Whole genome sequencing and assembly

Fresh young leaf was collected from a single A. rumphiana tree from the Ranong Biosphere Reserve, Thailand, and placed into liquid nitrogen. All necessary permissions were obtained for the sample collection. Frozen leaf was used for DNA extraction with the standard CTAB method followed by clean-up using a DNeasy Mini spin column from Qiagen.

DNA was used to prepare a linked-read library [11,12] using 10x genomics and sequenced on an Illumina HiSeq following Illumina protocols for 150 bp paired-end reads. De novo genome assembly was performed using supernova v2.1.1 [13] with default settings and the ‘pseudohap output’ style using an estimated genome size based on the A. marina genome (JACDXK000000000). Supernova includes a scaffolding step making use of the linked read data, so the output is referred to as ‘Supernova-scaffolds’ instead of ‘contigs’. The genome assembly was checked using Merqury v1.3 [14] with all raw reads and the Supernova-scaffolds allowing the program to calculate the optimum k-mer size. The genome was scaffolded using RagTag v2.0.1 [15] with A. marina (JACDXK000000000) as the reference to generate chromosome level scaffolds and these are referred to as ‘RagTag-scaffolds’ to avoid confusion with the output from Supernova. The default settings for RagTag were used with the instruction to not split contigs (Supernova-scaffolds) for cases where contigs (Supernova-scaffolds) had split mapping results. The reliability of this RagTag-scaffolding step was investigated by mapping the sequence of the RagTag-scaffold assembly against A. marina to identify any RagTag-scaffolds with discontiguous mapping. In addition, the published A. rumphiana genome (GWHBCJH00000000) [8] was scaffolded to A. marina using the same RagTag method to identify any cases of discontiguity.

Annotation

Annotation was performed for A. rumphiana on the repeat-masked genome using EvidenceModeler v1.1.1 [16] using gene sets and protein sets from related species and ab initio gene prediction. Protein sequences were obtained from public databases for Avicennia marina, Oryza sativa, Mimulus guttatus, Sesamum indicum, Populus trichocarpa and Eucalyptus grandis and aligned to the genome using AAT [17]. The ab initio prediction program Augustus v3.3.3 was used to predict genes based on the genome assembly [16]. All gene predictions were then combined by EvidenceModeler to generate consensus gene models using equal weights for each evidence type. Genome completeness was estimated by comparing each annotation gene set against the embryophyta_odb10 data set using Benchmarking Universal Single-Copy Orthologs (BUSCO) v4.0.5 [18]. The predicted gene sets for each species were functionally annotated using OmicsBox v2.0.10 (https://www.biobam.com/omicsbox) [19].

The annotated proteins were blasted to the A. marina proteins to identify homologous genes. Putative salt genes were identified by blast match to known salt genes based on the list of known salt genes from Natarajan et al. [5]. A two sample t-test was performed on the percent identity match of the blast result with all known salt genes as the ‘Salt’ group and all genes excluding the known salt genes as the ‘All’ group using the ‘t.test’ function in R.

Repeat sequence annotation

De novo repeat families were identified and classified for A. rumphiana using RepeatModeler version 2.0.1 [20]. These repeat sequences were then aligned to Genbank’s non-redundant protein database (using BLASTX with an E-value cutoff of 1 × 10 − 6) to exclude repeat sequences that contain large families of protein-coding genes.

Comparative genetics

The Supernova-scaffolds of A. rumphiana were compared to A. marina using the nucmer program of MUMmer v3.23 [21] to identify collinearity within Supernova-scaffolds to estimate the suitability of A. marina for reference based scaffolding. Nucmer was also used for the A. rumphiana Supernova-scaffolds of our assembly and the published A. rumphiana assembly (GWHBCJH00000000). The nucmer outputs of whole-sequence alignment were plotted in R to identify rearrangements relative to A. marina to produce Fig 1 and S1 Fig. Genome mapping statistics were calculated using the CIGAR string from mapping the Supernova-scaffolds, RagTag-scaffolds, or contigs (GWHBCJH00000000) against the reference (either A. marina genome or A. rumphiana PacBio contigs (GWHBCJH00000000)) using BWA v0.7.18 mem [22]. The CIGAR string was parsed to extract start and end locations of the mapped sequences, dropping unmapped reads, with the percent identity calculated from the number of insertions and deletions. To avoid double counting any individual base, the resulting alignment-interval table was then reduced using the GenomicRanges ‘reduce’ function in R to collapse overlapping intervals into a single interval. This reduction means that each base of the query sequence is counted only one time, even if it maps multiple times, allowing for calculation of the amount of query sequence that can map to the reference sequence. The reduce function loses the percent identity and location of each individual alignment, but the total percent identity of the mapped sequence can be estimated by taking the sum of each mapped sequence length multiplied by its percent identity divided by the sum of all mapped lengths.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Avicennia rumphiana RagTag-scaffold sequence mapped against Avicennia marina chromosome sequence.

