De-novo whole genome assembly of the orange jewelweed, Impatiens capensis Meerb. (Balsaminaceae) using nanopore long-read sequencing

Sampling and long read sequencing of Impatiens capensis

Tissue (leaf) samples from a specimen maintained at the Kemper Missouri Native Shade Garden, (Missouri Botanical Garden, Missouri, USA; Accession number 2001-1051) was obtained by one of the authors (SRG) and shipped to Dovetail Genomics (Scotts Valley, CA, USA) for sequencing. (While the sequencing and Hi-C refined assembly was outsourced to Dovetail, the bioinformatics for the annotation and assembly QC/summaries were done in-house by the authors). Approximately 0.98 g of plant tissue was obtained, and genomic DNA (gDNA) extracted from the sample using the Nanobind Plant Nuclei Big DNA Kit (Circulomics Inc., San Diego, CA, USA), following the protocol detailed in Workman et al. (2018) that was specifically developed to obtain high molecular weight (HMW) gDNA from recalcitrant plant species for long read sequencing. (We would like to note here that multiple attempts at whole genome sequencing of I. capensis using PacBio technology by Dovetail proved unsuccessful. We understand from them that this was likely due to the unexpected and repeated premature separation of the polymerase from the library). A total of 11,200 ng of HMW gDNA was extracted from the sample, determined by measuring gDNA concentration with a Qubit 4 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA). Next, 100 ng of the DNA was used for library preparation, using the 1D library preparation protocol (Oxford Nanopore Technologies, Oxford, England), following the manufacturer’s instructions. The library was sequenced for 48 h. on a Minion portable device with a MinION R9 flow cell (FLOMIN106, R9.4 chemistry) (Oxford Nanopore Technologies, Oxford, England), controlled with the software MiniKnow. The raw signal (FAST5 files) was basecalled using the software Guppy, version 4.0.1 (Oxford Nanopore Technologies, Oxford, UK) and a total of 3,311,362 sequencing reads in FASTQ format were obtained (available in the short read archive (SRA) repository at GenBank: accession number SAMN30713540).

De novo genome assembly using long reads

The nanopore ONT sequencing run generated a total of 3,311,362 reads (17.264 Gbp, read N50: 13,385 bp, longest read: 142,336 bp, mean read length: 5,213 bp, mean read quality (Q-score): 9.3, median read length: 2,116 bp, median read quality: 9.1). QC steps followed using porechop and fastp resulted in a total of 3,311,956 “clean” reads. Summary details regarding the assembly can be found in Fig. S1, which gives a BlobPlot of the assembly, with the base coverage of the sequence on the Y-axis and the GC content on the X-axis, along with the corresponding histograms. Sequences are represented by circles in the scatterplot that are colored by taxonomic group and the diameter of each circle is proportional to the sequence length. The count, span, and N50 values are given for each taxon in the legend. BlobTools QC indicated relatively low contamination in the assembled genome with Proteobacteria, Ascomycota, Arthropoda, Chordata, Bactereoidetes, and Chlorophyta (Fig. S1), which were filtered out from the assembly.

The circular snail plot describing the assembly statistics for the de novo genome sequence of I. capensis is shown in Fig. 2 and summarized in Table 1. The table shows the statistics of the de novo assembled genome after adapter-trimming, reads cleaning, and decontamination. It comprised 8,612 contigs that were a total of 725,987,335 bp in length, with N50 of 848,459 bp, and N90 of 29,117 bp. The assembled genome size of I. capensis in our study is slightly smaller than the 817 Mb reported for the same species when estimated using flow cytometry (Bai et al., 2012), but is well within the observed range reported for other congeneric species (range: 414 Mb in I. pallida—(ibid.) to 3,188 Mb in I. omeiensis (Vesely et al., 2012)).

The contigs were further assembled into scaffolds by Dovetail using their proprietary HiRise software (Putnam et al., 2016) in conjunction with Hi-C data, which accounted for 574,941,529 bp of the genome (or 79.19% of the genome length), as shown in Table 1 (right column). The N50 and N90 statistics for the scaffold-level assembly were 32,415,000 and 85,265 bp, respectively. The total number of scaffolds assembled from the 8,612 contigs was 2,594, out of which, 19 were greater than 1 mb in length. There was a slight change in the GC content in the scaffold-level assembly, likely due to the 20% of the genome that was not assembled into scaffolds.

Another assessment of the quality of the genome assembly is in terms of the extent to which a set of orthologous genes that are universal to a given taxonomic clade, known as BUSCOs (Manni et al., 2021), can be identified. The I. capensis genome was benchmarked using the set of 2,326 BUSCOs in the clade, Eudicota. Table 1 shows the numbers (and the corresponding percentages) of these BUSCOs identified under various categories by the contig and scaffold assemblies. Even though the BUSCO statistics are relatively low, they are impressive considering that the sequencing and assembly were entirely de novo, with a divergence time between Impatiens and the reference genome of tomato as high as 108 million years (Kumar et al., 2022).

