A DNA barcode reference of Asian ferns with expert-identified voucher specimens and DNA samples

Abstract

Ferns belong to species-rich group of land plants, encompassing more than 11,000 extant species, and are crucial for reflecting terrestrial ecosystem changes. However, our understanding of their biodiversity hotspots, particularly in Southeast Asia, remains limited due to scarce genetic data. Despite harboring around one-third of the world’s fern species, less than 6% of Southeast Asian ferns have been DNA-sequenced. In this study, we addressed this gap by sequencing 1,496 voucher-referenced and expert-identified fern samples from (sub)tropical Asia, spanning Malaysia, the Philippines, Taiwan, and Vietnam, to retrieve their rbcL and trnL-F sequences. This DNA barcode collection of Asian ferns encompasses 956 species across 152 genera and 34 families, filling major gaps in fern biodiversity understanding and advancing research in systematics, phylogenetics, ecology and conservation. This dataset significantly expands the Fern Tree of Life to over 6,000 species, serving as a pivotal and global reference for worldwide barcoding identification of ferns.

Background & Summary

A diverse modern land plant lineages, ferns are estimated to include more than 11,000 species¹. These plants are abundant and most diverse in tropical and insular regions in the world². Our understanding of the hyper-diversity in these areas began during floristic investigations of the 19^th and 20^th centuries, before molecular techniques became available. However, tropical ferns from these diversity hotspots remain scarce in DNA databases. This is especially true for Southeast Asia, where around one-third of world fern species are concentrated³, but less than 6% of species have been sequenced⁴.

DNA barcoding — sequencing DNA regions demonstrated to be of broad taxonomic utility — has proven to be an effective tool to evaluate genetic diversity from taxon-wide collections⁵. However, choosing standard loci for DNA barcoding is critical to compare taxa and samples with diverse origins. Like many other plants, plastid rbcL (ribulose-1,5-bisphosphate carboxylase/oxygenase large subunit) and trnL-F (the intergenic spacer between tRNA-Leu and tRNA-Phe genes; sometimes referred to the region extending to the tRNA-Leu intron) are commonly used DNA barcodes in ferns because they have universal primers^6–8, and, as plastid DNA regions, they are uniparentally inherited (reviewed in Kuo et al.⁹). DNA-referencing of these two barcodes has been widely employed^6,10–14 and highly successful in fern phylogenetic studies¹⁵; to date, more than 5,600 fern species are available with at least one of the two barcode sequences in GenBank¹⁵. trnL-F has been shown to have greater interspecific and intraspecific variation relative to rbcL because the former is include (a) non-coding spacer(s), whereas the latter is a protein-coding gene. trnL-F has been previously shown in studies including smaller sampling of ferns to have higher species discrimination rates than rbcL^8,16. In comparison, rbcL is useful as a phylogenetic marker at deeper divergence levels due to its slower evolution rate, and is the most-frequently sequenced genetic region in ferns¹⁵ as well as a core DNA barcode in land plants¹⁷. Therefore, sequencing both regions is highly recommended for fern DNA barcoding projects, which can serve to both identify species and expand the phylogenetic sampling of the fern tree of life. trnH-psbA, another frequently used non-coding DNA barcode in plants¹⁷, is not a prevalent choice for ferns due to its relatively slow substitution rate in most species^8,18. Other proposed plant DNA barcodes, such as nrITS and matK, lack (one of) the above-mentioned advantages, and are thus not prioritized in fern DNA barcoding studies^8,19.

DNA barcoding has been successfully applied in various ecological and floristic surveys of ferns, and is particularly useful for DNA-identification of their cryptic gametophyte stage^{7,13,16,20,21}. Fern gametophytes are free-living but frequently have too few morphological features to be reliably determined to species¹⁶. With DNA approaches, phenological studies and habitat investigations of fern gametophytes in the field can be accomplished based on reliably identified samples (e.g. Quinlan et al.²² and Wu et al.⁷). Notably, some ferns have populations consisting of long-lived gametophytes but producing no spore-producing individuals, which are referred to as ‘independent gametophytes’^23–25. To confirm their species identities, such a molecular identification tool is indispensable. Additionally, fern DNA barcodes have been used to study novel ecological links between these plants and other organisms, including insect pollinators²⁶ and rhizobium bacteria²⁷. As demonstrated in earlier research, prolific production coupled with high dispersibility means that fern spores provide a key signature reflecting environmental changes^28,29. The emerging field of environmental DNA research relies on further development and publication of DNA barcodes, enabling ecologists to readily monitor environmental dynamics and to expand documentation of biodiversity³⁰. A comprehensive and global database of fern DNA barcodes is therefore essential.

Here, we present a voucher-referenced and expert-identified collection of Asian ferns with DNA barcode regions rbcL and trnL-F, encompassing 1,496 samples from 956 species, including hybrid taxa. Of particular value is the large proportion of samples from fern diversity hotspots in South-eastern Asia, including the Philippines, Vietnam, and Malaysia, which fill major gaps in our understanding of these plants, and will facilitate future research of understanding fern diversity there. Furthermore, this DNA barcode dataset also serves as valuable resources for advancing investigation in systematics, phylogenetics, and conservation genetics of ferns from these biodiversity hotspots. Among these samples, 292 species were sequenced for the first time with these DNA barcodes, and will contribute to a notable expansion (4.6%) to the Fern Tree of Life (FTOL)¹⁵. The incorporation of our new sequence dataset with those already existing in FTOL offers the pivotal global database for fern barcode identification.

