Curation of microarray oligonucleotides and corresponding ESTs/cDNAs used for gene expression analysis in zebra finches

Zebra finches represent a major model organism for studying vocal learning [1–6], a trait that provides a basis for spoken language acquisition in humans. Studies in finches have led to insights into the molecular machinery that underlies learned vocalizations [7–19], including the transcriptome of the vocal control circuitry [7, 8, 11–16, 18–25] and the identification of convergent molecular specializations of the vocal control systems of birds and humans [7]. Such studies were largely based on the Songbird array v2 [16], a ~ 44,000 60-mer oligonucleotide array designed with eArray 5.4 (Agilent Technologies) and sequences from three cDNA collections [11, 16, 23]. Initial cDNA annotations were made before the zebra finch genome was available through BLAST searches of annotated cDNA/EST databases. Later efforts aligned oligo and EST sequences to the zebra finch genome (Taegut1; [26]), and assigned Ensembl model annotations to oligos that mapped to within 5 kb (or ESTs within 3 kb) of those models [7, 25]. However, this effort did not take into account strand information, did not detect ESTs/oligos intronic to gene models, or assigned ESTs/oligos to models that were incorrectly annotated. Other oligos were derived from cDNA cloning artifacts, or erroneous sequence selection. By removing and correcting these errors, we generated what we consider the most thorough and accurate constitutive transcriptome of the zebra finch song control system [27]. We describe this curation effort below.

For 27,974 out of the 37,609 oligos (74%) that passed initial filters we provide the consensus gene symbol as in previous efforts [7, 32] (Table 7), based on the Human Genome Nomenclature Consortium (HGNC; 2018). For the remaining 9635 oligos (26%) that define the constitutive transcriptome of the finch song system [27], we inspected alignments against Taegut1, and annotated sequences based on association with a gene model (Ensembl or finch-/xeno-RefSeqs on the correct strand). For ESTs corresponding to 3′-UTRs, we BLAT-aligned sequences to chicken (Galgal5) to try to connect them to chicken gene models by ‘walking’ the extensive chicken ESTs/mRNAs collection. In total, 3750 oligos annotations were confirmed by direct inspection (Table 8), and an additional 130 oligos further confirmed by synteny (Table 9), which required additional alignments and neighbor gene comparisons with other avian (e.g. chicken, Tibetan tit, other finches, budgerigar, starling, falcon) and non-avian (i.e., alligator, lizard, mouse, human) genomes. We provided correct annotations for 1529 unannotated or misannotated oligos (Table 10), including cases of improper Ensembl model assignment (e.g. wrong strand) or intronic location to a model, determining orthology for another 58 oligos (Table 11). 1533 oligos associated with loci with no orthologs in other organisms (Table 12) were named unknown. Lastly, we updated 2635 oligos to an HGNC symbol, or a consensus NCBI:Gene name (Table 13).

Our findings highlight the need for accurate curations to avoid propagating errors in gene identification and bioinformatics. This partial curated dataset (~ 35% of oligos on this array) serves as a reference for correcting errors from previous studies, and a roadmap for future oligo curations. We anticipate for the 27,975 oligos not examined here, 32% will require further curation, and 27% will require updated gene symbols (Table 7).