Abstract
Premise of the study Universal probesets for targeted sequence capture have facilitated phylogenomic research into diverse plant groups with limited genomic resources, including from low-quality DNA typical of herbarium specimens. Here, we leverage the Angiosperms353 loci to infer the first phylogeny of Otoba (Myristicaceae), a Neotropical tree genus that is ecologically dominant in low-to-mid elevation wet forests, exclusively from herbarium specimens.
Methods We use a combination of Angiosperms353 loci, obtained via targeted sequence capture, and plastid sequences to resolve the phylogeny of Otoba using concatenated and species tree methods. We subsequently use this phylogeny to infer biogeography and trait evolution using phylogenetic comparative methods.
Key results Recovery success of loci is correlated with age of herbarium specimens and average annual precipitation. Despite a large amount of missing data, we resolve the phylogeny of Otoba into three major subclades, each structured by geography. We show that Otoba’s crown radiation occurred on the western slopes of the Andes in the late Miocene, and from there, migrated into Central America at least twice; the genus was only able to cross to the eastern slopes of the Andes a single time. Trait evolution has been dynamic across vegetative and reproductive traits, with multiple origins of most discrete traits investigated, including ecologically important aril color.
Conclusions Otoba is recent, rapid radiation whose evolution is tied to landscape change, including Andean uplift, in the northern Neotropics. Its dynamic morphological evolution is consistent with sorting of ancestral traits during recent speciation events. In one of the first herbariomic studies exclusively using herbarium tissue from specimens collected in the wet tropics, this study demonstrates the promise of Angiosperms353 loci in resolving shallow species-level relationships, even from low-quality DNA.
The Northern Neotropics experience a dramatic recent geological history, punctuated by periods of rapid mountain uplift in the Andes (Hoorn et al., 2010) and Central America (De Boer et al., 1995) and a potentially gradual closing of the Isthmus of Panama (Bacon et al., 2015; O’Dea et al., 2016). Among the more notable geographic features of the Neotropics are the Andean mountains of western South America, the longest north-to-south oriented mountain chain in the world and the second tallest of any globally. These mountains began their rise in the Paleocene, but major bursts of mountain building occurred more recently, 4 to 12 My ago. This uplift not only changed local topography, but also impacted continental-scale climate and the entire landscape of the Neotropics (Hoorn et al., 2010). It is thus not surprising that the Andes are known to be important in structuring biogeography and species relationships in plants that occur in montane Andean habitats (Pennington et al., 2010; Särkinen, Pennington, et al., 2012; Lagomarsino et al., 2016; Hoorn et al., 2019), as well as in extra-Andean plants, including lowland tropical rainforest plants of the Amazon basin (Antonelli et al., 2009; Dick et al., 2012).
Both the Isthmus of Panama and the Andean mountains are important geological features of the northern Neotropics that promote speciation in allopatry following long distance dispersal (Gentry, 1982; Antonelli et al., 2009). Given their height and extreme habitat heterogeneity, the Andean mountains are a particularly important barrier to species movement, especially to tropical species that typically have relatively limited environmental preferences (Janzen, 1967). While there are widespread species that occur on either side of the Andes, including Cordia alliodora (Boraginaceae) (Rymer et al., 2013), Symphonia globulifera (Clusiaceae) (Dick and Heuertz, 2008), and Schizolobium parahyba (Fabaceae) (Turchetto-Zolet et al., 2012), it is more common that genera, not species, have trans-Andean distributions. Many of these groups are also found in Central America, which may have been facilitated by the closing of the Isthmus of Panama between 3 and 15 Mya (Coates and Stallard, 2013; Bacon et al., 2015). Myristicaceae, a pantropical family of mid-canopy and canopy trees with high species richness in the Neotropics (ter Steege et al., 2006), includes multiple genera with trans-Andean distributions.
Across their full diversity (comprising 21 genera and ca. 500 species), Myristicaceae are notable for their importance in ethnobotany, including as food plants (e.g., nutmeg and mace, Myristica fragrans), timber species (e.g., Virola surinamensis), and hallucinogenic drugs (e.g., epená, Virola sp. (Alrashedy and Molina, 2016)). Myristicaceae also have multidimensional ecological importance. For example, due to their large, arillate seeds, they are some of the most important food sources in the lowland tropics for large-bodied birds, such as toucans and mot-mots, primates, and bats, which, in turn, act as important seed dispersers (Forget et al., 2000; Russo, 2003; Giraldo et al., 2007; Melo et al., 2009; Moreira et al., 2017). Though pollination is less well studied, the small, usually imperfect flowers, of which thousands can be in bloom at a single time on an individual plant (Kanstrup and Olesen, 2000), are known to be generalist pollinated by various small insects including beetles, flies, and thrips in Myristica (Armstrong and Irvine, 1989; Sharma and Armstrong, 2013), Virola (Jardim and Mota, 2007). A similar set of small, generalist pollinators is likely common throughout the distribution of the family. Myristicaceae are usually dioecious, with individual plants producing either carpellate or staminate flowers, which results in differential resource allocation, with pistillate plants (which give rise to the fruit) investing more in stem growth than staminate plants (Queenborough et al., 2007a). Further, the six Neotropical genera of Myristicaceae, Bicuiba, Compsoneura, Iryanthera, Osteophloeum, Otoba, and Virola, are important components of wet tropical forests at low to mid-elevations, and individual species can be among the most abundant in lowland tropical rainforests (ter Steege et al., 2006). Neotropical Myristicaceae are an important system for understanding the ecological processes that allow species co-existence in hyperdiverse communities in the western Amazon Basin (Queenborough et al., 2007b; c).
Despite its relatively low species richness, Otoba (Fig. 1) has among the broadest ecological tolerances of all Neotropical Myristicaceae. This genus of ca. 10 species are distributed from Nicaragua to Brazil, with the highest species richness in low Andean montane forests and lowland rainforests, especially of the Chocó region and western Amazon (Santamaría-Aguilar, Jiménez, et al., 2019). They are abundant in low Andean montane forest, and include the highest elevation occurrence of any member of Myristicaceae (Jaramillo-Vivanco and Balslev, 2020). However, species can also be found in lowland rainforests, and Otoba is one of the ten most abundant genera in western Amazonia (ter Steege et al., 2006; Guevara Andino et al., 2017). Individual species can be some of the most common in many forests, including O. parvifolia in Madre de Díos, Peru (Pitman et al., 2017; Swamy, 2017) and Madidí, Bolivia (Macía, 2008), O. glycycarpa in Yasuní, Ecuador (Guevara Andino et al., 2017), and high várzea forest of the Amazonian floodplain in Brazil and Bolivia (Wittmann et al., 2006). Otoba parvifolia, a wide-spread species of the Western Amazon, shows high intraspecific genetic differentiation (Honorio Coronado et al., 2019).
