011, 2020; this study) confirm its position in M. subsection Multicaulis, and this is further supported by the geographic distribution of species (Schenk and Hufford 2020). Mentzelia decapetala was recovered as sister to the remaining members of clade-2 in some, but not all analyses (Fig. 3; Figs. S1, S2), however, the species is morphologically and geographically distinct from M. subsection Multicaulis. Based on the inconsistency of our results as well as the results of previous studies (Hufford et al. 2003;Schenk andHufford 2011, 2020), we recognize M. decapetala as its own clade, although it remains possible that the species might have evolved as sister to either subsections. Additional sampling across the distribution of M. decapetala is needed to fully understand its relationship with the rest of M. section Bartonia. Despite this, the distinct morphology and distribution of M. decapetala encourages us to recognize the species as a monotypic subsection. The recognition of the subsections as natural groups brings clarity to the diversity of Mentzelia section Bartonia, and for this reason, we propose to formally name them as subsections.
Mojave and Pinnatisect Clades-Phylogenomic studies of Mentzelia section Bartonia have been conducted with both RADSeq (Cohen and Schenk 2022) and genome skimming approaches (Cohen and Schenk 2022;Moore et al. 2023), which explored phylogenetic relationships among the "Mojave clade" and "pinnatisect clade," respectively. Below, we compare the relationships we recovered with Angiosperms353 to those recovered in previous studies.
MOJAVE CLADE-In Mentzelia subsection Multicaulis, Schenk and Hufford (2020) recovered a polytomy of seven species from in and around the Mojave Desert that they colloquially named the "Mojave clade" (their clade 2c), which included: M. canyonensis, M. hualapaiensis J.J.Schenk, W.C.Hodgs. & L.Hufford, M. leucophylla, M. oreophila, M. polita A.Nelson, M. puberula J.Darl, and M. tiehmii N.H.Holmgren & P.K.Holmgren. Cohen and Schenk (2022) brought resolution to this clade with RADseq and redefined the "Mojave clade" to include all the above species except M. hualapaiensis and M. canyonensis, which consistently grouped with M. memorabilis N.H.Holmgren & P.K.Holmgren in a sister clade that the authors referred to as the "Colorado Plateau clade." Consistent with Cohen and Schenk (2022), we recovered a monophyletic "Mojave clade" in all our analyses (Fig. 3; Figs. S1, S2). We did not, however, recover a "Colorado Plateau clade" in any of our analyses (Fig. 3; Figs. S1, S2), although this lack of resolution may be due to the absence of M. canyonensis in our datasets. Like Cohen and Schenk (2022), the relative placement of M. pterosperma Eastw. to the Mojave and Colorado Plateau clades was also inconsistent.
We recovered a polyphyletic M. oreophila that is consistent with the findings of both Cohen and Schenk (2022) and Schenk andHufford (2011, 2022). We sampled both Nevada and California populations of M. oreophila, including the type locality of Inyo County, California. Schenk 1006 (WS) from Clark County, Nevada, was recovered as sister to M. tiehmii from Nye County, Nevada. Mentzelia tiehmii is a state-listed gypsum endemic that closely resembles small forms of M. oreophila and occurs in the Great Basin near the northernmost edge of the M. oreophila distribution (Schenk and Hufford 2020). The specimen Cohen 120 (RSA) from Inyo County, California, on the other hand, was recovered as sister to the federally endangered M. leucophylla from Nye County, Nevada, near the California border. Mentzelia oreophila and M. leucophylla are morphologically similar but distinguished by geographical distribution and leaf trichome density, with M. leucophylla being confined to Ash Meadows National Wildlife Preserve and having dense hairs that form a white cast (Schenk and Hufford 2020). Previous hypotheses have suggested that M. oreophila and M. leucophylla may be the same species (Prigge 1993;Holmgren and Holmgren 2002;Holmgren et al. 2005), however, Schenk and Hufford (2020) argued that phylogenetic, morphological, distribution, and ecological differences are substantial enough to continue recognizing M. leucophylla at the specific level. Our analyses corroborate that the specimen Schenk 1006 represents a new species, which we are currently describing.
