Comparative genomic data of the Avian Phylogenomics Project

Background The evolutionary relationships of modern birds are among the most challenging to understand in systematic biology and have been debated for centuries. To address this challenge, we assembled or collected the genomes of 48 avian species spanning most orders of birds, including all Neognathae and two of the five Palaeognathae orders, and used the genomes to construct a genome-scale avian phylogenetic tree and perform comparative genomics analyses (Jarvis et al. in press; Zhang et al. in press). Here we release assemblies and datasets associated with the comparative genome analyses, which include 38 newly sequenced avian genomes plus previously released or simultaneously released genomes of Chicken, Zebra finch, Turkey, Pigeon, Peregrine falcon, Duck, Budgerigar, Adelie penguin, Emperor penguin and the Medium Ground Finch. We hope that this resource will serve future efforts in phylogenomics and comparative genomics. Findings The 38 bird genomes were sequenced using the Illumina HiSeq 2000 platform and assembled using a whole genome shotgun strategy. The 48 genomes were categorized into two groups according to the N50 scaffold size of the assemblies: a high depth group comprising 23 species sequenced at high coverage (>50X) with multiple insert size libraries resulting in N50 scaffold sizes greater than 1 Mb (except the White-throated Tinamou and Bald Eagle); and a low depth group comprising 25 species sequenced at a low coverage (~30X) with two insert size libraries resulting in an average N50 scaffold size of about 50 kb. Repetitive elements comprised 4%-22% of the bird genomes. The assembled scaffolds allowed the homology-based annotation of 13,000 ~ 17000 protein coding genes in each avian genome relative to chicken, zebra finch and human, as well as comparative and sequence conservation analyses. Conclusions Here we release full genome assemblies of 38 newly sequenced avian species, link genome assembly downloads for the 7 of the remaining 10 species, and provide a guideline of genomic data that has been generated and used in our Avian Phylogenomics Project. To the best of our knowledge, the Avian Phylogenomics Project is the biggest vertebrate comparative genomics project to date. The genomic data presented here is expected to accelerate further analyses in many fields, including phylogenetics, comparative genomics, evolution, neurobiology, development biology, and other related areas. Electronic supplementary material The online version of this article (doi:10.1186/2047-217X-3-26) contains supplementary material, which is available to authorized users.


Data description
Here we presented the genomes of 48 bird species, representing 36 orders of birds, including all Neognathae and two of the five Palaeognathae orders, collected by the Avian Genome Consortium ( [1], full author list of the Consortium provided in Additional file 1 and data in GigaDB [2]). The Chicken, Zebra finch, and Turkey genomes (sequenced using the Sanger method) were collected from the public domain. Another three genomes, the Pigeon, Peregrine Falcon and Duck, have been published during the development of this project [3][4][5], and five genomes, the Budgerigar, Crested Ibis, Little Egret, Emperor and Adele penguins, are reported in companion studies of this project [6,7]. The data downloads for the remaining 38 genomes are released here.

Genome sequencing
Tissue samples were collected from multiple sources, with the largest contributions from the Copenhagen Zoo (Denmark) and the Louisiana State University (USA). Most DNA samples were processed and quality control performed at the University of Copenhagen (Dr. Gilbert's lab, Denmark) and Duke University (Dr. Jarvis' lab, USA). The collected samples were then used for constructing pair-end libraries and sequenced using Illumina HiSeq 2000 platforms at the BGI (China). For the highcoverage birds, multiple pair-end libraries with a series of up to 9 insert sizes (170 bp, 500 bp, 800 bp, 2 kb, 5 kb, 10 kb and 20 kb) were constructed for each species, as part the first 100 species of the G10K project. For four birds (Anas platyrhynchos, Picoides pubescens, Ophisthocomus hoazin and Tinamus guttatus), libraries of some insert sizes were not constructed due to limited sample amounts or the sequencing strategies applied to those species. In addition, for the budgerigar genome, Roche 454 longer reads of multiple insert sizes were used [6]. For the low-coverage genomes, libraries of two insert sizes (500 bp and 800 bp) were constructed. The sequencing depths for high-coverage genomes were 50X to 160X, whereas the sequencing depths for low-coverage genomes were 24X to 39X. An effort was made to obtain DNA samples from tissues with associated museum voucher specimens with high quality metadata.

Genome assembly
Before assembly, several quality control steps were performed to filter the low-quality raw reads. The clean reads of each bird were then passed to SOAPdenovo v1.05 [8] for de novo genome assembly. We tried different k-mers (from 23-mer to 33-mer) to construct contigs and chose the k-mer with the largest N50 contig length. In addition, we also tried different cut-offs of read pairs for different libraries to link contigs into scaffolds. The assembly with the largest N50 length was finally used. All the assemblies have similar genome sizes, ranging from 1.04-1.26Gb (Table 1). The high-coverage genomes have a N50 scaffold length of >1 Mb, except for the White-throated Tinamou (Tinamous guttatus) with a scaffold N50 of 242 Kb and Bald Eagle (Haliaeetus leucocephalus) with a scaffold N50 of 670 Kb, due to no 10 kb and 20 kb libraries for these two genomes. For lowcoverage genomes, the scaffold N50 lengths ranged from 30 kb to 64 kb. The N50 contig lengths for high-coverage genomes were from 19 kb to 55 kb, and the low coverage genomes were from 12 kb to 20 kb. The Parrot and Ostrich genomes were further assembled with the aid of optical mapping data, thus achieving much larger scaffold N50 sizes.