The whole-genome sequence of the RagTag-scaffolds of A. rumphiana on the y-axes mapped against the whole-genome sequence of A. marina on the x-axes. Cases where multiple A. rumphiana RagTag-scaffolds map to a single A. marina chromosome appear as stacked boxes where the length of the y-axis is proportional to the amount of sequence that maps, resulting in small regions appearring as individual lines. Blue dots indicate direct sequence alignment, while red dots indicate inverted sequence alignment.

https://doi.org/10.1371/journal.pone.0318091.g001

We also mapped the raw 10x-sequencing reads against our genome assembly and the published genome assembly using BWA to report on the percent of raw reads that can map to each assembly.

The RAxML-NG program v1.0.2 [23] was used to construct a phylogenetic tree based on orthologous groups that were identified by OrthoFinder v2.5.5 [24]. The species included in the phylogenetic tree were Arabidopsis thaliana, Avicennia marina, Bruguiera cylindrica, Bruguiera gymnorrhiza, Bruguiera parviflora, Bruguiera sexangula, Ceriops decandra, Ceriops tagal, Ceriops zippeliana, Kendelia obovata, Populus trichocarpa, Rhizophora apiculata, Ricinus communis, and Oryza sativa. Protein sequences were aligned using MUSCLE [25] and a substitution model was estimated using the ModelTest-NG program v0.1.7 [26]. The divergence times in the phylogenetic tree were estimated by the Bayesian Relaxed Molecular Clock approach using MCMCtree v4.0 [27]. Divergence times were anchored using dated fossil records, the root node of the common ancestor of Rhizophoraceae, Euphorbiaceae (R. communis) and Salicaceae (P. trichocarpa) placed 105 - 120 MYA, fossils of the ancestor of Rhizophoreae from 48 - 56 MYA, and a fossil recognized as ancestral Rhizophora dated to 34 - 38 MYA. Colinear blocks were identified from the blastp alignment results using MCScanX [28] with regions of at least ten syntenic genes (with no more than six intervening genes allowed) considered as intra- or intergenomic homologous regions. Gene family size changes, based on orthogroups, were detected using CAFE5 [29] with the output from MCScanX to identify expansion and contraction of gene numbers within gene families and plotted onto the phylogenetic tree. GO term enrichment analysis of the expanded gene and contracted gene families was conducted using the TopGO package in R on the omicsbox GO annotations. Testing was performed following the topGO vignette and GO terms with low observed or expected counts ( < 5) were removed to avoid false positives.

The genes of our assembly and the published assembly (GWHBCJH00000000) were compared and grouped by performing two protein blasts (blastp -evalue 1e-06), one with each assembly as the reference, and a using CD-HIT v4.8.1 [30] with a similarity threshold of 95%. Groups from CD_HIT were analysed to identify singleton genes from each assembly and cases where multiple genes from one assembly grouped to a single gene from the other assembly.

Population analysis

Leaf samples of 29 A. rumphiana trees were collected from three sites in Thailand: Prachuap Khiri Khan, Chumphon, and Surat Thani. A reduced representation library was prepared from genomic DNA for each sample following the MGIEasy DNA Library Prep Kit Instruction Manual (MGI Tech, Shenzhen, China) using the restriction enzyme TaqI. The resulting reads were mapped to our A. rumphiana genome assembly using minimap2 v2.24 [31] with default parameters and variants were called using GATK v4.1 [32] with default parameters. The variants were then filtered to remove calls with less than 10 reads and to select SNPs where at least 20 of the 29 samples received a call. A principal components analysis was performed for the SNPs where all samples received a call using the ‘prcomp’ function in R and plotted using the R package ggplot2. The polymorphic information content of each SNP was calculated as a dominant marker type [33]. Population structure was calculated using the program STRUCTURE [34] with settings of no missing, minimum read depth of 10, and minor allele frequency set to 1%. The optimum k-value was determined from a k-value analysis using numbers from one to ten using 20 replicates per k-value.