Cohen and Schenk (2022) also investigated the two growth forms of M. polita, a state-listed narrow endemic with an affinity for limestone soils. Populations near the type locality of Clark County, Nevada, have erect stems while populations away from the type locality have prostrate stems. We sampled SYSTEMATIC BOTANY 74
[Volume 50 both growth forms of M. polita and found the species to be monophyletic and closely related to the California M. oreophila (Figs. 2, 3), which is consistent with Cohen and Schenk (2022). Our Angiosperms353 phylogeny is largely consistent with the relationships found in Cohen and Schenk (2022) and solidifies the monophyly of a core "Mojave clade," M. polita, and the polyphyly of M. oreophila. The results of both studies support that Angiosperms353 is comparable with RADseq for recently and rapidly radiating clades. PINNATISECT CLADE-Moore et al. ( 2023) used genome skimming to investigate relationships among 20 species in M. subsections Multiflora and Decapetala and tested the monophyly of a "pinnatisect clade", a hypothesized group of five species with pinnatisect leaves, including: M. conspicua Todsen, M. filifolia J.J.Schenk & L.Hufford, M. holmgreniorum J.J.Schenk & L.Hufford, M. laciniata (Rydb.) J.Darl., and M. sivinskii J.J.Schenk & L.Hufford. They recovered three data sets (nrDNA, an anonymous nuclear locus, and chloroplast genomes) all with highly discordant topologies, and recovered them as a clade or failed to reject them as a monophyletic group. Although we did not recover a strict pinnatisect clade, we did recover a clade with all pinnatisect members in it, with the addition of M. collomiae Christy from Coconino County, Arizona, and M. lagarosa from White Pine County, Nevada (Schenk 1158 [WS]; Figs. 2, 3). Like the nrDNA data set from Moore et al. (2023), we recovered M. collomiae and M. lagarosa as closely related to M. holmgreniorum. The inclusion of M. lagarosa in a broadened "pinnatisect clade" makes sense based on the species having nearly pinnatifid leaves. Mentzelia collomiae, on the other hand, has toothed leaves, but this may represent a character reversal in response to the volcanic cinder soil it inhabits, although this hypothesis would need formal testing.
While Schenk and Hufford (2020) recovered M. lagarosa as monophyletic in their Bayesian analysis, we recovered no such relationship in any of our analyses (Figs. 2, 3; Figs. S1, S2). Our sampling, however, was more widespread; we sampled Schenk 1158 (WS) from White Pine County, Nevada, and Hufford 4332 from San Miguel County, Colorado (WS; Figs. 2, 3), while Schenk and Hufford sampled from White Pine County, Nevada and Wayne County, Utah. Our results indicate that there may be geographic isolation in the Colorado M. lagarosa, but population sampling across the species range should be performed to elucidate the presence of any hidden species or hybridization. Although we found no evidence of hybridization of M. lagarosa with M. cronquistii (Table S3), hybridization or introgression may also explain why we recovered a polyphyletic M. lagarosa and why the placement of the species was so discordant between data sets in Moore et al. (2023).
Discordance and Hybridization-While HybPhaser analyses flagged M. longiloba var. longiloba (Schenk 1225 [WS]), M. humilis var. humilis (Schenk 1264 [WS]), and M. saxicola (Schenk 1256 [WS]) as putative hybrids, discordance between data sets and the recovery of polyphyletic species may also indicate potential hybridization. We discuss the relationships of these putative hybrids and their close relatives below.
MENTZELIA LONGILOBA-We found evidence that M. longiloba var. longiloba from Imperial County, California (Schenk 1225 [WS]) may represent an F1 hybrid between M. procera from Hudspeth County, Texas (Schenk 895 [WS]) and M. longiloba var. longiloba from San Bernardino County, CA (De Groot 2333 [WS]; Table S3). Introgression from M. procera may explain why Schenk and Hufford (2020) recovered only Texas M. procera, and no other populations, in a polytomy with M. longiloba var. longiloba. Hybridization between M. procera and M. longiloba var. longiloba may be limited to west Texas, however, proper population-level studies will be necessary to confirm this, especially in eastern Arizona where the ranges of the two species overlap (Schenk and Hufford 2020). Additional sampling of M. procera populations outside of Texas will be needed to confirm the relative placement of species in the phylogeny.