Repeat annotation
RepeatMasker [9] and RepeatModeler [10] were used to perform repeat annotations for the bird genomes. The overall annotated content of transposable elements (TE) range from within 2-9% of all bird genomes except Woodpecker (

Protein-coding gene annotation
We used the homology-based method to annotate genes, with gene sets of chicken, zebra finch and human in Ensembl release 60 [11]. Because the quality of homologybased prediction strongly depends on the quality of the reference gene sets, we carefully chose the reference genes for the annotation pipeline. The protein sequences of these three species were compiled and used as a reference gene set template for homology-based gene predictions for the newly assembled bird genomes. We aligned protein sequences of the reference gene set to each genome by TBLASTN and used Genewise [12] to predict gene models in the genomes. A full description of the homology-based annotations is in our comparative genomics paper [1]. All the avian genomes have similar coding DNA sequence (CDS), exon, and intron lengths (Table 3).

Syntenic-based orthlogous annotation
To obtain more accurate orthology annotations for phylogenetic analyses in [13], we re-annotated some genes of the Chicken and Zebra Finch based on synteny, thereby correcting errors in the annotations due to being annotated independently with different methods. We first ran bi-directional BLAST to recognize the reciprocal best hits (considered as pairwise orthologs) between our re-annotated chicken genome and each of the other genomes. Then we identified syntenic blocks by using pairwise orthologs as anchors. We only kept the pairwise orthologs with syntenic support. In addition, we also considered the genomic syntenic information inferred from the LASTZ genome alignments, and removed pairwise orthologs without genomic syntenic support. After the above filtering, all the remaining pairwise orthologs were combined into a merged list by using a chicken gene set as a reference. We also required each orthologous group to have members in at least 42 out of 48 avian species. Ultimately, we obtained a list of 8295 syntenic-based orthologs. We used the same methods to generate 12815 syntenic-based orthologs of 24 mammalian species. A full description of the synteny-based annotations is found in our phylogenomics paper [13].

Sequence alignments
Protein coding gene alignment CDS alignments for all orthologous genes were obtained by two rounds of alignments. In order to preserve the reading frames of CDS, we aligned the amino acid sequences and then back translated them into DNA alignments. In the first round of alignment, SATé-Prank [14] was employed to obtain the initial alignments, which were used to identify the aberrant over-aligned and underaligned sequences. The aberrant sequences were then removed, and the second round of alignment were performed by SATé-MAFFT [14] for the filtered sequences to create the final multiple sequence alignments. The default JTT model inside SATé [14] was used as we found it to fit the data best for most genes. We also used the same method to generate the alignments of mammalian orthologs. More details of the alignment are presented in Jarvis et al. [13].

Whole genome alignment
Whole genome alignments are very useful for comparative analyses, so we generated a multiple genome alignment of all 48 bird species. Firstly, pairwise alignments for each two genomes (with repeats masked) were produced by LASTZ [15], using chicken as the reference genome. Next chainNet [16] was introduced to obtain improved pairwise alignments. Finally, we used MULTIZ [17] to merge the pairwise alignments into multiple genome alignments. Approximately 400 Mb of each avian genome made it into the final alignment result. Thereafter, the alignment was filtered for over-and under-aligned errors, and for presence in 42 of 48 avian species. The resultant alignment was about 322 Mb, representing about one third of each genome, suggesting a large portion of the genome has been under strong constraints after different bird species diverged from their common ancestor. More details of the alignment are presented in Jarvis et al. [13].

dN/dS estimates
We deposit dN/dS estimates (ratio of non-synonymous versus synonymous substitution rates) of the protein coding genes from Zhang et al. [1]. The dN/dS ratios were estimated by PAML [18] program for the orthologs. Based on the CDS alignment of either protein coding data set, we used the one-ratio branch model to estimate the overall dN/dS ratios for each avian orthologous group and each mammalian orthologous group. In addition, to investigate the evolutionary rates in three major avian clades (Palaeognathae, Galloanserae and Neoaves), we used the three-ratio branch model, which estimated one identical dN/dS ratio for each clade. More details about dN/dS analyses are presented in Zhang et al. [1].

DNA sequence conservation
The overall level of conservation at the single nucleotide level could be estimated by PhastCons [19] based on multiple sequence alignments (MSA). First, the four-fold degenerate sites were extracted from 48-avian MSA and were used to estimate a neutral phylogenetic model by phyloFit [20], which is considered as the non-conserved model in PhastCons; we then ran PhastCons to estimate the conserved model. The conservation scores were predicted based on non-conserved and conserved models. We also used this method to estimate the sequence conservation for the 18-way mammalian genome alignments from the University of California at Santa Cruz (UCSC). Additional details of genome conservation are presented in the comparative genomics paper [1].

List of scripts used in avian comparative genome project
We also deposit the key scripts used in the avian comparative genome project in GigaDB [2], which include: 1) scripts for cleaning raw reads and assembling the genome using SOAPdenovo; 2) scripts for RepeatMasker and RepeatModeler repeat annotation; 3) scripts for homology-based protein-coding gene annotation and combining the gene annotation evidences into final gene sets; 4) scripts for generating whole genome alignment of multiple genomes; 5) scripts for running PAML to estimate branch model dN/dS ratios; 6) scripts for calculating conservation scores based on whole genome alignments and predicting highly conserved elements; 7) scripts for quantifying gene synteny percentages in birds and mammals; 8) scripts for identifying large segmental deletions from list of orthologous genes; 9) scripts for detecting gene loss in 48 avian genomes. We provide readme files in the script directories describing the usage of the scripts.

Availability and requirements
Download page for scripts: https://github.com/gigascience/paper-zhang2014 Operating system: Linux Programming language: Perl, R, Python Other requirements: Some pipelines need external bioinformatics software, for which we provided executable files in the directories. License: GNU General Public License version 3.0 (GPLv3) Any restrictions to use by non-academics: No