Methods

Sampling and specimen identification

We sampled a total of 1,496 fern collections across 390 localities in Malaysia, the Philippines, Taiwan, and Vietnam. They were collected during field expeditions spanning from 2005 to 2022, and were vouchered with specimens in Taiwan Forestry Research Institute Herbarium (TAIF). From each collection, tissue was preserved on silica gel for DNA extractions, which are also publicly available at the TPG website (https://www.twfern.org/DB/DNACollection). Species-level identifications were conducted by experienced fern taxonomists, relying on the morphology and genetic data of voucher specimens. A few collections may represent undescribed species or require further taxonomic investigation, so their identification was thus determined only to the generic level.

DNA extraction, amplicon preparation, and sequencing

The workflow of this study is summarized in Fig. 1. First, fresh or silica-dried leaf tissues were used for DNA extractions of the 1,496 fern samples following the CTAB protocol by Kuo³¹. To amplify these DNA barcodes, various PCRs of rbcL and trnL-F amplicons were carried out according to three different sequencing methods, including (1) the traditional Sanger sequencing, and the multiplexing strategies utilizing high-throughput next-generation sequencing (NGS), (2) PacBio CCS (circular consensus sequencing) and (3) Illumina MiSeq. For Sanger and PacBio CCS, longer amplicons (~1 kbp) were amplified and sequenced, with trnL intron alongside trnL-F. For Illumina MiSeq, shorter amplicons (<600 bp) were amplified and sequenced, and the rbcL region was split into two amplicons, rbcLN and rbcLC (Supplementary Figure 1). Dual 8nt-indexed primer sets were employed for the multiplexed amplicons for PacBio CCS and Illumina MiSeq. For these primers, the conservative priming regions were same as Wu et al.⁷ or identified using their approach, with different 8nt-indexes added to the 5′ ends. In addition, for Illumina MiSeq, we co-amplified trnL-F and rbcLN in the same PCR reaction for each of individual DNA sample. The details of primer sets and thermal conditions of PCR cycles were provided in Supplementary Tables 1 and 2, respectively. Each PCR reaction comprised 7.5 μL 2 × SuperRed PCR Master Mix (BIOTOOLS Co., Ltd., New Taipei, Taiwan), 2 μL DNA template (10 ng/μL), 0.75 μL of each primer (10 nM), and ddH2O added to a total volume of 15 μL.

Fig. 1.

The workflow of collecting fern DNA barcodes. The fern diagrams were downloaded from https://www.phylopic.org/ or modified from Vasco et al.⁴⁷.

For Sanger sequencing, PCR products were first purified using ExoSAP‐IT (Thermo Fisher Scientific, Waltham, Massachusetts, USA), and then sequenced by an ABI 3730XL DNA Analyzer (Thermo Fisher Scientific, Waltham, Massachusetts, USA). For PacBio CCS and Illumina MiSeq, amplicon products were initially assessed through 1 × TAE 1% agarose gels. We then pooled these products to achieve similar molecular concentrations according to their estimated DNA concentrations. The DNA fragments with target sizes in these multiplexed amplicon pools were isolated by electrophoreses with 1 × TAE 0.8% agarose gels and purified using Geneaid Large DNA Fragments Extraction Kit (Geneaid, New Taipei, Taiwan). For the isolated DNA products with low O.D. values (i.e. A260/280 and A260/230 < 1.8), further purification was conducted using 1 × AMPure XP Beads (Beckman Coulter, Brea, California, USA) before NGS library constructions. PacBio CCS library preparation and sequencing were carried out at the Sequencing and Genomic Technologies Core Facility of the Duke University Center for Genomic and Computational Biology, utilizing a single SMRT cell on a PacBio sequel sequencer with 3.0 chemistry (Pacific Biosciences, Menlo Park, California, USA). The fastq reads of PacBio CCS were then employed for the downstream demultiplexing and ASV (amplicon sequence variant) generation (see below). We constructed PCR-free libraries for Illumina MiSeq using KAPA Dual-Index Adapter Kit (Roche, Basel, Switzerland). DNA molecular concentrations of these Illumina libraries were measured using Sequencing Library qPCR Quantification (Illumina, San Diego, California, USA). The libraries were then sequenced on Illumina’s MiSeq PE300 platform using Reagent Kit v3 (600-cycle; Illumina, San Diego, California, USA). To obtain DNA barcode sequences from the NGS fastq reads, adapter sequences were first trimmed by fastp³². Cutadapt³³ was then applied for demultiplexing and removal of primer sequences, and dada2³⁴ was finally used to generate their ASV sequences. The most abundant ASV from each sample was selected as the DNA barcode sequence for further analyses.