Although not flagged by HybPhaser as a putative hybrid, Mentzelia longiloba var. chihuahuaensis had discordant relationships between data sets that may be indicative of hybridization. Mentzelia longiloba var. chihuahuaensis (Schenk 898 [WS]) and M. procera (Schenk 895 [WS]) have overlapping distributions in west Texas where they have the potential for hybridization (Schenk and Hufford 2020); however, we did not have the sampling to formally test this hypothesis with HyDe. We recovered Mentzelia longiloba var. chihuahuaensis as sister to M. longiloba var. longiloba in our strictest data set of 75 loci (Fig. S2), suggesting that it is likely the true sister to M. longiloba var. longiloba, which does not contradict previous studies (Schenk andHufford 2011, 2020); however, additional population sampling of M. longiloba var. chihuahuaensis across its distribution will be necessary to fully understand its relationship to M. longiloba var. longiloba.
MENTZELIA HUMILIS-Populations of Mentzelia humilis var. humilis in northern New Mexico (Schenk 1264 [WS]) are sometimes mistaken for M. perennis and M. todiltoensis N.D.Atwood & S.L.Welsh due to similarities in floral, leaf, and growth forms (Schenk and Hufford 2020). Despite morphological similarities and overlapping distributions, HyDe analyses found no signs of hybridization among these three taxa (Table S3).
We recovered a monophyletic Mentzelia humilis var. humilis in our 108-locus analysis, however, we recovered M. humilis var. guadalupensis as sister to M. humilis var. humilis from Culberson County, Texas (Schenk 1264 [WS]) in our 75-and 238-locus analyses, which may indicate gene flow between the taxa. We investigated M. humilis var. humilis 1264 as a putative hybrid with M. humilis var. guadalupensis but found no significant signs of hybridization (Table S3). Since M. humilis var. humilis 1264 was flagged as a putative hybrid by HybPhaser, it remains possible there is an unknown parent we did not investigate, although we sampled the most likely parents, M. perennis and M. todiltoensis (Schenk and Hufford 2020). More comprehensive sampling of the M. humilis varieties and its close relatives will be necessary to elucidate the presence of any cryptic taxa or hybridization.
MENTZELIA PERENNIS-We sampled two populations of Mentzelia perennis to represent two distinct laminal forms. The plants from the type locality (White Sands, NM) have mostly entire to slightly lobed leaves, while plants outside the type locality have deeply lobed to pinnatisect leaves that are similar to M. humilis (Schenk and Hufford 2020). In this study, Spellenberg 10586 (ID, NMC) from Sierra County, New Mexico represents the type locality and entire leaves, while Schenk 2642 (UNM) from Sierra County, New Mexico represents pinnately lobed leaves. In previous studies, populations representing both laminar types of M. perennis have been recovered as closely related to M. todiltoensis (Schenk andHufford 2011, 2020). Our results, however, suggest that only the pinnately lobed M. perennis (Schenk 2642) is closely related to M. todiltoensis while the entire-margined M. perennis (Spellenberg 10586) is more closely related to M. humilis (Figs. 2, 3). Future studies should focus on deeper sampling of M. perennis, M. humilis, and M. todiltoensis to investigate geographic isolation and hybridization among these species.
MENTZELIA SAXICOLA-The close relationship recovered between Mentzelia saxicola (Fig. 2), M. mexicana, and M. longiloba var. chihuahuaensis was not found in previous studies (Schenk andHufford 2011, 2020); however, the taxa do have overlapping ranges in the Chihuahuan Desert, and M. mexicana and M. saxicola are notably morphologically similar (Thompson and Powell 1981;Schenk and Hufford 2020). We found substantial evidence that M. saxicola from Hudspeth County, Texas (Schenk 1256 [WS]) had large genomic contribution from both M. mexicana from Coahuila, Mexico (Granados & Schenk 808 [MEXU]) and M. longiloba var. chihuahuaensis from Brewster County, Texas (Table S3). A hybrid origin would explain why, in only the 108-locus tree, M. longiloba chihuahuaensis is recovered in a clade with M. saxicola and M. mexicana and not with M. longiloba var. longiloba. Mentzelia saxicola and M. procera are also known to hybridize in Hudspeth County, Texas (Schenk and Hufford 2020), although we sampled both species from this locality and we did not find evidence of hybridization between them (Table S3).