Otoba is distinct among Myristicaceae in many regards. Like other members of the nutmeg family, Otoba is characterized by a strong aromatic scent from essential oils, a pagoda-like growth form (i.e., “Myristicaceous growth”, or Massart’s model (Hallé et al., 1978)), dioeciousness with small, trimerous flowers (Armstrong and Tucker, 1986) (Fig. 1A-C), red, dilute latex (Fig. H), and a characteristic valvate capsule that opens to reveal a large, arillate seed (Fig. 1D-E). However, Otoba can be distinguished from these genera by a variety of traits. Within Neotropical Myristicaceae, Otoba is notable for its low-montane distribution (vs. the more common lowland rainforest) and seeds that most commonly have white arils (vs. typically brightly colored, as in mace). It has bifacial pollen with continuous tectum on the distal pole and reticulate tectum on the proximal pole (Sauquet and Le Thomas, 2003), a unique set of characters within Myristicaceae that is more similar to African members of the family than other Neotropical genera. It is also the only genus of Neotropical Myristicaceae with conduplicate vernation (Fig. 1).
These differences are perhaps not surprising given its relatively distant relationship to other Neotropical Myristicaceae. Myristicaceae is sister to the rest of Magnoliales (Qiu et al., 2006; Soltis et al., 2007; Massoni et al., 2014), a magnoliid clade that represents one of the oldest angiosperm orders (Magallón et al., 2015). Compared to other families within its order, Myristicaceae have low levels of genetic diversity. This may be a product of the relatively recent origin of extant Myristicaceae, compared to other Magnoliales, in the Miocene or late Oligocene (15–18 My: (Doyle et al., 2004; Massoni et al., 2015). This is corroborated by recent molecular dating analyses across angiosperms that suggest that the family originated at least 20 My after the crown-group of Magnoliales at 109 My (Magallón et al., 2015). Despite these young molecular age estimates, fossil evidence suggests that the family has existed since at least the early Eocene, though it is possible that Myristicacarpum chandlerae represents a stem lineage (Doyle et al., 2008). Within the family, Myristicaceae is split into three major clades: the Malouchoids, Pycnathoids, and Myristicoids (Doyle et al., 2004). While most Neotropical members of this family are included in the Myristicoids, Otoba is nested within the otherwise African Pycnathoids. This, with the relatively young crown age of Myristicaceae, suggests that it represents the product of long-distance dispersal to the Neotropics from Africa, followed by in-situ diversification (Sauquet et al., 2003; Doyle et al., 2004).
Despite their ecological importance and past work into the phylogeny of Magnoliales and Myristicaceae, there is no species-level phylogeny of Otoba to date. Further complicating this endeavor, species of Otoba are morphologically similar and, as is common for tropical trees, poorly represented in herbaria, making systematic treatment challenging (Gentry, 1979; Bebber et al., 2010).Like many other primarily tropical groups, Myristicaceae has few genomic resources, with only a single publicly available transcriptome (Myristica fragrans; (Carpenter et al., 2019). Luckily, the recent development of universal probe sets that target loci of phylogenetic utility across angiosperms, including Angiosperms353, has facilitated the phylogenomic analysis of such understudied lineages. Further, the molecular technique they rely upon, targeted sequence capture, is robust to the low-quantity, low-quality DNA from herbarium specimens. This is important, as inclusion of herbarium specimens is often necessary to achieve robust taxon sampling for clades that grow in regions that are difficult to reach or to collect in for a variety of reasons, including international collaborations, government permitting, or regional unrest (Rabeler et al., 2019). While herbarium specimens serve as a valuable resource to improve taxonomic sampling, DNA from herbarium specimens collected in the wet tropics have been shown to perform relatively poorly due to preservation issues (Brewer et al., 2019) posing a challenge to robust genetic sampling. We take advantage of the universal nature of the Angiosperm353 probeset to infer the first phylogenetic hypotheses for Otoba. This is also the first phylogenomic study that relies exclusive on DNA extracted from herbarium specimens collected in the wet tropics. Using this phylogenetic framework, we discuss implications for biogeography and trait evolution of Otoba.
MATERIALS AND METHODS
Taxon Sampling
Twenty accessions of Otoba representing nine (O. acuminata, O. cyclobasis, O. glycycarpa, O. gordoniifolia, O. gracilipes, O. latialata, O. novogranatensis, O. parvifolia, and O. vespertilio) of the ten accepted species and two undescribed species (Otoba sp. nov.) were sampled. All accessions came from herbarium specimens; voucher information may be found in Appendix S1 (see the Supplemental Data with this article). Herbarium acronyms follow Index Herbariorum (Thiers, constantly updated: http://sweetgum.nybg.org/science/ih/). To serve as outgroups, data from the following transcriptomes available on 1KP project (Carpenter et al., 2019; One Thousand Plant Transcriptomes Initiative, 2019); <https://db.cngb.org/onekp/>) were gathered for Myristicaceae (Myristica fragrans), the broader Magnoniales (Magnolia maudiae, Annona muricata), and Laurales (Cassytha filiformis, Sassafras albidum, and Persea borbonia).
DNA extraction, library prep, target enrichment, and sequencing
Dried leaf tissue was weighed to obtain 500 mg, and tissue was homogenized using an MP Biomedicals FastPrep-24TM 5G Instrument. DNA extraction followed a modified sorbitol extraction protocol (Štorchová et al., 2000). Double-stranded DNA concentration was quantified using an Invitrogen Qubit 4 Fluorometer, and fragment size was assessed on a 1% agarose gel. For samples with a high concentration of large fragments (>800 bp), the DNA was sheared using a Bioruptor Pico (Diagenode Inc., Denville, New Jersey, United States) to obtain an average fragment size of ~ 500 bp. Library preparation was carried out using KAPA Hyper Prep and KAPA HiFi HS Library Amplification kits (F. Hoffmann-La Roche AG, Basel, Switzerland) and with iTru i5 and i7 dual-indexing primers (BadDNA, University of Georgia, Athens, Georgia, United States). Library preparation with KAPA Hyper Prep followed the manufacturer’s protocol (KR0961 – v8.20), except for the following modifications: reaction volumes were halved (25 μL starting reaction, instead of 50 μL), and bead-based clean-ups were performed at 3X volume rather than 1X volume to preserve more small fragments from degraded samples. Library amplification reactions were performed at 50 μL. Target enrichment was carried out using the MyBaits Angiosperms353 universal probe set (Däicel Arbor Biosciences, Ann Arbor, MI; (Johnson et al., 2019)). Target enrichment followed the modifications to the manufacturer’s protocol outlined in (Hale et al., 2020); i.e., pool of 20-24 samples and RNA baits diluted to ¼ concentration). Unenriched DNA library was added to the cleaned, target enriched pool to increase the amount of off-target, chloroplast fragments in the sequencing library. DNA libraries were sent to Novogene Corportation Inc., (Sacramento, California, United States) for sequencing on an Illumina Hiseq 3000 platform with 150 bp paired-end reads.
Sequence processing, assembly, and alignment
Raw sequence reads were demultiplexed by Novogene Corportation Inc., (Sacramento, California, United States). Adapter sequence removal and read trimming were performed using illumiprocessor v2.0.9 (Faircloth et al., 2012; Faircloth, 2016), a wrapper for trimmomatic v0.39 (Bolger et al., 2014). The default settings were used and reads with a minimum length of 40 bp kept.