Genome assembly, scaffolding and comparison

The genome assembled into 24,094 Supernova-scaffolds totalling 499.6 Mb, the largest Supernova-scaffold was 2.74 Mb and the N50 was 294 kb, from 110.9 Gb of paired-end sequence data. Merqury analysis showed that the genome assembly was 95% complete with optimum k-mer size 19.43 and had a k-mer multiplicity of one (S1 Fig). Our assembly was similar to the published A. rumphiana genome (GWHBCJH00000000), which was 470.6 Mb in 304 contigs with the longest contig at 21.1 Mb, it was assembled using PacBio, but not scaffolded into chromosomes [8]. To investigate the size difference between assemblies we mapped our assembly against the published assembly (GWHBCJH00000000) and found that 91.2% of our assembly mapped to the published assembly (GWHBCJH00000000) with an average identity of 99.89%, consistent with these being two individuals of the same species. However, there were 17,727 Supernova-scaffolds, totalling 41.16 Mb, that did not map to the published genome assembly (GWHBCJH00000000) and 16 Mb (3.4%) of the published genome (GWHBCJH00000000) had no sequence that mapped to it. Both our assembly and the published assembly (GWHBCJH00000000) contained mitochondrial and chloroplast sequence, so the disparity is not a result of one assembly simply lacking these sequences. Mapping our raw reads to each assembly showed a 95.7% maprate to our assembly and a 93.3% maprate to the published assembly (GWHBCJH00000000). The Supernova-scaffolds were blasted against the NBCI nt database to exclude sequences that represented sample contamination and, therefore, represent novel presence/absence allelic variation. The blast results showed no major hits to any known sequence, however, a small percentage of the internal sequence of some large Supernova-scaffolds matched bacterial sequence at an average identity of 70-80%, suggesting the sequence integrated with the A. rumphiana genome at some point in the past.

We mapped the Supernova-scaffolds of our A. rumphiana assembly against the A. marina genome to identify sequence similarity. Approximately 68.7% of the A. rumphiana genome sequence can be aligned to A. marina sequence, covering 72% of the A. marina genome, with 87.7% average nucleotide identity. Accounting for unmapped sequence gives an overall genome similarity of 60.3% at the whole genome nucleotide level. To add some perspective to this statistic, we have previously applied the same whole-genome-mapping approach to two species of Brugiera that can produce a fertile hybrid offspring and found that approximately 73% of the sequence of one can map to the other [35]. In addition, we have mapped a genome assembly of a Thai Sonneratia ovata tree to the genome assembly of a Chinese Sonneratia ovata tree and found that 92% of one can map to the other [36]. This high similarity at the whole-genome nucleotide level shows how closely related these two species are, which is not surprising considering they occupy the same environment niche and were originally identified as the same species.

The genome size of A. rumphiana is similar to that of A. marina, which is 457 Mb reported by Natarajan et al. [5], 456 Mb reported by Friis et al 2021 or 480 Mb reported by Ma et al. [7], so reference based scaffolding of the genome assembly was investigated. The published A. rumphiana assembly (GWHBCJH00000000) was less fragmented and had the largest contigs, so the entire genome assembly of this was mapped against A. marina using NUCmer and plotted. The result showed a high level of whole-seqence contiguity to A. marina (S2 Fig) with each contig mapping to a single A. marina chromosome. Six of the largest contigs are completely contiguous to and cover a full A. marina chromosome, with the majority of chromosomes covered by 2–4 contigs. Some chromosomes show small inversions that occur towards the middle of the chromosome, which likely represent centromere sequence. All 32 A. marina chromosomes are covered. If the A. rumphiana genome was structurally different to A. marina we would see parts of each A. rumphiana (GWHBCJH00000000) contig mapping to multiple A. marina chromosomes, especially the contigs that represent entire A. marina chromosomes. This high sequence contiguity shows that A. marina was not only similar enough to A. rumphiana to allow for it to be used as a reference for reference based assembly, but that it was remarkably similar considering how long ago the two species diverged. Following from this finding we used A. marina as a reference to scaffold our assembly of A. rumphiana and obtained 32 chromosome-level RagTag-scaffolds totalling 447.3 Mb, leaving 52.3 Mb of unplaced sequence (Fig 1). The resulting RagTag-scaffolds were aligned back to A. marina and shown to have the same level of sequence contiguity as the published A. rumphiana genome (GWHBCJH00000000).