Data verification

For DNA barcodes obtained through the NGS strategies we evaluated the read abundance and proportion of the best ASV sequence per sample. Sequences with abundances below 30 reads for Illumina MiSeq and 10 reads for PacBio CCS were excluded from further analysis due to potential contamination. For Illumina MiSeq, we considered the process of index hopping, where indexed pair-end reads from different samples could contaminate each other. Additionally, samples with the best ASV sequences accounting for proportions lower than 0.9 were inspected to identify potential contamination by nonspecific PCR products or other DNA sources. By these, some rbcL amplicons were found likely to contain plastid-derived copies from mitochondrial genomes. In such cases, we manually retrieved sequences from the original ASVs and found the correct copy. Finally, for both rbcL and trnL-F barcodes, we used MUSCLE³⁵ to align all DNA sequences, and FastTree2³⁶ to reconstruct a preliminary phylogeny. These analyses aimed to identify samples that were potentially misidentified or mislabelled. For the formal phylogeny, we first aligned all specimens within each family using MAFFT³⁷. We merged these family-level trnL-F alignments into a single alignment using the MAFFT–merge argument, and aligned rbcL with the same outgroups as those used in Nitta et al.¹⁵. We then inferred maximum-likelihood (ML) phylogenies each with 1000 ultrafast bootstrap replicates (UFBS)³⁸ using IQTREE³⁹

Data Records

In total, we included 1,492 rbcL and 1,340 trnL-F DNA barcode sequences (Table 1) from 956 identified species across 152 genera, 34 families, and 11 orders. Among them, 22 and 23 species/taxa belong to hybrids and species complexes, respectively (Supplementary Tables 3 and 4). Except for 21 rbcL and 12 trnL-F sequences published earlier (Supplementary Table 5), all are newly generated in this study. Sequence information including their voucher, GenBank accession numbers, and sequencing methods are provided in Supplementary Table 5. Three alignment files including all DNA barcode sequences are available on Figshare⁴⁰. More detailed voucher information, including specimen records from TAIF and links to voucher images, is provided in the GBIF occurrence dataset⁴¹, in which DNA barcode sequences are also gathered. The raw reads resulting from Illumina MiSeq and PacBio libraries had been deposited in NCBI Sequence Read Archive (SRA)⁴².

Table 1.

Number of DNA barcode sequences by different sequencing methods.

	rbcL	trnL-F
Sanger sequencing	305	379
PacBio CCS	239	241
Illumina MiSeq	948	720

Technical Validation

The ML trees generated from the two DNA barcodes are available on Figshare⁴⁰ and Fig. 2. These phylogenies provided strong resolution and branch supports identifying the systematic placement of 1,496 fern samples (Fig. 2), and aligned well with modern classification of ferns^1,43–45. From the family-level backbone, the combined rbcL + trnL-F tree shows over 90% of nodes with UFBS values above 90. Individually, the rbcL and trnL-F trees resolved 84% and 67% of nodes with similarly high supports. At the intergeneric level, the rbcL + trnL-F phylogeny resolved 89% of nodes with UFBS values above 90, while single-barcode rbcL and trnL-F trees resolved 83% and 81% with such high support. At the infrageneric level, the rbcL + trnL-F tree supported the monophyly of 75% of species with multiple collections, after we excluded hybrids and species-unidentified samples. On this scale, the rbcL and trnL-F trees respectively supported monophyly in 72% and 73% of species.

Fig. 2.

Maximum-likelihood rbcL + trnL-F phylogeny noted with the order-level taxonomy sensu PPG I. For details about each test, see Methods and Technical Validation. The fern diagrams were downloaded from https://www.phylopic.org/ or modified from Vasco et al.⁴⁷ and Dong et al.⁴⁸. The Psilotales and Equisetales are not colored.

Author contributions

L.-Y.K. designed this study and managed the progress of the project; L.-Y.K., C.-Wei C., Z.-X.C., T.-C.H., Y.-H.C., Y.-S.C., H.T.L., T.-T.K., A.M.A.M., F.P.C., V.B.A., and Y.K.T. conducted the field works and collected specimens and the DNA samples; L.-Y.K., C.-Wei C., Z.-X.C., T.-C.H., Y.-H.C., Y.-S.C., S.F., and M.S. identified the specimens; L.-Y.K., Y.-H.H., P.-J.X., and T.-T.K. performed the experiments; L.-Y.K., S.-K.T., C.-Wei C., J.H.N. complied the datasets; J.H.N. and L.-Y.K. analysed the data; L.-Y.K. prepared the draft of manuscript with the significant inputs from M.S., T.-T.K., S.F. and J.H.N.

Code availability

The customized shell script for demultiplexing and removal of primer sequences within NGS reads using cutadapt (v.3.5), and the R (v.4.2.0) script for the generation of dada2’s ASV are available at https://github.com/lykuofern/1.5KP_datapaper. The pipeline for technical validation was designed using the ‘targets’ R package⁴⁶ and is available from https://github.com/joelnitta/bifa_barcodes. A Docker image to run the code is available from https://hub.docker.com/r/joelnitta/bifa_barcodes.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41597-024-04161-8.