HybPiper v. 1.3.1 (Johnson et al., 2016) was used to assemble and extract target regions. Read mapping, contig assembly and coding sequence extraction were performed running the reads_first.py script. The intronerate.py script was run to extract introns and intergenic sequences flanking targeted exons. The retrieve_sequences.py script was run first with the “dna” argument to extract coding regions and subsequently with the “supercontig” argument to extract both coding and non-coding regions as a single concatenated sequence for each target gene. Individual genes were aligned using MAFFT v. 7.310 (Katoh and Standley, 2013). Alignments were visually inspected in AliView v. 1.18.1 (Larsson, 2014) to identify alignment errors, assembly errors, and areas that were difficult to align. Alignment errors were manually corrected and assembly errors, as well as areas that were difficult to align, were removed from individual alignments. Outgroup sequences were added to cleaned alignments and aligned using MUSCLE v.3.8.31 (Edgar, 2004) as the default aligner program in AliView (Larsson, 2014). Summary statistics on gene alignments were obtained using AMAS (Borowiec, 2016), including length, missing data, and number of parsimony informative sites.
Off-target chloroplast reads were extracted using FastPlast v1.2.6 (https://zenodo.org/record/973887). For all samples there was insufficient data to produce a fully-assembled chloroplast genome. The SPAdes-assembler built into FastPlast iteratively used k-mer lengths of 55, 87, and 121. Assembled contigs from the iteration using k-mer length 87 were mapped to a reference plastome obtained from GenBank (Clark et al., 2016): Horsfieldia pandurifolia (GenBank accession number NC_042225.1). Once mapped, contigs were cleaned by eye to remove assembly errors before generating a consensus sequence. Consensus sequences for each sample were aligned visually against the Horsfieldia plastome, as alignment algorithms performed poorly with the large amounts of missing data over long sequences. Plastomes for Annona muricata (MT742546.1), Cassytha filiformis (MF592986.1), Magnolia maudiae (MN990580.1), and an unverified plastome for Myristica yunnanensis (MK285565.1) were added as additional outgroups.
Assessing the impact of specimen age and climate on capture success
The collection year of each voucher specimen was recorded and the annual precipitation (mm) at the collection locality extracted from the WorldClim 2.0 30s Bioclimatic variable layer (Fick and Hijmans, 2017) using R package raster (Hijmans et al., 2015). Linear regressions were performed for collection year and annual precipitation number versus number of target loci recovered and average sequence length recovered for each sample to determine if the age of specimen and/or the amount rainfall at the collection locality affected the success of target sequence capture. The relationship between age, precipitation, and the number of ungapped basepairs in cleaned chloroplast sequences was also examined to assess the effect of these factors on off-target sequence capture.
Phylogenetic analyses
Gene tree reconstruction
Maximum likelihood (ML) estimation of gene trees was performed for each nuclear locus, a dataset with all nuclear loci combined, the chloroplast genome, and a dataset with chloroplast and nuclear data combined. Alignments were processed with trimAl (Capella-Gutiérrez and Silla-Martínez, 2009) assigning a gap threshold of 15% or 20% to each column, depending on the number of taxa in the alignment. Thresholds were chosen to maintain columns with data for four or more individuals. Alignments were analyzed using RAxML v8.2.12 (Stamatakis, 2014) under the GTR model with optimization of substitution rates and site-specific evolutionary rates. For combined datasets and the chloroplast, analyses were first run with all individuals and the program RogueNaRok v.1.0 (Aberer et al., 2011) was used to identify individuals that negatively impacted phylogenetic inference. Individuals identified by RogueNaRok and or those with little data (total bp <1% of aligned length) were excluded from further analyses.
Multispecies Coalescent
Trees were generated under the multispecies coalescent model in ASTRAL-III (Zhang et al., 2018). Twenty random bootstrap trees were selected from each gene tree analysis of the Angiosperm353 loci, and used as the input for ASTRAL.
Divergence time estimation
Divergence times were estimated on the ML chloroplast-nuclear combined tree using penalized likelihood via the chronos() function in the R package ape (Paradis and Schliep, 2019). Crown ages from the literature (Magallón et al., 2015; Massoni et al., 2015) for Laurales + Magnoliales, Laurales, Magnoliales, and Myristicaceae were applied as secondary calibrations (Table 2). Because (Massoni et al., 2015) presented five different calibration schemes, and therefore five sets of dates for each node, we calculated the mean age for our calibrations as the average of their mean ages estimated by BEAST across the different schemes. The minimum and maximum for each node were selected as the youngest and oldest date, respectively, in the 95% confidence interval of any scheme across the different analyses. To estimate the error surrounding dates at uncalibrated nodes, like the crown age of Otoba, the median/mean, minimum and maximum value from each study was applied as an absolute age at the corresponding nodes.
Ancestral State Reconstruction
Ancestral character estimation was performed using the morphological characters scored in a recently published taxonomic revision of Otoba (Jaramillo-Vivanco and Balslev, 2020), including 10 discrete characters and 18 continuous characters. Character states were applied to the ML chloroplast and nuclear combined topology calibrated to maximum ages in (Magallón et al., 2015) and trimmed to include one representative of each species. Otoba vespertilio was not included in ancestral state reconstructions due to the uncertainty surrounding its phylogenetic placement.
Because Otoba is small genus, and our phylogeny included 8 species, the following discrete characters were simplified from (Jaramillo-Vivanco and Balslev, 2020) to reduce the number of possible characters states: petiole wingedness; pubescence on the underside of leaves; anther shape and attachment; aril color. The degree to which the petiole is winged was simplified from four categories (“obscurely”, “somewhat”, “winged”, and “not winged”; (Jaramillo-Vivanco and Balslev, 2020) to three (obscurely to somewhat winged, winged, and not winged). Pubescence on the underside of leaves was similarly simplified from “pubescent”, “glabrescent”, “somewhat pubescent”, and “densely pubescent” (Jaramillo-Vivanco and Balslev, 2020) to just two character states: glabrescent to somewhat pubescent and pubescent to densely pubescent; O. novogranatensis ranges from glabrescent to densely pubescent and was thus coded as occupying both states. Anther shape and attachment did not vary (i.e, anthers were either globose and dorsally attached or reniform and basally attached, but never reniform and dorsally attached or globose and basally attached), so they were combined into a single character. Aril color was altered to have one state representing pale arils (“white”, “white-yellow”, and “yellow”) and one representing darker arils (“orange-reddish” and “red”). Aril lacination was excluded from this study as information is only available for O. acuminata, O. lehmannii, and O. vespertilio. For continuous characters, the midpoint was taken for measurements given as a range in (Jaramillo-Vivanco and Balslev, 2020) and two-dimensional traits (e.g., ovary size [length × width (mm)] and seed size [length × width (mm)]) were separated into two traits (e.g., ovary length and ovary width).
