Assembly of chloroplast genomes with long- and short-read data: a comparison of approaches using Eucalyptus pauciflora as a test case

Background
Chloroplasts are organelles that conduct photosynthesis in plant and algal cells. The information chloroplast genomes contain is widely used in agriculture and studies of evolution and ecology. Correctly assembling chloroplast genomes can be challenging because the chloroplast genome contains a pair of long inverted repeats (10–30 kb). Typically, it is assumed that the gross structure of the chloroplast genome matches the most commonly observed structure of two single-copy regions separated by a pair of inverted repeats. The advent of long-read sequencing technologies should remove the need to make this assumption by providing sufficient information to completely span the inverted repeat regions. Yet, long-reads tend to have higher error rates than short-reads, and relatively little is known about the best way to combine long- and short-reads to obtain the most accurate chloroplast genome assemblies. Using Eucalyptus pauciflora, the snow gum, as a test case, we evaluated the effect of multiple parameters, such as different coverage of long-(Oxford nanopore) and short-(Illumina) reads, different long-read lengths, different assembly pipelines, with a view to determining the most accurate and efficient approach to chloroplast genome assembly.

Results
Hybrid assemblies combining at least 20x coverage of both long-reads and short-reads generated a single contig spanning the entire chloroplast genome with few or no detectable errors. Short-read-only assemblies generated three contigs (the long single copy, short single copy and inverted repeat regions) of the chloroplast genome. These contigs contained few single-base errors but tended to exclude several bases at the beginning or end of each contig. Long-read-only assemblies tended to create multiple contigs with a much higher single-base error rate. The chloroplast genome of Eucalyptus pauciflora is 159,942 bp, and contains 131 genes of known function.

Conclusions
Our results suggest that very accurate assemblies of chloroplast genomes can be achieved using a combination of at least 20x coverage of long- and short-reads respectively, provided that the long-reads contain at least ~5x coverage of reads longer than the inverted repeat region. We show that further increases in coverage give little or no improvement in accuracy, and that hybrid assemblies are more accurate than long-read-only or short-read-only assemblies.