An alternative graphical view of the genome assembly was provided by Dovetail Genomics in their HiRise scaffolding report in the form of a “link density histogram” (Fig. 3A), which plots the position of each mate pair (the two ends of the sequenced DNA fragments) with the position of the corresponding sequence in the assembly. Although the diploid number of chromosomes in I. capensis is known to be 20 (n = 10) from cytological observations (Zinov’eva-Stahevitch & Grant, 1984), our assembly appears to have converged on 13 large scaffolds (Fig. 3A). Accordingly, we present the lengths of these 13 scaffolds (as well as the 14^th scaffold—only to demonstrate the dramatic drop in length) in Table 2 and a circular view of these 13 scaffolds (which constitute most of the genome; Fig. 2) in the form of a Circos plot with five tracks (Fig. 3B), created using the circlize package in R as described in Gu et al. (2014). Interestingly, all gene-related tracks showed elevated values at both edges of each scaffold—a curious feature that appears to be true in other plants as well, such as pineapple (Yow et al., 2022), maize (Lin et al., 2021), and cotton (Peng et al., 2022). Unsurprisingly, there appears to be some associated increase in the GC content at the scaffold edges as well. The figure also shows the composition of the four nucleotides, depicted in the form of a pie chart in the center of the Circos plot.

Genome annotation of I. capensis

The non-redundant reference gene set in the nuclear genome of I. capensis contained a total of 26,921 protein-coding genes and 79.10% (n = 21,294), 79.27% (n = 21,340), 61.77% (n = 16,630), 80.97% (n = 21,799), and 92.22% (n = 24,826) of these genes with structural annotation matched entries in the NCBI refseq plant (Protein), InterPro, Pfam, SignalP, and TargetP databases, respectively. As expected, there were multiple annotations per protein coding gene. A total of 399 rRNA genes were annotated using RNAmmer, and 978 tRNAs were detected by tRNAscan-SE.

Table S1 shows the annotation of the I. capensis genome with respect to repeat elements. It lists the number and length for each type of repeat element found, along with the percentage of the genome covered by each element. The analyses conducted in the platform GenSAS indicated that 71% (513,815,555 bp) of the genome was composed of repetitive elements, of which, 19.50% were not annotated. Considering only classified repetitive elements (80.50%), the most abundant were retroelements LTR Gypsy/DIRS1 (24.27%) and LTR Ty1/Copia (22.25%), which were considerably more common than the DNA transposons hobo-Activator (0.12%) and Tourist/Harbinger (0.12%). Low complexity DNA Satellites (n = 156; 0.001%) and rolling circles (n = 444; 0.07%) were much less common. The assembly also contained a total of 365,658 microsatellite sequences, most of them dinucleotides (n = 119,744).

Gene ontology based annotation

Among the three sources (Interproscan, BLAST, and DIAMOND), 22,773 genes out of the 26,921 genes included in the GFF file for the assembly received from Dovetail Genomics, were annotated in some way (Table S3), representing 84.6% of the predicted genes (Fig. S2A), although not all of them had GO annotations. In total, 19,671 unique predicted genes could be mapped to at least one GO term, representing 86.4% of the annotated genes and 73.1% of the predicted genes. Figures S2A–S2C show the overlap among the three sources of annotation of the predicted genes, the genes with at least one GO prediction, and those mapped to unique GO terms, respectively. Altogether, 3,466 unique GO terms mapped to these 19,671 predicted genes. Using Revigo (Supek et al., 2011), we removed redundant GO terms to obtain 3,167 unique terms, of which, 424, 1,144, and 1,599 were found to belong to the Cellular Component, Molecular Function, and Biological Process categories, respectively. The remaining are obsolete, duplicated, or otherwise absent from the GO ontology used by Revigo. Summing annotations from all three tools, these GO terms represent a total of 65,211 molecular function annotations corresponding to 20,046 unique genes, 28,391 biological process annotations corresponding to 12,696 unique genes, and 13,030 cellular component annotations corresponding to 7,851 unique genes. The number of gene annotations for the genome as a whole is larger than the number of genes in the genome as it includes multiple annotations made for each gene across all three GO namespaces; each gene annotation represents a unique combination of predicted gene and predicted GO annotation.

Figures 4A–4C are two-level pie charts that show the proportions of genes in each GO category and sub-category for molecular function, biological process, and cellular component, respectively. The overal counts are given in Fig. S3 and the break-up into categories and sub-categories can be seen in Tables S2A–S2C, for the unique GO terms associated with each cluster, as well as the number of unique genes associated with each cluster. Associated genes are given as percentages of the total number of predicted genes as well as percentages of the number of unique genes associated with each specific GO namespace.

Zinc ion binding and protein kinase activity appear to be the most common categories among the annotated genes, representing 21.54% and 17.64% of the genes annotated for molecular function (Fig. 4A and Table S2A). Interestingly, while protein kinase activity represented 31.38% of the terms out of the 1,144 GO annotations, zinc ion binding was seen in only 3.15% of them, indicating the disproportionately large number of genes involved in this function. While there did not appear to be any such glaring mismatches between the number of genes annotated for a given GO category and the number of GO annotations for that category in the case of biological function (Fig. 4B and Table S2B), unsurprisingly, a similar discord was seen for the cell membrane category in the cellular component GO term (Fig. 4C and Table S2C), attesting to the large number of genes involved in the membrane to regulate activities such as membrane trafficking (MartiniEre & Moreau, 2020).