Since O. novogranatensis was recovered as polyphyletic, the state for the broadly described O. novograntensis was applied to both O. novogranatensis populations in our tree, with the exception of aril color, which we know differs across inferred lineages (see (Santamaría-Aguilar, Jiménez, et al., 2019). Ancestral characters were estimated using the R package phytools (Revell, 2012); the ace() and the fastAnc() functions were used for discrete and continuous traits, respectively.
Biogeographic inference
We modeled biogeographic movements using BioGeoBEARS (Matzke, 2013, 2014; Massana et al., 2015) implemented in RASP v.4.0 (Yu et al., 2015). The same time-calibrated, trimmed topology used in ancestral state reconstructions was used for biogeographic reconstructions. To better understand how major geologic events, like the closure of the Isthmus of Panama and Andean orogeny, correlate with biogeographic events in the evolutionary history of Otoba, movement both between continents, as well as distribution on either side of the Andes were modeled. Each species was coded for occurrence in (A) Central America, (B) South America, or (AB) both. Species were also coded for their distribution on (A) the western side of Andes, including the Darién gap and Central America or (B) the eastern side of the Andes and western Amazonia. Six biogeographic models were tested with BioGeoBEARS (Matzke, 2013, 2014; Massana et al., 2015); the DIVA-like model was selected for reconstruction of continental movements and the BAYArea-like model with jump dispersal was selected for reconstruction of distribution around the Andes. A maximum of two ancestral areas was allowed for both analyses.
RESULTS
Summary statistics of data assembly
The number of Angiosperm353 loci captured, average sequence length, number of ungapped basepairs of chloroplast DNA (cpDNA) for each sample, collection year, and annual precipitation at the collection locality are listed in Table 1. A heatmap of the percent of the reference protein length recovered for each sample at each locus can be found in Appendix S2 and summary statistics for each locus in Appendix S3.
Capture success from herbarium specimens
We found variable success from hybrid enriched target sequence capture across samples. Half of the 20 samples submitted for sequencing recovered fewer than 10 Angiosperm353 loci; only 3 samples recovered more than 100 loci (Table 1). Success in gathering off-target chloroplast data did not necessarily correspond to success in capturing nuclear loci. For example, the sample with the most nuclear data– O. novogranatensis_WS36336 with 217 of the 353 targeted loci– did not recover useful chloroplast data. On the other hand, nearly half of the chloroplast genome was obtained for O. parvifolia_MS1182, despite recovering only 6 nuclear loci.
Specimens used for DNA extraction were collected between 1983 and 2015, and annual precipitation ranged from 1449 mm/year to 3870 mm/year (Table 1). There was a positive correlation with collection year and both the number of loci recovered and the average sequence length recovered--more recently collected specimens tended to recover longer sequences for more loci--though only the correlation between age and average sequence length was significant (Fig. 2). Otoba novogranatensis_LG20482 was an outlier; despite being the oldest specimen, this sample performed well in both target sequence capture and off-target capture of the chloroplast (Table 1; Figure 2). Annual precipitation was negatively, but not significantly, correlated with the number of loci, average sequence length. Higher rainfall in the collection locality reduced the performance of extracted DNA with target sequence capture. There was very little correlation between age or rainfall and off-target capture success (Fig. 2); the outlying sample Otoba novogranatensis_LG20482 did influence results. Without this sample, the weak negative correlation between collection year and off-target capture success became weakly positive and remained insignificant (cor=0.136, p-0.578). Meanwhile, the weak negative correlation between annual precipitation and off-target capture success became more negative, but remained insignificant (cor=−0.431, p=0.065).
Due to large amounts of missing data in both nuclear and chloroplast regions, the following samples were excluded from all analyses: O. gracilipes_DC884, O. novogranatensis_EB500, O. parvifolia_DN9151, O. sp. nov._RC5752, O. sp. nov._JP16902.
Otoba vespertilio, a recently described species from the Caribbean coast of Costa Rica and Panama (Santamaría-Aguilar, Jiménez, et al., 2019), was included in some analyses, despite large amounts of missing data, in an effort to place the species in the phylogeny (Figs. 3 and 4; Appenidces S4-S6). Maximum likelihood analyses were run with and without this species (Figs. 3 and 4; Appendices S4-S6). Overall, we were able to include 8 of the 10 described species of Otoba in phylogenetic analyses and 7 in comparative analyses.
Phylogenetic analyses
We found strong support for the monophyly of Otoba within Myristicaceae, albeit with limited outgroup sampling within the family (nuclear ML: 100, nuclear MSC: 1, chloroplast: 78, cp and nuclear combined ML: 100). Though the exact relationship between species varies across our individual analyses, we have strong support for three major clades of Otoba (Figs. 3 and 4). The first includes O. acuminata, O. cylclobasis, O. gordoniifolia, O. latialata, and O. novogranatensis specimens collected in Central America (nuclear ML: 94, nuclear MSC: 0.9, chloroplast: n/a, cp and nuclear combined ML: 100). The second includes O. parvifolia and O. glycycarpa (nuclear ML: n/a, nuclear MSC: 1, chloroplast: n/a, cp and nuclear combined ML: 75). The third includes individuals of O. novogranatensis collected in South America (nuclear ML: n/a, nuclear MSC: n/a, chloroplast: 63, cp and nuclear combined ML: 88).
Maximum likelihood analyses of concatenated nuclear loci recovered a poorly-supported grade of O. glycycarpa and O. parvifolia successively sister to a well-supported clade including O. cyclobasis, O. latialata, O. gordoniifolia, and O. novogranatensis (Fig. 3A). With O. vespertilio included, the widespread species O. parvifolia is paraphyletic with respect to O. vespertilio. Samples for another widespread species, O. novogranatensis, are monophyletic, but the nuclear dataset only included individuals from Central America. The ASTRAL-III tree also found a clade with O. cyclobasis, O. latialata, O. gordoniifolia, and O. novogranatensis with high support as well as a strongly-supported clade of O. glycycarpa, O. parvifolia, and O. vespertilio (Fig. 1B). Otoba parvifolia remains non-monophyletic in the ASTRAL-III results; however, relationships within the clade are poorly supported (Fig. 3B).
The topology of the ML chloroplast tree did not agree with either of the nuclear topologies; however, relationships in the chloroplast tree are, overall, poorly supported (Appendix S5). The strongly-supported nodes in the ingroup included Otoba as a clade (bootstrap support=78) and the sister pair O. acuminata and O. novogranatensis (CA, sample LG20482; bootstrap support=85). Albeit with weak support along the backbone, the South American samples of O. novogranatensis are inferred to be sister to the rest of the genus. The clade of South American O. novogranatensis samples is resolved as sister to the rest of the genus in the combined chloroplast and nuclear analyses (bootstrap support=88; Fig. 4). The topology for the remainder of Otoba from the chloroplast and nuclear combined dataset is congruent with analyses of nuclear data (Figs. 3 and 4). The clade including O. cyclobasis, O. latialata, O. gordoniifolia, and the Central American O. novogranatensis is again found with high support; a well-supported clade including O. glycycarpa, O. parvifolia is also recovered (Fig. 4). However, when O. vespertilio was included in the combined dataset, O. vespertilio nests within the South American O. novogranatensis rather than O. parvifolia (Appendix S5).