MENTZELIA PUMILA-Mentzelia pumila has a history of confounding taxonomy (Schenk and Hufford 2011) and was recovered as a monotypic lineage with no clearly-defined close relatives in Schenk and Hufford (2020). We recovered M. pumila within (Figs. 2, 3) or sister to (Fig. S2) M. speciosa in our 108-and 75-locus analyses, respectively, but sister to a clade comprised of M. laevicaulis (Douglas) Torr. & A.Gray and M. inyoensis Prigge in our 238-locus analysis (Fig. S1). Recovery of M. pumila within M. speciosa was unexpected, as Schenk and Hufford (2020) recovered M. speciosa as monophyletic; however, this close relationship does makes sense geographically because both M. pumila and M. speciosa are distributed in the middle Rocky Mountains (Schenk and Hufford 2020). On the other hand, M. pumila has notably similar seed morphology and chromosome numbers as M. laevicaulis (Hill 1975). We did not have the population sampling to test for hybridization in M. pumila with HyDe, but future studies should explore potential hybridization with M. speciosa and/or M. laevicaulis.
The goal of our study was to elucidate evolutionary relationships within Mentzelia section Bartonia using Angios-perms353 target-capture. Like Schenk and Hufford (2011), the phylogeny we recovered was well-supported at the base and the tips but lacked resolution or strong support along the backbone. Relationships between tips were mostly consistent across data sets, with most relationships being wellsupported, consistent with previous studies, and resolving previous polytomies, but some topological differences warrant further investigation (Figs. 2, 3; Figs. S1, S2). We also recovered several nonmonophyletic species that illuminated potential cryptic speciation or hybridization that should be investigated in future studies (Figs. 2,3;Figs. S1,S2). Overall, the Angiosperms353 approach, in conjunction with our rigorous bioinformatics pipeline, provided species-level resolution to this recent and rapidly evolving lineage from western North America, significantly progressing our understanding of its evolution. To provide stronger support for relationships across all clades, however, lineage-specific baits may be necessary to resolve relationships in areas of the phylogeny with little to no support. S1. Recovery statistics for the Angiosperms353 data of Mentzelia section Bartonia and outgroup. "Clean Reads" indicates the number of reads after quality control with fastp. "Genes With Sequences" is the number of Angiosperms353 loci recovered by HybPiper, regardless of sequence length. "Genes At 50%" is the number of loci with recovered sequences at least 50% of the targeted locus length. Paralog Warnings (Long) and Paralog Warnings (Depth) are flags by HybPiper. "Paralog Warnings (Long)" are the number of loci in which the specimen contains multiple contigs with sequences $ 75% of the reference. "Paralog Warnings (Depth)" are the number of loci in which the set of sequences for a specimen is extracted from contigs with a depth greater than 1 across $ 75% of the reference. "Paralogs Detected (HybPhaser)" are the number of loci determined to be paralogous for that specimen, as determined by HybPhaser, based on the specimen having a relatively high proportion of SNPs for that locus (more than 1.5 Ã IQR [interquartile range] above the third quartile of mean). "Allele Divergence" (AD) indicates the proportion of SNPs distributed across all loci, and "Locus Heterozygosity" (LH) indicates the proportion of loci containing SNPs. Both AD and LH are calculated by HybPhaser, with an em-dash (-) indicating specimens for which HybPhaser failed to detect SNPs. All specimens were sequenced on Illumina NovaSeq, except those sequenced on MiSeq ( Ã ) and NextSeq (1). Mentzelia collomiae is the only specimen to be sequenced on both NovaSeq and MiSeq platforms, therefore raw reads from both sequencing runs were bioinformatically combined before quality control steps.
