Brown Lab Plant Ecology & Genomics

News ·

poa: annotating plant organelle genomes, splicing and editing included

A new command-line annotator for plant mitochondrial and plastid genomes that rebuilds spliced genes, corrects for RNA editing and annotates the assembly graph itself.

Annotating a plant organelle genome sounds like it should be a solved problem. Plastomes are small and well behaved, and there are good tools for them. Plant mitogenomes are different. Several of their genes are trans-spliced, with exons scattered hundreds of kilobases apart, sometimes on opposite strands, and only joined up at the RNA level. C-to-U RNA editing creates start and stop codons that are not in the DNA. And the genomes themselves are often not a single circle but a graph of alternative structures (as has been shown quite a few times now).

poa, the plant organelle annotator, is my attempt to deal with all of that in one pipeline. It grew out of annotating over 1000 species for the database behind our organelle atlas, and that database now uses it. This database sort of mirrors the typical DToL pipelines, but is only for plant organellar DNA - currently there are no ways to store GFA’s on NCBI etc.

What it does

  • Core genes, tRNAs and rRNAs: protein-coding genes are called with nhmmscan against OatkDB’s land plant gene-family models, tRNAs with tRNAscan-SE and rRNAs with plant rRNA models.
  • RNA editing: orfedit aligns each gene’s protein profile to the genome, allowing for C-to-U edits. It finds the true start and stop codons, including ones that only exist after editing.
  • Spliced genes: transsplice joins the exons of the trans-spliced nad1, nad2, nad5 and rps3, and the cis-spliced nad4, nad7, ccmFc, cox2, rpl2 and rps10. It decides for each species and each intron whether that junction is cis or trans.
  • The assembly graph: given the GFA, poa maps every gene onto the graph’s unitigs as well as the linear contigs, and writes a GFF3 for each.
  • Both organelles, in one command, from oatk or any other assembler.

Why it is a step forward

Most organelle annotators cover one organelle. For example GeSeq, CPGAVAS2, PGA and Chloë are built for plastomes. And PMGA, MITOFY and RhoMitoAnnotator are built for mitogenomes. Many of the best are web servers, which are hard to run on a thousand genomes. poa does both organelles, runs from the command line, and bundles all of its reference data, so it is built for whole datasets.

The closest end-to-end tool is probably oatk, which we use to assemble our genomes. oatk’s gene calls are a by-product of assembly, and running the same gene models directly and then reconstructing the genes makes a real difference:

  • Fragmented genes become whole genes. oatk only ever found one piece of nad5. Calling genes directly found all four exon fragments in every species in our test set, and roughly doubled the share of trans-spliced genes we could reconstruct completely.
  • Missing genes come back. In some species oatk called none of nad1, nad2, nad5 or rps3, or no genes at all, despite a good assembly. poa finds them.
  • Coordinates on the graph. As far as we know, poa is the only annotator that gives gene positions on the assembly graph’s unitigs as well as the linear contigs. That matters when the “genome” is a set of alternative structures rather than one sequence.

poa works quite well

We tested poa against the curated RefSeq annotation of the Arabidopsis thaliana mitogenome, using an independent PacBio assembly from the Darwin Tree of Life:

  • 54 of 55 annotated genes found. The 55th is a naming difference.
  • Edit-corrected proteins of single-exon genes match the curated proteins at 99.5 to 100% identity.
  • nad4 and nad7 are rebuilt with every exon boundary exact to the base, and nad2 and rps3 with all their exons.

Across the full dataset, poa rebuilds complete spliced genes in 55% (nad5) to 81% (rps10) of the species that have them. Its weakest point is mosses (which I’m working on correcting now). The templates are mostly built from flowering plants, so some moss genes are placed less well. Lineage-specific templates are the next step.

Please try it!

mamba env create -f environment.yml
conda activate poa
pip install .
poa run --mito sample.mito.ctg.fasta --mito-gfa sample.mito.gfa \
        --pltd sample.pltd.ctg.fasta --pltd-gfa sample.pltd.gfa \
        -o out/ -t 8

An Arabidopsis mitogenome and plastome take about five minutes on four cores. A bioconda package and container images are on the way. poa is on GitHub, and I’d love to hear how it does on your genomes.

As it stands, I’ll keep updating the database live on this site, and attempt to get the most comprehensive set of assemblies and annotations for plant organellar genomes out there.

All news