Otoba lehmannii was not sampled in this study and we did not recover sufficient data to include O. gracilipes in our phylogenetic analyses. Based on the geographical structure of clades within Otoba, these species likely either belong to the larger western Andean/Central American clade or in the South American O. novogranatensis clade.
Divergence Time Estimation
Divergence times estimated from different calibrations across studies (Magallón et al., 2015; Massoni et al., 2015) were largely congruent (Table 2; Appendices S7-S12). The mean estimated ages based on (Massoni et al., 2015) were older than those based on (Magallón et al., 2015) and had a broader estimated minimum and maximum range. This is expected as (Massoni et al., 2015) estimated ages under five different timelines for the crown radiation of angiosperms: 130 Ma, 140 Ma, 150 Ma, 170 Ma, and 200 Ma. The estimated age of angiosperms from (Magallón et al., 2015) was 139.4 Ma. All estimates support the radiation of Otoba in the late Miocene (Fig. 6). The crown age for Otoba based on calibrations from Magallón et al. (2015) is estimated to be 7.28 Ma (7.04-7.51 Ma); whereas dates based on (Massoni et al., 2015) are 8.69 Ma (6.5 −11.36 Ma). The divergence between the western Andean/Central American clade and the eastern Andean/Amazonian clade is inferred to have occurred around 6.54 Ma (6.32-6.75 Ma) and 7.67 Ma (5.98-9.83 Ma) based on (Magallón et al., 2015) and (Massoni et al., 2015), respectively.
Ancestral State Reconstruction
Many of the discrete characters exhibited frequent transitions between states along the tree (Figure 5; Appendices S13-S22). These frequent shifts in character state among closely-related individuals, combined with short divergence times along the backbone of the tree, produced near-equivocal reconstructions of character states at ancestral nodes (Appendices S9-S18).
Similarly, closely-related species exhibited marked differences in continuous traits (Figure 5; Appendices S23-S40). All estimations of continuous characters resulted in a value at the root close to the average of the observed states in the genus (Appendices S19 - S36).
Traits with a resolved evolutionary history include: the presence of secondary to tertiary intramarginal veins, the shape and fusion of filaments, the presence of extrastaminal discs, the presence of prickles at the apex of the seed, and aril color (Appendices S10, S11, S12, S17, and S18). Within the more-resolved character histories, multiple derivations of traits are inferred. The presence of secondary to tertiary intramarginal veins emerged at least three times; they are present in O. gordoniifolia, populations of O. novogranatenis, and O. lehmannii (not sampled). Fused, bottle-shaped filament columns are the most common state in Otoba. Filament shape transitioned from the tapered bottle shape to cylindrical in O. cyclobasis. Fusion of filaments was lost multiple times in populations of O. novogranatensis. Extrastaminal discs evolved twice: once in O. acuminata and once in O. cyclobasis. Prickles at the apex of the seed were partially lost in O. cyclobasis and O. parvifolia and lost completely in O. glycycarpa; Otoba vespertilio (not included in ancestral reconstructions) also displays a reduction in seed prickles. Finally, arils are ancestrally pale (white to yellow) in Otoba; red arils have evolved at least once in Central American O. novogranatensis. One additional species not included in ancestral reconstructions, O. gracilipes, also has red arils.
Biogeographic Reconstruction
South America and, more specifically, the western side of the Andes is inferred to be the ancestral area of Otoba (Fig. 6). Jump dispersal to the eastern side and the Amazonian basin best explains the divergence of the O. glycycarpa-O. parvifolia clade (Fig. 6B). In the western-Andean clade including O. acuminata and O. cyclobasis, a South American ancestor is inferred with expansion into Central America via widespread ancestors (Fig. 6A).
DISCUSSION
There were many challenges, both methodological and biological, inherent to our goal of inferring the first phylogeny of the ecologically dominant Neotropical tree genus Otoba. Our use of a universal probe set meant lower specificity in target capture, and that variation within loci would potentially be low. Methodological challenges extended to our exclusive use of degraded, low quantity DNA extracted from herbarium specimens— a challenge exacerbated by Otoba’s distribution in low- to-mid-elevation moist tropical forests, where specimens are particularly difficult to preserve and dry. Biologically, Otoba’s Andean-centered distribution suggests that it is likely a young clade with a phylogeny characterized by short branch lengths and young speciation events, a result borne out in our analyses. Despite these significant challenges, we were able to produce a framework phylogeny for Otoba— the first phylogeny for any Neotropical group of Myristicaceae. Our phylogenetic results consistently included two subclades: 1) an eastern Andean/Amazonian clade comprising O. parvifolia and O. glycycarpa and 2) a northern Andean/Central American clade comprising O. acuminata, O. cyclobasis, O. gordoniifolia, O. latialata, and some individuals of O. novogranatensis. The 6-9 Ma divergence of the eastern Andean/Amazionian clade from the western Andean/Central American clade suggests that long-distance dispersal across the Andes, rather than vicariance via mountain uplift, produced the present distribution.
Landscape evolution and dispersal limitation drive biogeography in Otoba
The Neotropics have a complex geological history (Hoorn et al., 2010), including the rapid uplift of the Andes and the closure of the Isthmus of Panama. These events have had dramatic impacts on the evolution of the taxa that inhabit this region (Hughes et al., 2006; Antonelli et al., 2009; Bacon et al., 2015). We find compelling evidence that both have played important roles in the shaping biogeographic history of Otoba.
Otoba is a product of long-distance dispersal from Africa to the Neotropics. It is likely that the dispersal event that resulted in stem lineage Otoba was from western Africa, where the closest relatives of Otoba, Pycnanthus and Ceolocaryon, occur today, to eastern South America (Doyle et al., 2004) in the last 19-7.2MA (Table 1). During this time period, range expansion across the northern Neotropics would have been facilitated by a more-or-less contiguous swathe of lowland rainforest and a relatively low height of the Andes, including a low elevation gap in the region that comprises the current border area between Ecuador and Peru (Hoorn et al., 2010). By the time that crown Otoba originated ca. 7.2Ma, we infer that this lineage was restricted to the western side of the Andes; the long, empty stem lineage of Otoba is consistent with high rates of past extinction explaining this pattern. Our ancestral state reconstructions and divergence time estimates point to a western Andean origin of Otoba (Fig. 6) in the late Miocene to early Pliocene. This is a region of origin shared with relatively few other Neotropical tree groups, including a clade of Annonaceae (Pirie et al., 2018); more tree clades that have been investigated have Amazonian origins, including the Brownea clade (Fabaceae; (Schley et al., 2018) and Neotropical Chrysobalanaceae (Chave et al., 2020). Following establishment, Otoba seem to be one of few lineages that experience in situ diversification within the Chocó biogeographic region (Pérez-Escobar et al., 2019).