Specimen Clean Reads Genes With Sequences Genes At 50% Paralog Warnings (Long) Paralog Warnings (Depth) Paralogs Detected (HybPhaser) Allele Divergence Locus Heterozygosity M. albescens 912 2,227,842 348 309 0 3 10 0.048 20.73 M. argillicola 4156 11,616,024 352 336 5 17 33 0.504 87.35 M. argillosa 4145 5,889,742 351 330 3 34 0 --M. candelariae 1145 9,806,526 351 331 3 9 9 0.055 23.28 M. canyonensis 25716 23,206 37 4 0 0 0 --M. chrysantha 811 110,866 196 59 0 0 22 0.502 84.29 M. collomiae 1662 Ã 5,584,678 349 326 4 14 16 0.228 55.99 M. conspicua 1624 501,622 325 188 0 0 10 0.452 69.11 M. cronquistii 3692 Ã 2,144,052 331 251 1 3 15 0.773 88.25 M. decapetala 3753 352,110 305 164 0 0 17 0.746 84.29 M. densa 1811 11,501,858 350 331 4 24 0 --M. filifolia 2506 6,196,804 348 307 1 23 0 --M. flumensevera 4147 2,341,312 351 332 7 12 0 --M. goodrichii 4144 8,792,682 351 334 7 62 0 --M. holmgreniorum 2509 4,114,840 351 330 3 31 0 --M. hualapaiensis 5995 1,762,998 350 324 6 15 24 0.291 67.48 M. humilis var. guadalupensis 12440 10,656,326 351 329 4 9 15 0.804 95.21 M. humilis var. humilis 892 5,190,376 349 317 4 11 29 0.635 89.85 M. humilis var. humilis 1264 5,462,922 350 328 3 14 8 0.843 92.94 M. integra 1728 10,410,086 351 336 10 31 5 0.136 26.88 M. integra 1737 Ã 415,368 265 123 0 0 0 --M. inyoensis 10764 4,396,978 348 317 2 8 10 0.190 37.54 M. laciniata 1609 7,096,118 351 334 1 8 7 0.042 17.96 M. laciniata 2673 2,301,896 350 331 0 5 0 --M. laevicaulis var. laevicaulis 1857 3,652,830 351 329 5 14 11 0.166 59.7 M. laevicaulis var. parviflora 1070 5,687,850 351 336 3 12 17 0.072 34.34 M. lagarosa 1158 7,686,418 351 332 4 14 24 0.537 82.34 M. lagarosa 4332 9,572,544 351 335 16 43 0 --M. leucophylla 2247 1 19,360,964 351 325 8 17 0 --M. librina 4258 5,208,592 346 306 1 16 3 1.051 90.36 M. longiloba var. chihuahuaensis 898 3,421,310 351 330 12 34 6 1.808 99.7 M. longiloba var. longiloba 1225 4,029,734 351 333 10 39 5 0.989 74.63 M. longiloba var. longiloba 2333 9,671,538 351 334 15 55 16 0.587 83.28 M. longiloba var. longiloba 4813 7,531,206 351 335 7 25 0 --M. longiloba var. pinacatensis s.n. 7,190,992 348 324 5 36 12 0.892 84.73 M. longiloba var. yavapaiensis 1011 9,873,978 350 335 23 52 0 --M. marginata 963 2,559,502 351 328 2 7 18 0.568 90.45 M. memorabilis 4151 8,964,842 352 333 4 20 29 0.754 91.52 M. mexicana 808 1 15,324,096 348 322 8 20 0 --M. multicaulis 14485 16,964,774 344 324 8 147 5 0.931 84.73 M. multiflora 777 1 26,381,858 350 331 20 37 12 0.373 60.00 M. nuda 911 13,129,672 350 333 5 21 18 0.659 68.07 M. nuda 2665 8,563,402 350 333 4 12 14 0.433 64.11 M. oreophila 120 1,635,446 343 307 3 6 14 0.16 39.19 M. oreophila 1006 13,011,326 350 333 14 43 26 0.325 66.97 M. paradoxensis 972 Ã 872,706 312 203 0 0 0 --M. perennis 2642 2,312,220 349 326 9 34 10 0.502 78.38 M. perennis 10586 2,940,612 349 318 3 13 0 --M. polita (prostrate) 7,313,732 351 334 6 13 18 0.339 63.36 M. polita 6453 (type form; erect) 8,770,608 351 330 4 9 9 1.31 80.93 M. procera 895 1 2,633,162 249 105 2 2 0 --M. pterosperma 14259 3,734,858 351 337 3 10 17 0.483 85.37 M. puberula 1224 1,123,438 344 313 2 4 22 0.512 89.82 M. puberula 3632 1,164,588 348 328 3 7 0 0.038 8.17 M. puberula 16686 2,868,834 350 320 2 6 21 0.372 81.63 M. pumila 978 Ã 1,372,364 321 224 0 1 0 --M. reverchonii 4310 5,388,304 351 330 2 16 25 0.692 91.64 M. rhizomata 4135 7,471,098 351 335 10 27 10 0.050 20.25 M. rusbyi 2511 3,148,478 351 331 1 6 11 1.586 94.63 M. saxicola 813 4,131,008 350 333 1 6 0 --(Continued) FABRE ET AL.: PHYLOGENOMICS OF MENTZELIA SECTION BARTONIA 81 2025]