Genome annotation and salt genes

The A. rumphiana genome was annotated and found to have 37,347 genes with 22,414 of those contained within RagTag-scaffolds (S1 File). Functional analysis of these genes identified 30,511 genes with a known annotation. The gene set was assessed for completeness using BUSCO and found to be 94.5% complete with 89.5% single copy and 5% duplicated. This is a larger number of genes compared to the published A. rumphiana genome (GWHBCJH00000000) or the A. marina genome, which had 31,499 and 31,477 genes, respectively. The larger number of genes between the published A. rumphiana genome (GWHBCJH00000000) and ours can be explained by a difference in gene annotation parameters and the larger size of our assembly. Protein blast of the two A. rumphiana annotations against each other identified 21,380 genes in common, and 21,269 of these received a functional annotation (S1 Table). Grouping the genes using CD_HIT identified 50,529 gene clusters with 14,867 gene clusters that contained at least one gene from each assembly. Of the 14,867 gene clusters, there were 13,558 clusters that contained a single gene from each assembly and 1309 clusters that contained multiple genes. The multi-gene clusters contained 2576 genes from our assembly and 2086 genes from the published assembly and there were multiple cases where several smaller genes from our assembly grouped to a much longer gene from the published assembly, which partly explains why our assembly contained more genes. There were 19,388 genes from our assembly and 15,347 genes from the published assembly that formed single gene clusters. The majority of the genes from our assembly that did not receive a functional annotation (6164 of the 6536 genes) were in this group of single gene clusters.

Comparison of the genes showed that 22,554 A. rumphiana genes were homologous to 22,029 A. marina genes, which leaves a large number of genes that are too different between the species to be identified as homologous. This may make it more simple to identify salt genes that both species have in common, but makes it more likely that the exact mechanism of salt tolerance is at least slightly different between A. rumphiana and A. marina. The homologous gene set was searched for salt genes listed in Natarajan et al. [5] to identify salt genes that both species had in common. The list of A. marina salt genes contained 614 genes that were identified as playing some kind of role in salt tolerance and there were 671 A. rumphiana genes that had a best-hit blast match to 467 of these genes. This left 205 A. marina genes that did not have a best-hit A. rumphiana match, however a blast of A. marina against A. rumphiana shows that only 98 of these salt genes truly had no A. rumphiana match. Considering how much convergent evolution has occurred in mangrove species, this does not necessarily mean that the most important salt tolerance genes are included in the list of homologous genes. Nevertheless, it is interesting to identify which known salt genes are most similar between the two species and which, if any, have experienced gene duplication. Not only was there a large difference between the number of salt genes that were similar enough to be identified as a homologue, but the mean percentage similarity of the salt gene homologues was lower (86.8%) than the mean percent similarity of the entire gene set (91.1%), showing that the salt genes are more diverse than would be expected based on the entire gene set (p-val 4.12e–11). This is particularly surprising considering the high similarity in salt tolerance mechanisms and recent divergence time.

The gene families that form the salt secreting glands on the leaves were identified in the A. marina genome as involving ion channels, aquaporins, plasmodesmata, and vesicles [5]. Nine genes or gene families were identified in the A. marina genome as contributing to the salt secreting gland. To identify homologues of these genes, we performed a blast search using the A. marina gene and identified homologues based on sequence similarity. There were 11 genes identified as sodium/hydrogen exchanger or cation/proton exchanger, 18 genes identified as cation/H(+) antiporter, and a single gene identified as a cation-chloride cotransporter. Five genes were identified as S-type anion channel, with 19 genes identified as a cyclic nucleotide-gated ion channel, and 17 genes annotated as a glutamate receptor. There were 28 genes identified as an aquaporin, 7 genes identified as a chloride channel, and 5 genes identified as sodium transporter HKT1. How much of a role each of these genes play, if any, in salt excretion remains to be elucidated.