The Andes structure the ranges of species and clades in crown Otoba. Species either occur on the eastern slopes of Andes, extending into the western Amazon basin, or on the western slopes of the Andes and/or in Central America. The same pattern is observed within subclades, with closest relatives sharing distribution on one side of the Andes. During the time that elapsed between the dispersal from Africa and the origin of crown Otoba, the northern Andes gained approximately half their elevation (Garzione et al. 2017) and analogs to modern montane cloud forests formed (Hughes, 2016; Martínez et al., 2020). Thus, even though they had not yet reached their full height (Hoorn et al., 2010), the Andes would have represented a significant barrier to dispersal for low-to-mid-elevation groups like Otoba. Consistent with this scenario, we infer only a single dispersal event from the western side of the Andes to the eastern side; this dispersal occurred in the late Miocene to early Pliocene, resulting in two widespread, lowland species, O. glycycarpa and O. parvifolia (Fig. 6). It is unlikely that Andean uplift served as a vicariant event that split widespread populations; instead, we posit that long-distance dispersal across the mountains, likely mediated by bird or mammal seed dispersers, facilitated this disjunct distribution.
Movement between Central and South America has occurred more frequently than movement across the Andes during the evolutionary history of Otoba. Range expansion and long-distance dispersal have both played a role in movement into Central America. A widespread distribution in South and Central America is inferred for the most recent common ancestor of O. acuminata, O. novogranatensis (CA), O. gordoniifolia, and O. latialata in the late Miocene (Fig. 6). The current, widespread distribution of O. latialata in the Chocó-Darién moist forest from Colombia to Panama suggests the common ancestor of the clade may have occupied a similar distribution. If this is the case, a pattern of subset sympatry—when one daughter lineage inherits the ancestral range and the other daughter(s) inherit a portion of the ancestral range (Ree et al., 2005)—is consistent with our inferred biogeographic history. An additional long-distance dispersal event is assumed for O. vespertilio. While the phylogenetic position of O. vespertilio differs across our analyses (Fig. 3; Appendix S6), all possible placements are consistent with it representing a long distance dispersal event from South America to Central America. These events occurred within the last 10 million years, a time-frame that supports the role of the closure of the Isthmus of Panama in facilitating these movements (Montes et al., 2012; Bacon et al., 2013).
The observed biogeography patterns are likely explained by dispersal limitation. Otoba’s relatively large seeds are dispersed birds, primates, and bats (Giraldo et al., 2007; Nuñez-Iturri and Howe, 2007; Melo et al., 2009; Santamaría-Aguilar, Jiménez, et al., 2019), with a potential role of small mammals as secondary seed dispersers (Forget et al., 2002). These groups can often be dispersal limited, including across riverine barriers and fragmented habitats (Eberhard and Bermingham, 2005; Ripperger et al., 2013; Boubli et al., 2015) (though this is not always the case; see (Holbrook, 2011). The potential for dispersal limitation is observed within communities in which the relatively large seeds of Otoba parvifolia are dispersed at low frequency over typically short distances (Terborgh et al., 2011). On a continental scale and over evolutionary time, this has resulted in relatively few long distance biogeographic movements. Otoba’s two migrations into Central American were potentially facilitated by the continuous land bridge of more-or-less suitable habitat, while cold high-elevation habitats likely prevented more frequent traversing of the Andes. Further, the range of Otoba does not occupy all of the suitable habitat it presumably could based on distributions of extant species. Otoba does not occur in other low to mid-elevation moist forest habitats in the Neotropics, like the Atlantic coast forest in Brazil and the Caribbean islands, a pattern mirroring ecologically similar and closely related Virola (Santamaría-Aguilar, Aguilar, et al., 2019). Again, dispersal limitation may explain Otoba’s absence from these regions: the relatively large seeds of Otoba likely make dispersal events over water barriers or large stretches of unsuitable terrestrial habitat uncommon compared to groups that are dispersed by wind (Pérez-Escobar et al., 2017) or migratory passerine birds (Nathan et al., 2008). This is likely coupled to time limitation, given the young age of the genus and the stochastic nature of long-distance dispersal events (Nathan et al., 2008). However, high levels of seed-set, both in closed canopy forests and in treefall gaps, may make Otoba an effective colonizer of new habitats once they do arrive in a new region (Myster, 2020).
Dynamic evolution of morphological traits is common
We observe a dynamic pattern of character evolution across a broad suite of traits in Otoba. While frequent transitions between morphological characters often results in unresolved ancestral state reconstructions, mapping tip states reveals that no two species share a set of discrete traits (Fig. 5). Further, even though ancestral states were equivocal, the majority of the discrete traits that we investigated underwent convergent evolution within the genus, including: winged petioles, pubescence presence and color, filament fusion, anther shape and attachment, and gynoecium pubescence (Fig. 4). Coupled with short divergence times between speciation events after the crown age of Otoba (Fig. 6; Appendices S7 and S8), the observed variation across all traits is consistent with rapid morphological evolution coinciding with rapid lineage divergence upon establishment in South America. This pattern may also be explained by the sorting of ancestral variation (Pease et al., 2016), and it is likely that hemiplasy or parallel evolution underlie morphological evolution. Ecological opportunities on a new continent may have served to differentially select standing ancestral variation, resulting in the repeated evolution of many traits in Otoba.
An unconfirmed, but likely case of convergent evolution is aril color. Most species in Myristicaceae produce red arils that cover seeds to varying degrees. These bright, red arils that contrast with the green and brown of the capsule and seeds (Fig. 1E) serve as an attractant to frugivores who consume the nutritious aril, either with or without consuming the seed (Howe and Vande Kerckhove, 1981; Gautier-Hion et al., 1985). Most species of Otoba, however, produce white to yellow arils (Fig 1F); these pale colors still contrast the seed and capsule and attract frugivores (Gautier-Hion et al., 1985; Wheelwright and Janson, 1985). Two species have red arils: O. gracilipes and the Central American members of O. novagranatensis (supported as a distinct lineage in our analyses). Our field observations suggest that these differences in aril color coincide with differences in texture: red arils are generally thick and waxy, while whitish arils are thin and gelatinous. On a broad phylogenetic scale, this represents convergent evolution of red arils within Myristicaceae and likely serves as a shared ecological signal with other Neotropical members of the family, including the widespread Virola. Within the genus, bright red arils have presumably evolved twice independently (Fig. 5), though the exact phylogenetic placement of these independent origins remains unknown as O. gracilipes was not included in our analyses. We also see convergence in seed size and shape (Fig. 5), additional traits that are likely related to the mechanics of seed dispersal.