TABLE S1. (CONTINUED) Specimen Clean Reads Genes With Sequences Genes At 50% Paralog Warnings (Long) Paralog Warnings (Depth) Paralogs Detected (HybPhaser) Allele Divergence Locus Heterozygosity M. saxicola 1256 12,074,082 351 336 29 78 22 0.323 55.69 M. shultziorum 4140 81,074 192 53 0 0 14 0.236 38.55 M. sivinskii 2492 2,553,328 351 323 2 10 6 0.529 57.66 M. speciosa (M. sinuata form) 786 1,979,160 350 326 3 8 9 0.813 81.27 M. speciosa 790 12,004,932 341 325 15 39 15 0.798 93.73 M. speciosa 792 8,897,762 351 333 11 27 4 0.325 58.61 M. speciosa 1437 5,416,580 350 329 2 12 6 0.638 58.15 M. springeri 856 Ã 1,036,230 317 217 1 1 10 0.786 89.12 M. tiehmii 4157 13,071,692 349 331 8 35 11 0.751 89.79 M. todiltoensis 854 4,175,890 351 327 3 29 6 0.225 31.4 M. todiltoensis 2652 4,890,634 350 332 9 29 0 --M. uintahensis 961 9,577,866 349 326 7 16 13 0.126 27.46 Outgroup M. involucrata 4622 15,533,990 351 318 15 44 8 0.081 24.18 M. nesiotes 2680 8,838,620 351 328 3 12 27 0.495 71.74 M. reflexa 14325 7,731,924 351 324 1 11 1 2.210 98.81 M. tricuspis 553 15,188,018 351 325 43 113 7 2.358 97.87 M. tricuspis 3626 1,279,880 344 287 9 23 10 0.048 20.73 Average 6,466,205 337 301 6 21 10 0.436 49.78 TABLE S2. Summary statistics for alignments and gene trees from our three data sets: loci with individual paralogous samples removed by HybPiper (238 Loci), loci with higher-than-average bipartition support (108 Loci), and loci with no samples flagged as paralogous by either HybPiper or Hyb-Phaser (75 Loci). Alignment statistics were calculated with AMAS. Gene tree statistics were calculated with SortaDate (root-to-tip variance, tree length, bipartition support) and IQ-TREE (gene concordance factors [gCF] and site concordance factors [sCF]). Gene tree statistics are represented by the average 6 SD. All data sets included 76 total specimens. 238 Loci 108 Loci 75 Loci Alignments Average number of taxa per locus 74 69 68 Average locus alignment length (bp) 7383 6397 6421 Concatenated alignment length (bp) 1,476,961 690,863 481,597 Total matrix cells in concatenated alignment 112,249,036 52,505,588 36,601,372 Number of undetermined characters 67,349,374 31,513,332 21,558,423 Proportion of missing characters 60.00 60.02 58.90 Proportion of variable sites 0.23 0.23 0.22 Parsimony informative sites 117,955 55,998 34,444 Proportion of parsimony variable sites 0.08 0.08 0.07 GC content (%) 0.38 0.37 0.37 Gene Trees root-to-tip variance 0.002 6 0.030 0.0002 6 0.001 0.007 6 0.055 treelength 0.64 6 0.28 0.62 6 0.24 0.58 6 0.32 Bipartition support 0.15 6 0.05 0.19 6 0.03 0.16 6 0.05 gCF 19.00 6 24.10 22.88 6 26.89 20.35 6 25.19 sCF 43.34 6 15.61 40.92 6 15.50 43.11 6 19.15 TABLE S3. Individual hypothesis testing with HyDe based on the 108-loci data set. Mentzelia longiloba is abbreviated as M. lon. Parent 1 Hybrid Parent 2 Z-score p-value g