The Parrot Genome: De-Novo Sequencing and Assembly Of a Novel Avian Genome To Identify Regulatory Traits For Vocal Learning

Vocal learning is a rare trait necessary for learning how to speak and sing in humans and in several groups of birds: songbirds, hummingbirds, and parrots. Recent data indicate convergent molecular changes in genes involved in neural connectivity and brain metabolism in vocal learning birds and humans. These include changes in microsatellite DNA in regulatory promoter regions, which are difficult to sequence and assemble. Thus, we need genome sequencing and assembly approaches that can correctly read and assemble the regulatory regions of a genome. Here we sequenced the genome of the budgerigar, using multiple sequencing and assembly technologies. We used 15X coverage 454 medium (400 bp) and long (750 bp) reads, and IIlumina short (75 bp) reads, in both shotgun and paired end libraries (insert sizes ranging from 0.5-20Kb) and 7X Pacific Biosciences reads (mode read length of 1.3kbp). We generated assemblies using Celera Assembler (CABOG) and Newbler algorithms, and assessed assembly accuracy with highly homologous genes across birds and targeted single molecules of parrots. We estimate that 8X 454 coverage was enough to sequence over 95% of the ~1.2Gb parrot genome. Both assembly algorithms were able to generate scaffolds the sizes of chromosomes, but Celera outperformed Newbler in most metrics, generating assemblies that exceeded contiguity metrics of the published zebra finch and chicken draft genomes. Both algorithms assembled standard complexity genes correctly, but Celera assembled more of them. However, sequences with long (>500 bp) repetitive microsatellites were misassembled whenever only reads lengths of 400 bp or less were used. This problem was corrected by adding the longer 750 bp reads. Sequence reads with 60-70% GC, including those with known functional exons, were not well assembled by any of the algorithms. Stretches of genomic DNA > 500bp with 70-98% GC were sequenced at low abundance, or not all at, by all of the sequencing chemistries tested, resulting in poor assembly of the GC rich regions. These GC rich regions account for many of the gaps in the assemblies. Addition of Illumina paired end reads to 454 reads resulted in breakdown of scaffolds in a hybrid assembly. This problem was fixed by not using the Illumina paired-end information, but still using the Illumina reads for increasing coverage and accuracy. Importantly, we have been able to assemble the egr1 promoter region, an immediate early gene studied in vocal learning birds, which had not been assembled for the published zebra finch and chicken genomes. Our findings help generate a blueprint for lower cost sequencing and more accurate assembly of genomes for analyzing regulatory regions.