Convergence in fruit traits, including seed size and aril color and texture, may be linked to a specialized dispersal syndromes in Otoba. Differences in overall seed morphology throughout the genus, including size, aril color and texture, potentially represent specialization with different classes of dispersers. For example, species with bright red arils, including Virola surinamensis, are consumed by birds (Howe and Kerckhove, 1980; Howe and Vande Kerckhove, 1981), while species with whitish arils are consumed by bats (Melo et al., 2009; Santamaría-Aguilar, Jiménez, et al., 2019); while more comparative field studies are needed to confirm the most effective dispersers across species, these observations are broadly consistent with global patterns of mammal- and bird-dispersed fruits (Sinnott-Armstrong et al., 2018). While seed dispersal is not directly tied to reproductive isolation, theoretical models support mutualisms between animal dispersers and flowering plants coupled with repeated habitat fragmentation as a mechanism that promotes speciation during range contractions, allowing for coexistence of diverged populations upon habitat reunification (Kakishima et al., 2015). This is a likely potential scenario in Otoba, given that major mountain-building events were occurring the northern Andes and, along with them, reorganization of river systems, were occurring during the late Miocene and early Pliocene, when many of the divergence times between species are inferred (Hoorn et al., 1995, 2010; Struth et al., 2015). Fluctuating landscapes and fluvial barriers at the time may have fragmented populations, promoting divergence between those populations and securing species boundaries upon secondary contact. Supporting this, large rivers are important barriers to dispersal and gene flow in tropical birds (Hayes and Sewlal, 2004; Burney and Brumfield, 2009; Fernandes et al., 2014; Oliveira et al., 2017; Sandoval-H et al., 2017); by extension, they should also be significant barriers to the plants that they disperse as well (Nazareno et al., 2017; Dambros et al., 2020).
The need for future phylogeography of widespread species
Otoba includes a combination of narrowly endemic and widespread species. We find strong evidence for polyphyly of one widespread taxon, O. novogranatensis, and for paraphyly of another, O. parvifolia. This non-monophyly is likely both a product of biological processes (e.g., very large population sizes maintained over very large distances in O. novogranatensis (Pennington and Lavin, 2016) and an artifact of insufficient taxonomic study (Lagomarsino and Frost, 2020).
The polyphyletic Otoba novogranatensis is one of the most collected species of its genus. A preliminary revision of the resulting specimens shows variation across many traits. For example, South American specimens differ from Central American specimens in their thicker pericarp, pubescent ovaries, anthers that can be unfused to the base (de Candolle, 1856; Jaramillo et al., 2004; Jaramillo-Vivanco and Balslev, 2020), and generally white arils. Thus, the two distinct lineages of O. novogranatensis that we resolve (Fig. 4), corresponding to South American and Central American accessions, are supported by morphology. There are additional differences within Central American O. novogranatensis that are likely to be taxonomically informative as well, including the size, shape, and pubescence of leaves, as well as the length of staminate inflorescences, perianth, and anthers varies across the Pacific (e.g., R. Aguilar 638 [INB/CR]) and Caribbean slopes (e.g., I. Chacón 1360 [INB/CR]) of Costa Rica; montane populations within Costa Rica exhibit additional differentiation. Together, these observations suggest that phylogeographic analysis of O. novogranatensis is necessary to adequately determine the number of lineages that are currently described under this umbrella taxon. This would facilitate future taxonomic efforts to recircumscribe this species complex to reflect evolutionary relationships, which will likely entail the description of at least one new Central American species.
Similarly, the paraphyly of O. parvifolia is not surprising, but for different reasons. Our phylogenetic results suggest that O. glycycarpa has recently diverged from O. parvifolia, and has since become widespread as well. Our taxon sampling includes multiple accessions of O. parvifolia and a single accession of O. glycycarpa; these species determinations were based on differences in pericarp thickness, gynoecium pubescence, and color of foliar pubescence (Jaramillo et al., 2000). In all analyses, these accessions are each others’ closest relatives, though O. parvifolia is almost always resolved as a grade (Fig. 3, 4; but see Appendix S4 where it is resolved as a clade). The geographic structure of these species are unresolved across our analyses; while Bolivian and Peruvian accessions of O. parvifolia are often resolved as sister lineages that are more distantly related from Ecuadoran and Brazilian accessions O. parvifolia and O. glycycarpa, there is variation across our analyses and support for these relationships is often low. A previous study based on limited genetic data has suggested that O. parvifolia and O. glycycarpa are genetically indistinguishable, with the implication that they represent a single species (Honorio Coronado et al., 2019). This low genetic variability between species, which is consistent with the short branch lengths in our analyses, may be the product of species misidentification, lack of informative variation in the loci used, introgression, or sorting of ancestral variation in the short time since speciation (ca. 3.7 Ma). The latter two mechanisms would not be surprising: our analyses suggest that these species are recently diverged, overlapping in both geographic occurrence, and have overall similar morphology. While our sampling did not allow us to test the monophyly of O. glycycarpa, morphological evidence supports the distinctness of these species (Jaramillo et al., 2000). Of particular note, O. glycycarpa has one of the thickest pericarps of all Otoba, which has implications for potential efficacy of seed dispersers. We consider O. parvifolia to be a distinct species form O. glycycarpa, and its paraphyly is indicative of its complex evolutionary history (Freudenstein et al., 2016). Future ecological and phylogeographic research could target what allows morphologically similar closest-related species to co-occur throughout large swathes of western Amazonia.
Targeted sequence capture promotes herbariomics—with caveats
Hybrid-enriched target sequence capture with Angiosperms353 has proven useful for generating phylogenetically informative data at multiple taxonomic scales and from different sources of DNA (i.e., silica-dried tissue versus herbarium specimens; (Brewer et al., 2019; Shee et al., 2020; Valderrama et al., 2020). To our knowledge, our study represents a new milestone for Angiosperms353 phylogenomics: the first exclusively herbariomic dataset for a wet tropical genus. DNA from herbarium specimens collected in the wet tropics performs poorly compared extractions from other climates and silica-gel dried tissue (Brewer et al., 2019). This is because high humidity in moist tropical forests and often remote localities extend drying times and/or delay access to drying apparatuses. Resultantly, collections are often treated with ethanol to prevent mold and decay until they can be dried. Both storage of tissues in humid conditions after collection (Adams, 2011) and preservation in ethanol (Doyle and Dickson, 1987; Pyle and Adams, 1989) degrades DNA, resulting in damaged and fragmented genetic material (Särkinen, Staats, et al., 2012). Even though current short-read sequencing techniques perform well with herbarium specimens as compared to conventional Sanger sequencing (Bakker et al., 2015), the level of degradation common among wet tropical specimens led (Brewer et al., 2019) to recommend the use of silica tissue for wet tropical groups. While we agree that this is a best-practice, as in other herbarium-based studies (Brewer et al., 2019; Shee et al., 2020), we were able to extract useful phylogenomic data from herbarium specimens using Angiosperms353.
Herbarium samples are an increasingly useful source of DNA for phylogenomic studies, extending the utility of our historic natural history collections (Lendemer et al., 2020), but they are not a panacea. Even high efficiency target capture will fail when DNA is low quality, as is typical in herbarium specimens collected in the wet tropics (Brewer et al., 2019), especially when they are collected in ethanol (Särkinen, Staats, et al., 2012)— a common scenario for Otoba, and likely universal in the herbarium specimens that we sampled. The specimens we included in our analyses span a breadth of age and environmental conditions at collecting localities. The oldest specimen that we included was 37 years old, collected in 1983; surprisingly, it resulted in the most genetic data (Fig. 2). However, not all of our samples generated useful sequences; we had to remove some of our samples completely due to poor quality reads and apparent contamination following visual inspection of alignments. In other cases, we were able to extract a handful of useful loci, but at a much lower quality than the majority of our included species.