Phylogenetic tree

A phyogenetic tree of 76 mangrove and mangrove related species was recently published [8]. Here we generated a phylogenetic tree, using single copy orthologues, of our A. rumphiana assembly compared to some mangroves that were newly sequenced since the previous large phyogenetic tree, plus a selection of previous samples to aid comparison (Fig 2). Mangroves are a prominent example of convergent evolution, so comparing the gene content of various mangrove species may give some insight into salt tolerance evolution. The results suggested that A. rumphiana and A. marina separated from each other 17.8 million years ago (MYA), consistent with the previous finding. The gene orthologue set was used to estimate gene expansion or contraction of gene numbers within gene families using a birth and death process to model gene gain and loss across this phylogenetic tree (Fig 2). A. rumphiana gained genes in 45 gene families and lost genes in 357 gene families, while A. marina gained genes in 265 gene families and lost genes in 59. GO term enrichment analysis of the expanded gene families showed enrichment for protein serine/threonine kinase activity and DNA-binding transcription factor activity as well as several underrepresented GO terms (S2 Table). While contracted gene familied were enriched for DNA binding, protein binding, mRNA binding, and also protein serine/threonine kinase activity. While the over- and under-represented GO terms are biologically important, they occur in too many different pathways to offer any specific insight into the evolution of this species.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Maximum-likelihood phylogenetic tree of Avicennia rumphiana.

Divergence times (million years ago) are indicated at each node. The number of gene families that have gained or lost genes are indicated as blue and red numbers, respectively.

https://doi.org/10.1371/journal.pone.0318091.g002

Repeat sequence analysis

The total interspersed repeats accounted for 46% of the genome sequence and the most common known repeat type was Ty1/Copia and Gypsy/DIRS1 LTR elements (Table 1). The majority of repeats were in the category ‘unclassified’ and account for approximate 25.6% of the genome followed by LTR elements. Approximately 3% of the genome consists of DNA transposons, with the most common identified sequence being hobo-Activator and Tourist/Harbinger transposons.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Repeat sequence statistics for the assembly of Avicennia rumphiana.

https://doi.org/10.1371/journal.pone.0318091.t001

Population analysis

Proper management of mangrove species, which are being threatened by rising sea levels and deforestation, can be aided by measuring the genetic diversity of each species in conservation sites. A population with a low level of genetic diversity can be at increased risk of collapse from mechanisms such as inbreeding depression and reduced adaptability to biotic or abiotic stresses. Genetic diversity of A. rumphiana in Thai mangrove conservation sites were estimated by SNP variation present in trees sampled from multiple sites in Thailand. The total filtered SNP set contained 7,382 SNPs where all 29 samples were called. The polymorphic information content (PIC) ranged from 0.034 to 0.5 with the median PIC being 0.44, indicating high genetic diversity. A principle components analysis (PCA) showed a loose clustering effect of the samples according to the geographic location from which they were sampled (Fig 3), whereas STRUCTURE analysis showed two mixing populations with Chumpon and Surat Thani forming a single group (S3 Fig). Increasing the K-value in STRUCTURE added separation to the Prachuap Khiri Khan samples rather than the Chumpon and Surat Thani samples. The samples were collected from the east coast of three bordering Thai provinces in the Malay peninsula: Prachuap Kiri Khan in the north, Chumpon, and Surat Thani in the south. There is some overlap between samples from Prachuap Kiri Khan with those from Chumpon, consistent with trees coming from neighboring locations, but otherwise the samples show separation based on location. The PIC values and PCA results showing sample separation show that a healthy level of genetic variation exists in the population. These findings support the IUCN rating of ‘vulnerable’ for this species, which is the least severe status in the ‘threatened’ category.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Principal component analysis of 29 Avicennia rumphiana trees from Thailand.

Individuals are color coded according to location on the east coast of the Malay Peninsula for the three bordering provinces of Prachuap Kjiri Khan (north), Chumpon, and Surat Thani (south) in Thailand. The sum of variance explained by PC1 and PC2 is 66.25%.

https://doi.org/10.1371/journal.pone.0318091.g003