In agreement with past studies, we found that capture success, as measured by the total number of loci obtained as well as their average length, decreased with increasing age of the specimen; we further found a correlation between environmental conditions (i.e., annual precipitation) at the site of collection and the quality of phylogenomic data (Fig. 2). However, we were only able to capture more than 10 loci from half of our initial 20 samples, and only three Otoba samples were successful in recovering over 100 loci. This limited capture success is most likely a product of low input DNA quality and not due to sequence divergence from the probeset, especially considering that we had similarly variable success in capturing the high-copy, off-target chloroplast genome. Plastomes from wet tropical specimens have also been found to have higher fragmentation rates and lower sequencing success as compared to plastomes from specimens collected in other climates (Bakker et al., 2015; Brewer et al., 2019). This compounded with a lower depth of sequencing coverage in the off-target chloroplast regions and resulted in higher sequencing error, which led to conflicting phylogenetic signal and low support. The chloroplast data were able to complement our Angiosperms353 data and allowed the inclusion of additional samples in phylogenetic analyses, but stronger, more consistent signal was recovered in Angiosperm353 target loci.
To extend the potential utility of voucher specimens collected in the future for genomic research, efforts should be made to collect leaf tissue in silica gel or other preservation technique, following best practices, including unique IDs to connect the herbarium voucher to this secondary product (Funk et al., 2017). However, many studies of tropical groups rely on museum specimens, as it is not feasible to collect living material for the taxonomic and/or geographic breadth of many clades. As this is often the case, we urge botanical collectors to include information about how voucher specimens were treated (e.g., specifying that specimens were collected and dried in ethanol). We have demonstrated that even though it may not be ideal, it is still possible to generate phylogenomic datasets that can resolve rapid radiations of wet tropical plant clades using the Angiosperms353 probeset.
Utility of Angiosperms353 as universal loci for plant phylogenomics, even in rapid radiations
Despite the fact that the Angiosperms353 loci were designed to be phylogenetically useful across all angiosperms, we were able to resolve relationships within a clade estimated to be between 6.5–11.3 My old (Table 2). This adds to a growing number of rapid radiations whose phylogenies have been resolved using these loci (Larridon et al., 2019). This performance goes against a preconception that universal loci are not useful for species-level phylogenomic analysis, probably borne of experience from the low-variation plastid loci that had universal utility via PCR-based Sanger sequencing (Shaw et al., 2005), as well as related calls to move away from the lofty goal of a universal set of loci for all flowering plants and instead develop lineage-specific loci with more phylogenetic utility (Hughes et al., 2006). However, targeted sequence capture allows the isolation of loci without finicky PCR probes at a percent-divergence that allows for efficient capture over substantial evolutionary distances. Further, universal probesets, including Angiosperms353, are designed such that multiple probe sequences for each locus are included to account for sequencing variability across taxa. Instead of being limiting, the universal nature of Angiosperms353 loci are proving to be informative for both very deep (e.g, (Dodsworth et al., 2019) and very shallow (e.g., within populations of rice, (Van Andel et al., 2019) phylogenetic splits. A further attractive aspect of Angiosperms353 data is that it is cost-effective to generate, especially if pre-sequencing molecular labwork is completed by the researcher and not outsourced (Hale et al., 2020).
While we were able to infer relationships in Otoba using Angiosperms353 loci, there was very limited variation across our dataset (i.e., the proportion of variable sites in concatenated target loci was 0.285). It has been shown in other taxa that custom bait kits are more informative than Angiosperms353 loci for species-level phylogenetics (Jantzen et al., 2020), though this is not always the case (Larridon et al., 2019). It is thus possible that a custom probe kit designed for Otoba and close relatives would have outperformed the Angiosperms353 loci, either alone or in combination. However, developing such custom loci is predicated on the existence of genomic resources, either pre-existing or newly generated, which would have been difficult for Otoba. While the number of angiosperms genomic resources is constantly growing (One Thousand Plant Transcriptomes Initiative, 2019), there are still no transcriptomes or nuclear genomes available for Otoba and data is limited for Myristicaceae overall: there is a single transcriptome available in the 1KP database (nutmeg, Myristica fragrans) and no other genomic resources. Further, we currently do not have access to fresh or silica-dried tissue from Otoba. Because this scenario is common— especially in tropical plant groups, which tend to be understudied (Goodwin et al., 2015), the universal utility and subsequent promise of assembling a standardized set of loci across analyses is a very desirable property of the Angiosperms353 loci. To facilitate standardized loci that can be combined across studies, the plant systematics community should strive for the development of searchable, long-term repositories for Angiosperms353 datasets. Any such future repository would further benefit from being linked to herbarium vouchers, which would help with reproducibility of research as well as continue to extend the use of herbarium specimens in science (Lendemer et al., 2020).
Conclusions
Otoba is a recent, rapid radiation with a trans-Andean distribution whose biogeographic history suggests that key aspects of the Neotropical landscape— namely the Andean mountains and the Isthmus of Panama— have been important barriers to dispersal. It is likely that traversing these barriers has happened only rarely in the history of the genus due to dispersal limitation, a product of its relatively large seeds that rely on large-bodied vertebrates for dispersal. Very short branch lengths separating species coincide with dynamic trait evolution— a pattern consistent with sorting of ancestral variation during rapid lineage diversification. Future research into Otoba would ideally tackle phylogeography of widespread species, including two species that we identify as non-monophyletic: O. novogranatensis and O. parvifolia, both of which can reach very high local abundance in the communities in which they occur. These insights were gained from a relatively small phylogenomic dataset obtained via targeted sequence capture of Angiosperms353 loci from DNA extracted from herbarium specimens. We observed an impact specimen age and the environmental conditions of their collection locality on data quality. While our study highlights the promise of Angiosperms353 in herbarium-based phylogenomics, it demonstrates the challenges inherent in studying rapid radiations broadly, and, more specifically, when using DNA extracted from herbarium specimens collected in the humid tropics.
Author Contributions
LAF, DASA, and LPL conceived of the research; LAF and DS collected data and performed analyses; LAF and LPL wrote the manuscript with significant feedback from DASA and DS.
Data Availability Statement
Illumina reads will be submitted to the NCBI Sequence Read Archive (SRA) and all other data formats (tree files, alignments, character matrices, etc.) will be uploaded on the Dryad Digital Repository and made available upon publication.
Additional Supporting Information may be found online in the supporting information
Acknowledgements
This research was funded by a Louisiana Board of Regents Research Competitiveness Subprogram grant and by the LSU Department of Biological Sciences. We would like to thank the Missouri Botanical Garden (MO) for their access to their important collections, and Reinaldo Aguilar, Rudy Gelis, Timothy Paine, and John Janovec for permission to use their photographs. We thank Brant Faircloth, Matthew Johnson, Carl Oliveros, and Jessie Salter for their guidance in library preparation and Brant Faircloth for access to laboratory equipment. This manuscript benefited from feedback from Laymon Ball, Janet Mansaray, and Diego Paredes-Burneo.